Ok

En poursuivant votre navigation sur ce site, vous acceptez l'utilisation de cookies. Ces derniers assurent le bon fonctionnement de nos services. En savoir plus.

Data Management Platform (DMP) Administrator's Guide S–2327–B RECORD OF REVISION S–2327–B Published November 2013. Updated to support new features in ESM XX-2.1.0. S–2327–A Published June 2013. Original printing.

Au format texte : R Data Management Platform (DMP) Administrator's Guide S–2327–B © 2013 Cray Inc. All Rights Reserved. This document or parts thereof may not be reproduced in any form unless permitted by contract or by written permission of Cray Inc. U.S. GOVERNMENT RESTRICTED RIGHTS NOTICE The Computer Software is delivered as "Commercial Computer Software" as defined in DFARS 48 CFR 252.227-7014. All Computer Software and Computer Software Documentation acquired by or for the U.S. Government is provided with Restricted Rights. Use, duplication or disclosure by the U.S. Government is subject to the restrictions described in FAR 48 CFR 52.227-14 or DFARS 48 CFR 252.227-7014, as applicable. Technical Data acquired by or for the U.S. Government, if any, is provided with Limited Rights. Use, duplication or disclosure by the U.S. Government is subject to the restrictions described in FAR 48 CFR 52.227-14 or DFARS 48 CFR 252.227-7013, as applicable. The following are trademarks of Cray Inc. and are registered in the United States and other countries: Cray and design, Sonexion, Urika, and YarcData. The following are trademarks of Cray Inc.: ACE, Apprentice2, Chapel, Cluster Connect, CrayDoc, CrayPat, CrayPort, ECOPhlex, LibSci, NodeKARE, Threadstorm. The following system family marks, and associated model number marks, are trademarks of Cray Inc.: CS, CX, XC, XE, XK, XMT, and XT. The registered trademark Linux is used pursuant to a sublicense from LMI, the exclusive licensee of Linus Torvalds, owner of the mark on a worldwide basis. Other trademarks used in this document are the property of their respective owners. AMD is a trademark of Advanced Micro Devices, Inc. Apple, Mac, and OS X are trademarks of Apple Inc. Bright Cluster Manager is a registered trademark of Bright Computing, Inc. CentOS is a trademark of the CentOS Project. DELL is a trademark of Dell, Inc. DataDirect Networks and DDN are trademarks of DataDirect Networks, Inc. Engenio is a trademark of Engenio Inc. NetApp and SANtricity are trademarks of NetApp Inc. GPFS is a trademark of International Business Machines Corporation. InfiniBand is a registered trademark and service mark of InfiniBand Trade Association. Intel is a registered trademark of Intel Corporation in the United States and/or other countries. ISO is a trademark of International Organization for Standardization (Organisation Internationale de Normalisation). Java, JRE, MySQL, and NFS are trademarks of Oracle and/or its affiliates. LSI, MegaRAID, and MegaCLI are trademarks of LSI Corporation. Lustre is a registered trademark of Xyratex and/or its affiliates. Moab is a trademark of Adaptive Computing Enterprises, Inc. Novell, Qlogic is a registered trademark of Qlogic Corporation. RSA is a registered trademark of RSA Security Inc. UNIX, the “X device,” X Window System, and X/Open are trademarks of The Open Group. VNC is a registered trademark of RealVNC Ltd. Windows is a registered trademark of Microsoft Corporation. RECORD OF REVISION S–2327–B Published November 2013. Updated to support new features in ESM XX-2.1.0. S–2327–A Published June 2013. Original printing. Contents Page Introduction [1] 15 1.1 Scope of the Manual . . . . . . . . . . . . . . . . . . . . . . . 15 1.2 Related Publications . . . . . . . . . . . . . . . . . . . . . . . 15 1.3 External Services Node Name Changes . . . . . . . . . . . . . . . . . . 16 1.4 System Overview . . . . . . . . . . . . . . . . . . . . . . . . 17 1.5 Software Components . . . . . . . . . . . . . . . . . . . . . . 18 1.5.1 Determining the Current Software Releases . . . . . . . . . . . . . . . 18 1.5.2 Bright Cluster Manager Software Overview . . . . . . . . . . . . . . . 18 1.5.3 Cray Integrated Management Services (CIMS) Software Overview . . . . . . . . . 19 1.5.4 CDL Software Overview . . . . . . . . . . . . . . . . . . . . 19 1.5.4.1 Cray XE and Cray XK CDL Software . . . . . . . . . . . . . . . 19 1.5.4.2 Cray XC30 CDL Software . . . . . . . . . . . . . . . . . . . 19 1.5.5 Cray Lustre File System (CLFS) Software . . . . . . . . . . . . . . . . 19 1.5.6 CLE, CDT, and CADE . . . . . . . . . . . . . . . . . . . . . 19 1.5.7 Installing DMP Software . . . . . . . . . . . . . . . . . . . . 20 1.5.7.1 Distribution Media . . . . . . . . . . . . . . . . . . . . . 20 1.5.8 DMP Tools and Utilities . . . . . . . . . . . . . . . . . . . . . 22 1.5.8.1 esdumpsys . . . . . . . . . . . . . . . . . . . . . . 22 1.5.8.2 ESMupdateimage . . . . . . . . . . . . . . . . . . . . 22 1.5.8.3 eswrap . . . . . . . . . . . . . . . . . . . . . . . 22 1.5.8.4 cray-esfs-catman . . . . . . . . . . . . . . . . . . . 23 1.5.8.5 esfsmon_failback . . . . . . . . . . . . . . . . . . . 23 1.5.8.6 update_excludelist . . . . . . . . . . . . . . . . . . 23 1.6 DMP Networks Overview . . . . . . . . . . . . . . . . . . . . . 23 1.6.1 CIMS Network Configuration . . . . . . . . . . . . . . . . . . . 25 1.6.2 CDL Network Configuration . . . . . . . . . . . . . . . . . . . 28 1.6.3 CLFS Network Configuration . . . . . . . . . . . . . . . . . . . 28 1.7 Hardware Components . . . . . . . . . . . . . . . . . . . . . . 30 1.7.1 Cray Management Server (CIMS) Hardware Overview . . . . . . . . . . . . 30 S–2327–B 3 Data Management Platform (DMP) Administrator's Guide Page 1.7.2 Cray Development and Login (CDL) Node Hardware Overview . . . . . . . . . . 32 1.7.3 Lustre® File Server (CLFS) Node Hardware Overview . . . . . . . . . . . . 32 1.7.4 Switches and PDUs . . . . . . . . . . . . . . . . . . . . . . 34 Bright Cluster Manager® [2] 35 2.1 Managing a System with Bright . . . . . . . . . . . . . . . . . . . . 35 2.1.1 Devices and Device Names in Bright . . . . . . . . . . . . . . . . . 35 2.1.2 Node Organization . . . . . . . . . . . . . . . . . . . . . . 36 2.1.3 Software Image Management . . . . . . . . . . . . . . . . . . . 39 2.1.4 Node Provisioning . . . . . . . . . . . . . . . . . . . . . . 40 2.1.4.1 Preboot Execution Environment (PXE) Booting . . . . . . . . . . . . . 41 2.1.4.2 Booting From the Hard Drive . . . . . . . . . . . . . . . . . . 42 2.1.4.3 The Boot Role . . . . . . . . . . . . . . . . . . . . . . 42 2.2 Bright License Management . . . . . . . . . . . . . . . . . . . . . 42 2.2.1 Displaying License Attributes . . . . . . . . . . . . . . . . . . . 42 2.2.2 Verifying A License . . . . . . . . . . . . . . . . . . . . . . 43 2.2.3 Installing the Bright License . . . . . . . . . . . . . . . . . . . 43 2.2.4 Reinstating an Expired License . . . . . . . . . . . . . . . . . . . 48 2.2.5 Reboot After Installing License . . . . . . . . . . . . . . . . . . . 48 2.3 Bright GUI . . . . . . . . . . . . . . . . . . . . . . . . . 49 2.4 The Command Shell . . . . . . . . . . . . . . . . . . . . . . . 49 2.4.1 Mixing cmsh and UNIX Shell Commands . . . . . . . . . . . . . . . 52 2.5 Installing and Running cmgui on a Remote System . . . . . . . . . . . . . . 53 2.6 Running cmgui and Connecting to a DMP System . . . . . . . . . . . . . . 54 2.7 Running cmgui From the CIMS . . . . . . . . . . . . . . . . . . . 56 CIMS Configuration Tasks [3] 59 3.1 Software Installation . . . . . . . . . . . . . . . . . . . . . . . 59 3.1.1 Updating Slave Node RPMs from ESM Media . . . . . . . . . . . . . . 59 3.2 Administrative Passwords . . . . . . . . . . . . . . . . . . . . . 60 3.2.1 Managing Bright admin.pfx Certificates . . . . . . . . . . . . . . . 62 3.2.2 Changing the Password for the Baseboard Management Controller (BMC or iDRAC) . . . . 64 3.3 Changing CIMS Configuration Settings . . . . . . . . . . . . . . . . . . 65 3.4 Configuring the RAID Virtual Disks . . . . . . . . . . . . . . . . . . . 66 3.5 Configuring the LSI® MegaCLI™ RAID Utility . . . . . . . . . . . . . . . 69 3.6 Adding a New or Modified Disk Setup XML File to the Bright Database . . . . . . . . . 76 3.7 Changing BIOS for a DELL™ R720 CIMS Node . . . . . . . . . . . . . . . 77 3.8 Power Control . . . . . . . . . . . . . . . . . . . . . . . . . 87 4 S–2327–B Contents Page 3.9 Rebooting Slave Nodes . . . . . . . . . . . . . . . . . . . . . . 89 3.10 Shutting Down Slave Nodes . . . . . . . . . . . . . . . . . . . . 90 3.11 Network Settings . . . . . . . . . . . . . . . . . . . . . . . 91 3.11.1 The sipcalc Utility . . . . . . . . . . . . . . . . . . . . . 95 3.11.2 DNS Domains . . . . . . . . . . . . . . . . . . . . . . . 96 3.11.3 Adding a Network . . . . . . . . . . . . . . . . . . . . . . 96 3.11.4 Changing Node Host Names . . . . . . . . . . . . . . . . . . . 97 3.11.5 Adding Hostname to an Internal Network . . . . . . . . . . . . . . . 98 3.11.6 Changing External Network Parameters for the System . . . . . . . . . . . . 100 3.11.6.1 Changing the External Network Object Settings . . . . . . . . . . . . 100 3.11.6.2 Changing Network Settings for the CIMS . . . . . . . . . . . . . . 102 3.11.7 Using DHCP to Supply Network Values for the External Interface . . . . . . . . . 103 3.12 Isolate a Slave Node for Testing . . . . . . . . . . . . . . . . . . . 103 3.13 Changing Node Category . . . . . . . . . . . . . . . . . . . . . 106 3.14 Creating a CDL Node Group . . . . . . . . . . . . . . . . . . . . 106 3.15 Configuring Virtual Network Computing (VNC®) on the CIMS . . . . . . . . . . . 107 3.16 Adding a Managed Switch or Device to the Bright Configuration . . . . . . . . . . 111 3.17 DELL™ 5548 Switch Configuration . . . . . . . . . . . . . . . . . . 114 3.18 Zoning the QLogic® FC Switch . . . . . . . . . . . . . . . . . . . . 116 3.19 Configuring the InfiniBand (ib-net) Network for Slave Nodes . . . . . . . . . . 118 3.20 iDRAC Remote Console . . . . . . . . . . . . . . . . . . . . . 120 3.21 Configuring Administrator Email Alerts from the CIMS . . . . . . . . . . . . . 122 3.22 Configuring SSH Keys for eswrap on CDL and Internal Login Nodes . . . . . . . . 123 3.23 Save the System Configuration . . . . . . . . . . . . . . . . . . . . 126 3.24 Resizing Partitions on the CIMS . . . . . . . . . . . . . . . . . . . 128 3.25 Configuring DHCP to Allow Requests from Unknown Nodes . . . . . . . . . . . 136 3.26 Changing the CIMS Firewall Configuration . . . . . . . . . . . . . . . . 136 CDL Administration Tasks [4] 137 4.1 Creating a CDL Node in Bright Cluster Manager® . . . . . . . . . . . . . . . 137 4.2 Configuring Network Parameters for the Site User Network . . . . . . . . . . . . 140 4.3 Creating the Workload Manager Network (wlm-net) . . . . . . . . . . . . . 142 4.4 Configuring Bright Categories for the CDL Nodes . . . . . . . . . . . . . . . 145 4.5 Configuring Kdump on CDL Nodes (SLES) . . . . . . . . . . . . . . . . 148 4.6 Creating a kdump Crash Partition on a Slave Node Local Disk . . . . . . . . . . . 153 4.7 Seting CDL Device Parameters . . . . . . . . . . . . . . . . . . . . 155 4.8 Configuring eswrap . . . . . . . . . . . . . . . . . . . . . . 156 4.8.1 Support for Native SLURM . . . . . . . . . . . . . . . . . . . . 157 S–2327–B 5 Data Management Platform (DMP) Administrator's Guide Page 4.9 Configuring a CDL Node to Mount an External Lustre® File System . . . . . . . . . . 157 4.10 Mounting a Lustre® File System on a CDL Node . . . . . . . . . . . . . . . 159 4.11 Setting Up Exclude Lists . . . . . . . . . . . . . . . . . . . . . 160 4.11.1 Check Exclude Lists . . . . . . . . . . . . . . . . . . . . . 162 4.11.2 Changing Exclude Lists . . . . . . . . . . . . . . . . . . . . 162 4.11.3 Exclude User Home Directories . . . . . . . . . . . . . . . . . . 163 CLFS Administration Tasks [5] 165 5.1 Recommended MDS/MGT Volume Size . . . . . . . . . . . . . . . . . 165 5.2 Configuring CLFS Failover (esfsmon) . . . . . . . . . . . . . . . . . 165 5.2.1 Storage Configuration Overview . . . . . . . . . . . . . . . . . . 165 5.2.2 Failover Conditions . . . . . . . . . . . . . . . . . . . . . . 166 5.2.3 Failover Functional Tests . . . . . . . . . . . . . . . . . . . . 166 5.2.4 HA/Failover of I/O Paths for Servers Connected to Storage Arrays . . . . . . . . . 166 5.2.5 Disk Failure and Rebuild: RAID Hardware Capability . . . . . . . . . . . . 167 5.2.6 Failover Features and Bright Monitoring . . . . . . . . . . . . . . . . 167 5.2.6.1 esfsmon healthcheck . . . . . . . . . . . . . . . . . 168 5.2.7 Configuring the esfsmon healthcheck Monitor . . . . . . . . . . . . 169 5.2.7.1 Installing esfsmon_healthcheck and esfsmon_action Scripts . . . . . 169 5.2.8 Configure esfsmon.conf . . . . . . . . . . . . . . . . . . . 171 5.2.9 Activating esfsmon . . . . . . . . . . . . . . . . . . . . . 174 5.2.10 Configuring esfsmon healthcheck in Bright . . . . . . . . . . . . 174 5.2.11 Controlling the Bright Cluster Manager Failover Monitor (esfsmon) . . . . . . . 176 5.2.12 Avoiding Inadvertent Failover Operations . . . . . . . . . . . . . . . 177 5.2.13 Tuning the esfsmon:filesystem Check Interval . . . . . . . . . . . . . 177 5.2.14 Returning a Node to the CLFS Service . . . . . . . . . . . . . . . . 178 5.2.15 OSS Failover . . . . . . . . . . . . . . . . . . . . . . . 178 5.2.15.1 Failover Actions . . . . . . . . . . . . . . . . . . . . . 179 5.2.15.2 Corrective Actions . . . . . . . . . . . . . . . . . . . . 179 5.2.15.3 Recovery Actions . . . . . . . . . . . . . . . . . . . . . 179 5.2.16 MDT/MGS Failover . . . . . . . . . . . . . . . . . . . . . 179 5.2.16.1 Failover Actions . . . . . . . . . . . . . . . . . . . . . 180 5.2.16.2 Corrective Actions . . . . . . . . . . . . . . . . . . . . 180 5.2.16.3 Recovery Actions . . . . . . . . . . . . . . . . . . . . . 180 5.2.16.4 MGS Failover . . . . . . . . . . . . . . . . . . . . . . 180 5.3 Configuring Kdump on CentOS™ (Optional) . . . . . . . . . . . . . . . . 181 5.4 Configuring a NetApp™ Storage System . . . . . . . . . . . . . . . . . 186 5.4.1 Installing SANtricity Storage Manager Software for NetApp Devices . . . . . . . . 186 6 S–2327–B Contents Page 5.4.2 Configuring LUNs for NetApp Devices . . . . . . . . . . . . . . . . 187 5.4.3 Host Mappings . . . . . . . . . . . . . . . . . . . . . . . 189 5.4.4 Configuring Remote Logging of NetApp™ Storage System Messages . . . . . . . . 190 5.4.5 Add a NetApp RAID Storage System to Bright . . . . . . . . . . . . . . 190 5.5 Rediscovering the New LUNs . . . . . . . . . . . . . . . . . . . . 192 5.6 Partitioning the LUNs . . . . . . . . . . . . . . . . . . . . . . . 192 5.7 Adding Nodes to CLFS Categories . . . . . . . . . . . . . . . . . . . 193 5.8 Configuring Bright Categories for CLFS Nodes . . . . . . . . . . . . . . . . 194 5.9 Configuring Lustre® File Systems on CLFS Nodes . . . . . . . . . . . . . . . 197 5.10 Configuring Lustre® File Systems on MDS and OSS Nodes . . . . . . . . . . . . 199 5.11 Creating a Generic CLFS Node in Bright Cluster Manager® . . . . . . . . . . . . 200 5.12 Configuring Node Finalize Script for MDS Nodes . . . . . . . . . . . . . . . 203 5.13 Configuring Node Finalize Script for OSS Nodes . . . . . . . . . . . . . . . 205 5.14 Configure LDAP on MDS Nodes . . . . . . . . . . . . . . . . . . . 206 5.15 Configuring the Site User Network (for MDS Nodes Only) . . . . . . . . . . . . 208 5.16 Mounting a CLFS Lustre® File System on a Cray CLE System . . . . . . . . . . . 210 5.17 Configuring Multipath on CLFS nodes . . . . . . . . . . . . . . . . . . 215 5.18 Migrating from Lustre® 1.8.x to 2.4 . . . . . . . . . . . . . . . . . . 220 5.18.1 Related Publications . . . . . . . . . . . . . . . . . . . . . 220 5.18.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . 221 5.18.3 FIDS . . . . . . . . . . . . . . . . . . . . . . . . . 221 5.18.4 Quota Support . . . . . . . . . . . . . . . . . . . . . . . 222 5.18.5 Performance Expectations . . . . . . . . . . . . . . . . . . . . 223 5.18.5.1 Object Index Creation and Repair . . . . . . . . . . . . . . . . 223 5.18.5.2 Adding FIDs to inodes . . . . . . . . . . . . . . . . . . . 223 5.18.5.3 Space Usage Statistics . . . . . . . . . . . . . . . . . . . 224 5.18.6 Upgrade Procedure . . . . . . . . . . . . . . . . . . . . . . 224 5.19 SCSI RDAC Driver Kernel Parameters for Fibre Channel Storage . . . . . . . . . . 228 5.20 Configuring Lustre® Monitoring Tool (LMT) and Cerebro on the CIMS . . . . . . . . . 229 5.20.1 Configuring Cerebro . . . . . . . . . . . . . . . . . . . . . 230 5.20.2 Starting Cerebro and LMT . . . . . . . . . . . . . . . . . . . . 230 5.20.3 Configuring the MySQL Database for Cerebro and LMT . . . . . . . . . . . 231 5.20.4 Managing the Data . . . . . . . . . . . . . . . . . . . . . . 233 5.20.5 Stopping the Cerebro Service . . . . . . . . . . . . . . . . . . . 234 5.20.6 Deleting the LMT MySQL Database . . . . . . . . . . . . . . . . . 234 5.20.7 Managing Cerebro with Bright . . . . . . . . . . . . . . . . . . 234 S–2327–B 7 Data Management Platform (DMP) Administrator's Guide Page Monitoring and Troubleshooting [6] 237 6.1 Check Device Status . . . . . . . . . . . . . . . . . . . . . . . 240 6.2 Check Power Status . . . . . . . . . . . . . . . . . . . . . . . 240 6.3 Check Node Health Status . . . . . . . . . . . . . . . . . . . . . 240 6.4 Monitoring Configuration with cmgui . . . . . . . . . . . . . . . . . . 241 6.5 Check System Status . . . . . . . . . . . . . . . . . . . . . . . 245 6.6 Monitoring Health and Metrics . . . . . . . . . . . . . . . . . . . . 247 6.6.1 Set Email Alerts when a Node Goes Down . . . . . . . . . . . . . . . 250 6.7 Log Files . . . . . . . . . . . . . . . . . . . . . . . . . . 251 6.8 Bright Logs . . . . . . . . . . . . . . . . . . . . . . . . . 251 6.8.1 View the Event Log . . . . . . . . . . . . . . . . . . . . . . 251 6.8.2 View the rsync Log . . . . . . . . . . . . . . . . . . . . . 252 Appendix A Changing BIOS for a DELL™ R720 Managed CDL Node 253 Appendix B Changing BIOS for a DELL™ R815 Managed CDL Node 261 Appendix C Changing BIOS for a DELL™ R720 Managed CLFS Node 269 Procedures Procedure 1. Installing the Bright license on a CIMS . . . . . . . . . . . . . . . 44 Procedure 2. Reinstating an expired license . . . . . . . . . . . . . . . . . 48 Procedure 3. Installing and running cmgui . . . . . . . . . . . . . . . . . 53 Procedure 4. Starting the cmgui and connecting to the system . . . . . . . . . . . . 54 Procedure 5. Install and run cmgui from the CIMS . . . . . . . . . . . . . . . 56 Procedure 6. Updating slave node RPMs from ESM media . . . . . . . . . . . . . 59 Procedure 7. Changing DMP system passwords . . . . . . . . . . . . . . . . 61 Procedure 8. Revoking administration certificates . . . . . . . . . . . . . . . 64 Procedure 9. Changing the password on the BMC (iDRAC) . . . . . . . . . . . . . 64 Procedure 10. Setting up RAID virtual disks . . . . . . . . . . . . . . . . . 66 Procedure 11. Installing the MegaCLI utility on the CIMS . . . . . . . . . . . . . 69 Procedure 12. Installing the MegaCLI utility on slave node . . . . . . . . . . . . . 70 Procedure 13. Configuring the megaraid healthcheck in Bright . . . . . . . . . . . 72 Procedure 14. Changing the disk setup XML file for a category . . . . . . . . . . . . 76 Procedure 15. Changing the system setup for the CIMS . . . . . . . . . . . . . . 77 Procedure 16. Rebooting slave nodes . . . . . . . . . . . . . . . . . . . 89 Procedure 17. Changing node hostnames . . . . . . . . . . . . . . . . . . 97 Procedure 18. Isolating slave node for testing . . . . . . . . . . . . . . . . . 103 Procedure 19. Change node configuration . . . . . . . . . . . . . . . . . . 106 8 S–2327–B Contents Page Procedure 20. Creating a CDL node group . . . . . . . . . . . . . . . . . 106 Procedure 21. Starting the VNC server . . . . . . . . . . . . . . . . . . 107 Procedure 22. Connecting to VNC server through an ssh tunnel, using the vncviewer . . . . 109 Procedure 23. Connecting to the VNC server through an ssh tunnel . . . . . . . . . . 109 Procedure 24. Connecting an Apple® Mac® OS X system to the VNC server through an ssh tunnel . 110 Procedure 25. Connecting to the VNC server through an ssh tunnel with Windows® . . . . . . 110 Procedure 26. Adding a Mellanox IS50XX series switch to the Bright configuration . . . . . . 111 Procedure 27. Configuring the Dell 5548 1GbE switch . . . . . . . . . . . . . . 114 Procedure 28. Configuring zoning for a QLogic SANbox switch using QuickTools utility . . . . 116 Procedure 29. Creating a backup of your QLogic switch configuration . . . . . . . . . . 117 Procedure 30. Configuring the InfiniBand (ib-net) network in Bright . . . . . . . . . 118 Procedure 31. Using the iDRAC7 web interface and remote console . . . . . . . . . . 120 Procedure 32. Configuring administrator email alerts from the CIMS using cmgui . . . . . . 122 Procedure 33. Configuring administrator email alerts from the CIMS using cmsh . . . . . . 122 Procedure 34. Configuring SSH Keys for eswrap on CDL and internal login nodes . . . . . . 123 Procedure 35. Save the system configuration settings to an XML file . . . . . . . . . . 127 Procedure 36. Load system configuration settings from an XML file . . . . . . . . . . 127 Procedure 37. Resizing partitions on a CIMS . . . . . . . . . . . . . . . . . 128 Procedure 38. Configuring DHCP to allow requests from unknown nodes . . . . . . . . . 136 Procedure 39. Creating a CDL node in Bright . . . . . . . . . . . . . . . . 137 Procedure 40. Setting network parameters for site-user-net . . . . . . . . . . 140 Procedure 41. Creating the workload manager network (wlm-net) . . . . . . . . . . 142 Procedure 42. Configure category settings for the CDL image . . . . . . . . . . . . 145 Procedure 43. Configuring kdump on CDL nodes (SLES) . . . . . . . . . . . . . 148 Procedure 44. Create a kdump /var/crash partition on a slave node . . . . . . . . . 153 Procedure 45. Setting CDL slave node parameters in Bright . . . . . . . . . . . . . 155 Procedure 46. Configuring eswrap . . . . . . . . . . . . . . . . . . . 156 Procedure 47. Configuring a CDL node to mount an external Lustre file system . . . . . . . 158 Procedure 48. Mounting a Lustre file system on a CDL node . . . . . . . . . . . . 159 Procedure 49. Installing esfsmon_healthcheck and esfsmon_action . . . . . . 169 Procedure 50. Configuring esfsmon healthcheck in Bright . . . . . . . . . . 174 Procedure 51. Tuning the esfsmon:filesystem check interval . . . . . . . . . . . 177 Procedure 52. Configuring kdump on CentOS . . . . . . . . . . . . . . . . 181 Procedure 53. Installing SANtricity storage management software . . . . . . . . . . . 186 Procedure 54. Creating a volume group for NetApp devices . . . . . . . . . . . . . 187 Procedure 55. Creating and configuring volumes for NetApp devices . . . . . . . . . . 188 Procedure 56. Adding a NetApp RAID storage system to Bright . . . . . . . . . . . 190 S–2327–B 9 Data Management Platform (DMP) Administrator's Guide Page Procedure 57. Rebooting the CLFS and verifying LUNs are recognized . . . . . . . . . 192 Procedure 58. Adding nodes to a CLFS category . . . . . . . . . . . . . . . . 193 Procedure 59. Configuring Bright CLFS categories . . . . . . . . . . . . . . . 194 Procedure 60. Configuring Lustre file systems on CLFS nodes from the CIMS . . . . . . . 197 Procedure 61. Configuring Lustre file systems on MDS and OSS nodes . . . . . . . . . 199 Procedure 62. Creating a CLFS node in Bright . . . . . . . . . . . . . . . . 200 Procedure 63. Configuring the node finalize script for MDS nodes . . . . . . . . . . . 204 Procedure 64. Configuring the node finalize script for OSS nodes . . . . . . . . . . . 205 Procedure 65. Configure LDAP on MDS nodes . . . . . . . . . . . . . . . . 207 Procedure 66. Configuring the site user network (for MDS nodes only) . . . . . . . . . . 208 Procedure 67. Mounting a Lustre file system on a CLE system . . . . . . . . . . . . 210 Procedure 68. Configuring multipath on CLFS nodes . . . . . . . . . . . . . . 216 Procedure 69. Upgrading Lustre 1.8.x to Lustre 2.4 . . . . . . . . . . . . . . . 224 Procedure 70. Adding SCSI RDAC kernel parameter to ESF software image . . . . . . . . 228 Procedure 71. Configuring the MySQL database for Cerebro and LMT . . . . . . . . . . 231 Procedure 72. Managing Cerebro on a slave node with Bright . . . . . . . . . . . . 234 Procedure 73. Managing Cerebro for a category . . . . . . . . . . . . . . . . 235 Procedure 74. Monitoring configuration setup . . . . . . . . . . . . . . . . 242 Procedure 75. Check power status . . . . . . . . . . . . . . . . . . . . 247 Procedure 76. Monitor system health and metrics . . . . . . . . . . . . . . . 248 Procedure 77. Display the metric status . . . . . . . . . . . . . . . . . . 248 Procedure 78. Dump health check data . . . . . . . . . . . . . . . . . . 249 Procedure 79. Dump metric data . . . . . . . . . . . . . . . . . . . . 249 Procedure 80. Set email alerts when a node goes down . . . . . . . . . . . . . . 250 Procedure 81. View the event log . . . . . . . . . . . . . . . . . . . . 251 Procedure 82. View the rsync log . . . . . . . . . . . . . . . . . . . 252 Procedure 83. Changing a R720 slave node's BIOS and iDRAC settings . . . . . . . . . 253 Procedure 84. Changing a R815 slave node's BIOS and iDRAC settings . . . . . . . . . 261 Procedure 85. Changing a R720 CLFS node's BIOS and iDRAC settings . . . . . . . . . 269 Examples Example 1. List Software Images . . . . . . . . . . . . . . . . . . . . 40 Example 2. Set PXE label for a node . . . . . . . . . . . . . . . . . . . 41 Example 3. Display license attributes in cmsh . . . . . . . . . . . . . . . . 43 Example 4. Verifying a License . . . . . . . . . . . . . . . . . . . . 43 Example 5. Mixing cmsh and UNIX Commands . . . . . . . . . . . . . . . 52 Example 6. Using UNIX output in cmsh commands . . . . . . . . . . . . . . 53 Example 7. Using a foreach loop to invoke commands . . . . . . . . . . . . . 53 10 S–2327–B Contents Page Example 8. CIMS configuration settings . . . . . . . . . . . . . . . . . . 65 Example 9. The sipcalc Utility . . . . . . . . . . . . . . . . . . . 95 Example 10. Changing the Default Setting of a Network . . . . . . . . . . . . . 97 Example 11. Changing node host names with cmsh . . . . . . . . . . . . . . . 98 Example 12. Adding host names to an internal network using cmsh . . . . . . . . . . 100 Example 13. Changing the Network Settings for the CIMS . . . . . . . . . . . . . 103 Example 14. Changing SNMP community strings for devices . . . . . . . . . . . . 111 Example 15. Starting cerebrod manually . . . . . . . . . . . . . . . . . 231 Example 16. cmsh monitoring actions Mode . . . . . . . . . . . . . . 238 Example 17. cmsh monitoring healthchecks mode . . . . . . . . . . . 239 Example 18. cmsh monitoring metrics Mode . . . . . . . . . . . . . . 239 Tables Table 1. Data Management Platform Node Name Changes . . . . . . . . . . . . . 16 Table 2. CIMS Network Interfaces . . . . . . . . . . . . . . . . . . . 27 Table 3. DMP Slave Node Categories . . . . . . . . . . . . . . . . . . . 36 Table 4. Node Groups . . . . . . . . . . . . . . . . . . . . . . . 39 Table 5. Command Shell Object Descriptions . . . . . . . . . . . . . . . . . 52 Table 6. Network Configuration Settings . . . . . . . . . . . . . . . . . . 94 Table 7. healthcheck Alert Levels . . . . . . . . . . . . . . . . . . 241 Figures Figure 1. DMP System Hardware Overview . . . . . . . . . . . . . . . . . 18 Figure 2. DMP Hardware and Networks Overview . . . . . . . . . . . . . . . 25 Figure 3. CIMS Network Interfaces and Default Addresses . . . . . . . . . . . . . 26 Figure 4. CDL Network Interfaces and Default Addresses . . . . . . . . . . . . . 28 Figure 5. CLFS Network Interfaces and Default Addresses . . . . . . . . . . . . . 29 Figure 6. CIMS Node Hardware Overview . . . . . . . . . . . . . . . . . 31 Figure 7. CDL Node Hardware Overview . . . . . . . . . . . . . . . . . . 32 Figure 8. CLFS OSS Node Hardware Overview . . . . . . . . . . . . . . . . 33 Figure 9. CLFS MDS Hardware Overview . . . . . . . . . . . . . . . . . 34 Figure 10. PXE Boot Menu . . . . . . . . . . . . . . . . . . . . . 41 Figure 11. Bright License Request Form . . . . . . . . . . . . . . . . . . 46 Figure 12. Bright cmgui Window . . . . . . . . . . . . . . . . . . . 49 Figure 13. Connecting cmgui to a DMP System . . . . . . . . . . . . . . . 55 Figure 14. Connecting cmgui to a DMP System . . . . . . . . . . . . . . . 55 Figure 15. cmgui Splash Screen . . . . . . . . . . . . . . . . . . . . 56 Figure 16. cmgui Window . . . . . . . . . . . . . . . . . . . . . 57 S–2327–B 11 Data Management Platform (DMP) Administrator's Guide Page Figure 17. cmgui Connect-to-Cluster . . . . . . . . . . . . . . . . . . 57 Figure 18. cmgui Authentication Menu . . . . . . . . . . . . . . . . . . 63 Figure 19. Final RAID Configuration Settings . . . . . . . . . . . . . . . . 68 Figure 20. Dell 720 BIOS Menu . . . . . . . . . . . . . . . . . . . . 78 Figure 21. Dell 720 BIOS Boot Settings . . . . . . . . . . . . . . . . . . 78 Figure 22. Dell 720 BIOS Boot Sequence . . . . . . . . . . . . . . . . . . 79 Figure 23. Dell 720 BIOS Boot Settings . . . . . . . . . . . . . . . . . . 80 Figure 24. Dell 720 BIOS Serial Communication Settings . . . . . . . . . . . . . 80 Figure 25. Dell 720 BIOS Serial Port Address Settings . . . . . . . . . . . . . . 81 Figure 26. Dell 720 BIOS iDRAC Settings . . . . . . . . . . . . . . . . . 82 Figure 27. Dell 720 BIOS iDRAC Name . . . . . . . . . . . . . . . . . . 82 Figure 28. Dell 720 BIOS Enable IPMI over LAN (SOL) . . . . . . . . . . . . . 84 Figure 29. Dell 720 BIOS iDRAC LCD Settings . . . . . . . . . . . . . . . . 85 Figure 30. Dell 720 BIOS MBA Configuration Settings . . . . . . . . . . . . . . 86 Figure 31. Dell 720 System BIOS Settings . . . . . . . . . . . . . . . . . 87 Figure 32. Bright Network Configuration GUI . . . . . . . . . . . . . . . . 92 Figure 33. Network Settings GUI . . . . . . . . . . . . . . . . . . . . 93 Figure 34. Adding a Network . . . . . . . . . . . . . . . . . . . . . 96 Figure 35. Changing Node Host Names with cmgui . . . . . . . . . . . . . . 97 Figure 36. Adding a Hostname to an Internal Network . . . . . . . . . . . . . . 99 Figure 37. Changing External Network Object Parameters . . . . . . . . . . . . . 101 Figure 38. Changing Network Settings for the CIMS . . . . . . . . . . . . . . . 102 Figure 39. iDRAC Login Page . . . . . . . . . . . . . . . . . . . . . 120 Figure 40. iDRAC System Summary Page . . . . . . . . . . . . . . . . . 121 Figure 41. iDRAC Remote Console Window . . . . . . . . . . . . . . . . . 121 Figure 42. Administrator's Email Address . . . . . . . . . . . . . . . . . 122 Figure 43. Default Node in Bright . . . . . . . . . . . . . . . . . . . . 138 Figure 44. Setting up Exclude Lists in cmgui . . . . . . . . . . . . . . . . 162 Figure 45. Setting RDAC Kernel Parameters for Fibre Channel . . . . . . . . . . . . 229 Figure 46. Bright Monitoring Configuration . . . . . . . . . . . . . . . . . 242 Figure 47. Monitoring Configuration Category . . . . . . . . . . . . . . . . 243 Figure 48. Monitoring Configuration Metric . . . . . . . . . . . . . . . . . 244 Figure 49. Monitoring Configuration Action . . . . . . . . . . . . . . . . . 245 Figure 50. Dell 720 System BIOS Settings . . . . . . . . . . . . . . . . . 254 Figure 51. Dell 720 Boot Sequence BIOS Settings . . . . . . . . . . . . . . . 255 Figure 52. Dell 720 Serial Device BIOS Settings . . . . . . . . . . . . . . . . 256 Figure 53. Dell 720 Serial Communication BIOS Settings . . . . . . . . . . . . . 256 12 S–2327–B Contents Page Figure 54. Dell 720 iDRAC BIOS Settings . . . . . . . . . . . . . . . . . 257 Figure 55. Dell 720 iDRAC BIOS LCD Settings . . . . . . . . . . . . . . . . 258 Figure 56. Dell 720 MBA Configuration Menu BIOS Settings . . . . . . . . . . . . 259 Figure 57. Dell 720 Legacy Boot Protocol BIOS Settings . . . . . . . . . . . . . 260 Figure 58. Dell 815 Boot Settings Menu . . . . . . . . . . . . . . . . . . 262 Figure 59. Dell 815 Boot Sequence Menu . . . . . . . . . . . . . . . . . . 262 Figure 60. Dell 815 Boot Sequence Settings . . . . . . . . . . . . . . . . . 263 Figure 61. Dell 815 Integrated Devices (NIC) Settings . . . . . . . . . . . . . . 263 Figure 62. Dell 815 Serial Communication BIOS Settings . . . . . . . . . . . . . 264 Figure 63. Dell 815 Embedded Server Management Settings . . . . . . . . . . . . 265 Figure 64. Dell 815 User-defined LCD String Settings . . . . . . . . . . . . . . 265 Figure 65. Dell 815 DRAC LAN Parameters Settings . . . . . . . . . . . . . . 266 Figure 66. Dell 815 DRAC IPv4 Parameter Settings . . . . . . . . . . . . . . . 267 Figure 67. Dell 720 System BIOS Settings . . . . . . . . . . . . . . . . . 270 Figure 68. Dell 720 Boot Sequence BIOS Settings . . . . . . . . . . . . . . . 271 Figure 69. Dell 720 Serial Device BIOS Settings . . . . . . . . . . . . . . . . 272 Figure 70. Dell 720 Serial Communication BIOS Settings . . . . . . . . . . . . . 272 Figure 71. Dell 720 iDRAC BIOS Settings . . . . . . . . . . . . . . . . . 273 Figure 72. Dell 720 iDRAC BIOS LCD Settings . . . . . . . . . . . . . . . . 274 Figure 73. Dell 720 MBA Configuration Menu BIOS Settings . . . . . . . . . . . . 275 Figure 74. Dell 720 Legacy Boot Protocol BIOS Settings . . . . . . . . . . . . . 276 S–2327–B 13 Introduction [1] The Cray Data Management Platform (DMP) integrates Cray XC30, or Cray XE and Cray XK systems into a customer user environment using 1U and 2U commodity rack-mounted service nodes. DMP nodes run a combination of commercial off-the-shelf Linux™ software and Cray proprietary software to replicate the Cray Linux Environment (CLE) for application development and testing. These specialized external login and service nodes expand the role of the Cray system internal login nodes and provide a development platform for shared externalized file systems, data movers, or high-availability configurations. A Cray Integrated Management Services (CIMS) node runs the commercially available Bright Cluster Manager® software (Bright). Bright software enables administrators to manage many DMP nodes using the Bright CMDaemon (cmd), cluster management shell (cmsh) or the cluster management GUI (cmgui). Note: External shared file systems remain accessible to Cray DMP nodes, regardless of the state of the Cray system. An overview of the Bright software and how it is used to manage a Cray DMP system is discussed in Chapter 2, Bright Cluster Manager® on page 35. 1.1 Scope of the Manual This document provides procedures and common administration tasks that target Cray's implementation of Bright to manage a DMP system. Cray DMP systems are designed to function as special purpose service nodes for a Cray system, and are not designed as compute clusters. Note that Cray's implementation of Bright leverages its capability to manage Cray's specialized service nodes and not a traditional compute cluster. It follows that some features of Bright (such as cloudbursting) are not implemented in Cray DMP systems. Refer to the Bright Cluster Manager 6.0 Administrator Manual and Bright User Manuals that are stored on the CIMS in /cm/shared/docs/cm. These documents provide detailed information about the Bright software, and describe how to use the user interface (cmgui) and management shell (cmsh) to perform common administrative tasks. 1.2 Related Publications The following documents contain additional information that may be helpful: S–2327–B 15 Data Management Platform (DMP) Administrator's Guide • Installing Cray Integrated Management Services (CIMS) Software (S–2522) • Installing Cray Development and Login (CDL) Software (S–2520) • Installing Lustre File System by Cray (CLFS) Software (S–2521) • Installing CLE Support Package on a Cray Development and Login (CDL) Node (S–2528) • Managing System Software for the Cray Linux Environment (S–2393), which is provided with your CLE release package • Managing Lustre for the Cray Linux Environment (CLE) (S–0010), which is provided with your Cray Linux Environment (CLE) operating system release package • Bright Cluster Manager 6.0 Administrator Manual. PDFs are stored on the CIMS in /cm/shared/docs/cm • DELL™ R620, R720, R820, and R815 Hardware Owners Guides, available from 1.3 External Services Node Name Changes The Cray® External Services product line has been re-branded to Cray Data Management Platform (DMP). The node names in Bright Cluster Manager® (Bright) esms, esfs, and eslogin will continue to be used throughout the documentation examples as long as these node names are consistent with the installation and management software. The new nomenclature is used when nodes are described in a generic sense. Table 1. Data Management Platform Node Name Changes Old Name New Name Description esMS node CIMS node A Cray Integrated Management Services (CIMS) node is the centralized management node that aggregates data and manages slave node power, provisioning, reporting, health, and software images. esLogin node CDL node A Cray Development and Login (CDL) node is a secure multi-user application development platform with access to the same shared file system as internal Cray login nodes. esFS node CLFS node A Cray Lustre® File System (CLFS) node is a customized Lustre storage node that provides connectivity to DataDirect™ Networks (DDN) or NetApp™ storage devices. CLFS nodes are typically configured as MGS, MDS, OSS nodes. 16 S–2327–B Introduction [1] 1.4 System Overview Cray Data Management Platform (DMP) systems expand the role of Cray internal login nodes by providing a software development platform and shared storage for users. In order to expand the role of Cray system internal login nodes, a Cray DMP system must provide a software login and development platform for two distinct software environments (Cray XE and Cray XC30 systems), provide high-performance shared file systems that are accessible to both the Cray DMP system and Cray system. It follows, that to simulate the software development environment for both Cray XE and Cray XC30 systems, two DMP Cray Development and Login (CDL) node types must exist: one for Cray XE series systems and one for Cray XC30 series systems. The Cray Linux Environment (CLE) and Cray Developer Toolkit software for either a Cray XE or Cray XC30 system is installed on the CDL nodes, along with other commands and utilities to emulate the software development environment that exists on a Cray internal login node. DMP systems support Cray Lustre® file system (CLFS) nodes that are configured to support a shared Lustre file system for the Cray system and each CDL node. This file system remains available to developers on the DMP system regardless of the operational state of the Cray system. Cray DMP systems use Bright software on a Cray Integrated Management Services (CIMS) node to manage all of the nodes, software images, and networks in the system. Two CIMS nodes can be configured for high-availability (HA) configurations. Bright enables a system administrator to create custom categories and groups for system objects such as nodes, networks, software images, and other system entities, and manage them all from a single location. Using the Bright GUI (cmgui) or Bright command-line shell (cmsh), you can manage a very large system with many different node types, software images, networks, and hardware configurations from a single interface. S–2327–B 17 Data Management Platform (DMP) Administrator's Guide Figure 1. DMP System Hardware Overview eslogin-xc1 eslogin-xc2 eslogin1-xe1 eslogin2-xe2 esfs-mds001 esfs-mds002 esfs-oss001 esfs-oss002 esfs-oss003 esfs-oss004 esfs-oss005 esfs-oss006 Data Management Platform Shared Storage Cray System DDN Storage OST DDN Storage OST DDN Storage OST DDN Storage OST DDN SFA12000 DDN SFA12000 Cray XC30 login nodes in esLogin-XC category Cray XE/Cray XK login nodes in esLogin-XE category MDS nodes running CentOS and Lustre in esFS-MDS category OSS nodes running CentOS and Lustre in esFS-OSS category InfiniBand Fibre Channel InfiniBand Bright , SLES 11, Cray tools and utilities, storage management Maintenace network IPMI network Admin network GigE switch Maintenance network RAID and switch management IB switch FC switch NetApp E5400/E5500 2600 storage CIMS Primary CIMS Secondary 1.5 Software Components Each node type is supported by a separate software release that includes an operating system and custom Cray software to support its role in the DMP system. • Cray CIMS software (ESM), SLES, Bright Cluster Manager® software (Bright) • Cray CDL software (ESL) and SLES • Cray CLFS software (ESF) and Community Enterprise Operating System (CentOS)™ • Cray CDE and CADE 1.5.1 Determining the Current Software Releases The current ESM, ESF, or ESL software release installed is stored in /etc/opt/cray/release/. For example, the contents of /etc/opt/cray/release/ESMrelease file would contain ESM-XX-2.1.0-201310100612. Refer to Distribution Media on page 20 for a description of the release naming nomenclature. 1.5.2 Bright Cluster Manager Software Overview The Bright Cluster Manager® software (Bright) manages the hardware and software for every device or node in system as well as other objects such as node categories and networks. Bright is discussed in detail in Chapter 2, Bright Cluster Manager® on page 35. 18 S–2327–B Introduction [1] 1.5.3 Cray Integrated Management Services (CIMS) Software Overview Cray ESM software (ESM) includes SUSE Linux Enterprise Server 11 Service Pack 2 (SLES 11, SP2) and Bright 6.0. ESM software releases also include Cray tools and utilities, distributed replicated block device (DRBD), and parallel distributed shell (PDsh) software. Bright software is discussed in detail in Chapter 2, Bright Cluster Manager® on page 35. Refer to DMP Tools and Utilities on page 22 for a description of Cray DMP tools and utilities. 1.5.4 CDL Software Overview The Cray Development and Login (CDL) software (ESL) supports the development environment for both Cray XE and Cray XC30 systems. The Cray eswrap utility enables users on CDL nodes to run a subset of Cray internal login node commands. Refer to eswrap on page 22 for more information about the eswrap command. 1.5.4.1 Cray XE and Cray XK CDL Software Cray XE and Cray XK software releases provide SLES 11, SP1 as a base operating system and Lustre client software to support Lustre file systems. OpenFabrics Enterprise Distribution (OFED) is passed through directly from Novell. Cray Linux Environment (CLE) 4.x support package and the Cray Application Developer Environment (CADE) are installed to support the Cray XE and Cray XK software development environment. Cray XE and Cray XK ESL releases are tied to specific CLE releases. 1.5.4.2 Cray XC30 CDL Software The Cray XC30 ESL software release provides SLES 11, SP2 as a base operating system. Lustre client software is installed to support Lustre file systems. OFED is built by Cray from CLE and OFED. The Cray Linux Environment (CLE) 5.x support package and the Cray Developer Toolkit (CDT) software is installed to support the Cray XC30 software development environment. Cray XC30 ESL releases are tied to specific Cray CLE releases. 1.5.5 Cray Lustre File System (CLFS) Software The Cray ESF software release includes Community Enterprise Operating System (CentOS) 6 and Lustre software obtained directly from Wamcloud. Cray builds the Linux kernel and OFED package. 1.5.6 CLE, CDT, and CADE The Cray Linux Environment (CLE) is installed on CDL nodes for user application development and testing. S–2327–B 19 Data Management Platform (DMP) Administrator's Guide Cray Developer Toolkit (CDT) supports Cray XC30 systems, and the Cray Application Developer Environment (CADE) supports Cray XE and Cray XK systems. This software is installed from the Cray Linux Environment (CLE) Support Package software distribution. Note: All of the Lustre software components for a DMP system are currently provided with the CLE software. Refer to Installing CLE Support Package on a Cray Development and Login (CDL) Node for more information about the CLE4.x and CLE5.x releases for managed or unmanaged CDL nodes. 1.5.7 Installing DMP Software All Cray DMP software is installed from Cray DVD media or ISO file images. Software installation procedures are documented for each type of DMP node. Refer to the installation documentation listed in Related Publications on page 15. 1.5.7.1 Distribution Media The following release media is available for DMP system nodes: ESM Release Media (CIMS Nodes) • The Bright release media contains the Bright software and the SLES 11, SP2 operating system for the CIMS node. This item is available as the ISO bright6.0-sles11sp2.iso. Note: An initial installation requires both the ISO and the bootable DVD created from this ISO. The DVD is used to boot the CIMS and install the Bright software during an initial installation. • The Cray ESM release media also contains SLES security updates and CIMS software and tools. This item is available as an ISO such as ESM-XX-N.N.N-DatestampVer.iso. Note: Cray uses the following convention for ISO file names: PROD-ARCH-N.N.N-DatestampVer.iso PROD Product name, such as ESM ARCH Supported architecture: XX for all architectures, XC for Cray XC30 systems, and XE for Cray XE and XK systems N.N.N Release number, such as 2.1.0 for ESM XX-2.1.0 Datestamp Unique date stamp in the ISO name, in the format YYYYMMDDHHmm (such as 201309042055) Ver Installer version, such as a12 20 S–2327–B Introduction [1] ESL Release Media • The Cray ESL release media contains the CDL node software and tools for either the Cray XE, Cray XK, or Cray XC30 systems. For ESL XC-N.N.N, this item is available as the ISO such as ESL XC-1.3.0-DatestampVer.iso For ESL XE-N.N.N, this item is available as the ISO such as ESL XE-2.1.0-DatestampVer.iso • Also on the ESL release media is the SUSE Linux Enterprise Server operating system for either the Intel® or AMD® platforms. For ESL XC-N.N.N software releases (Intel platform), this item is available as an ISO such as : SLES-11-SP2-DVD-x86_64-GM-DVD1.iso For ESL XE-N.N.N software releases (AMD platform), this item is available as an ISO such as : SLES-11-SP1-DVD-x86_64-GMC3-DVD1.iso • CLE release media, which contains the CLE Support Package required for the Cray developer environment: • SUSE Linux Enterprise Software Development Kit (SDK) For ESL XC-N.N.N, this item is available as the ISO: SLE-11-SP2-SDK-DVD-x86_64-GM-DVD1.iso For ESL-XE-N.N.N, this item is available as the ISO: SLE-11-SP1-SDK-DVD-x86_64-GMC3-DVD1.iso • The Cray Developer Toolkit (CDT) or Cray Application Developer Environment (CADE): – For ESL XC-N.N.N, use the Cray Developer Toolkit (CDT) release media. – For ESL XE-N.N.N, use the Cray Application Developer's Environment (CADE) release media. ESF Release Media (CLFS Nodes) • The Cray ESF software release media contains CLFS software and CentOS 6.4. For ESF XX-N.N.N, this item is available as an ISO such as ESF-XX-2.1.0-DatestampVer.iso S–2327–B 21 Data Management Platform (DMP) Administrator's Guide 1.5.8 DMP Tools and Utilities Refer to the UNIX® man pages for each of the utilities in this section for command line options. 1.5.8.1 esdumpsys The esdumpsys command performs a dump of the specified CLFS and/or CDL in a Cray DMP system. esdumpsys can optionally generates a system memory dump if kdump is configured on the specified system(s). Refer to Configuring Kdump on CDL Nodes (SLES) on page 148 to configure kdump for SLES. Refer to Configuring Kdump on CentOS™ (Optional) on page 181 to configure kdump for CentOS. If kdump is not configured, the nodes will crash and reboot, but no memory dump will be generated. Information prints to the console in either case. The esdumpsys command uses ssh and scp to gather information from the specified systems. Password-less ssh is used to connect to the specified systems. 1.5.8.2 ESMupdateimage The ESMupdateimage command updates slave node images from the ESM media. Whenever you update your ESM software, and not your slave nodes, always run ESMupdateimage to update slave node RPMs. Refer to the man page or Updating Slave Node RPMs from ESM Media on page 59 for more information about the ESMupdateimage command. 1.5.8.3 eswrap The eswrap utility is a wrapper that lets users access a subset of Cray Linux Environment (CLE) and Programming Environment (PE) commands from a CDL node. eswrap uses Secure Shell (ssh) to launch the wrapped command on the Cray system, then displays the output on the CDL node so that it appears to the user that the wrapped command is actually running on a Cray internal login node. The CDL installation process creates a symbolic link for each wrapped command in the directory /opt/cray/eslogin/eswrap/default/bin. Each symbolic link points to the eswrap command, so that running a wrapped command (such as apstat) actually runs eswrap with the wrapped command as an argument. eswrap uses ssh to run the command on the specified node of the Cray system (by default, the internal login node with the host name login, unless the $ESWRAP_LOGIN environment variable specifies a different node). Refer to the man page for eswrap on the CDL for more information. 22 S–2327–B Introduction [1] 1.5.8.4 cray-esfs-catman The cray-esfs-catman utility is a script that helps change the Bright category settings for multiple CLFS nodes at the same time. cray-esfs-catman creates and runs the necessary Bright cmsh commands to change a category setting for either metadata server (MDS) or object storage server (OSS) nodes in the specified Lustre file system. 1.5.8.5 esfsmon_failback The esfsmon_failback command returns a failed CLFS node to operational status. When a CLFS node has been failed over to its backup node, the failed node is automatically powered down and placed into a failed node category. After the failed CLFS node has been repaired, the administrator must use esfsmon_failback to return the node to service. Refer to the esfsmon_failback man page on the CIMS for more information. Note: You must be root user to run the esfsmon_failback command. Refer to Configuring CLFS Failover (esfsmon) on page 165 for more information. 1.5.8.6 update_excludelist The update_excludelist script changes a Bright exclude list for all slave nodes in the specified category or categories. An exclude list controls which files in a slave image are retained or excluded during image synchronization, such as when a slave node is rebooted. update_excludelist composes and issues the necessary cmsh commands to change all nodes in a category at the same time. Refer to the man page on the CIMS for more information. 1.6 DMP Networks Overview Cray DMP software uses internal and external designations to classify networks. The esmain-net, ipmi-net, for example are internal networks accessible only to the CIMS. External networks in a DMP system are site-user-net, and site-admin-net, which enable users from outside the system to gain access. Refer to Network Settings on page 91 for common network configuration tasks. Figure 2 shows an overview of the hardware components and networks used in a Cray DMP system. The list below describes the primary networks used in a DMP system. The Bright software provides built-in classifications for the various networks in a system, and a Cray DMP system uses primarily the internal, external, and management classifications. There are other network classifications within Bright, such as cloud and global, but these are not used. There may be additional networks defined, depending on the requirements of the system. S–2327–B 23 Data Management Platform (DMP) Administrator's Guide esmaint-net An internal management network that connects the CIMS server with the slave nodes, switches, and RAID controllers. This network enables Bright to manage and provision the slave nodes and other devices in the DMP system. When using the Bright GUI (cmgui) or Cluster Management Shell (cmsh) this network is classified as the internal management network. ipmi-net Internal Intelligent Platform Management Interface (IPMI) or DELL™ Remote Access Controller (DRAC) network that provides remote console and power management of the slave nodes from the CIMS. site-admin-net External administration network used by site administrators to log in to the CIMS server (typically on the same network as the Cray SMW). The name and IP address of this network are customized during installation. Note: The CIMS IPMI interface (DRAC) may also be on this network to provide remote console and power management of the CIMS server. site-user-net External user (site) network used by the slave nodes. On CDL nodes, this network provides user access and authentication services such as LDAP. On CLFS MDS nodes, this network connects to the site LDAP for file ownership authentication. The name and IP address of this network are customized during installation. Note: Connections to additional site-specific networks are optional. wlm-net External user (site) network used by CDL nodes to access Cray SDB node or Cray internal login nodes. ib-net Internal InfiniBand® network used by the slave nodes for Lustre LNET traffic. failover-net Internal failover network used between two CIMS servers in an HA configuration for heartbeats between the active/passive CIMS nodes. This network does not connect to a managed switch. Note: Depending on system configuration, additional networks may be required. 24 S–2327–B Introduction [1] Figure 2. DMP Hardware and Networks Overview SAS/FC RAID controller A B OST OST SAS/FC A B MDT Ethernet switch eth0 eth2 eth0 eth2 eth0 DRAC eth2 eth0 DRAC eth2 esmaint-net esmaint-net ipmi-net ipmi-net IB IB eth0 DRAC eth1 eth0 DRAC eth1 IB IB eth0 ipmi-net DRAC IB eth0 DRAC IB esmaint-net ib-net esmaint-net esmaint-net External user site-user-net to NIS, LDAP server site-user-net site-user-net VLAN 4 eth1 DRAC DRAC eth1 SMW Workstation or PC Bright Cluster Manager site-admin-net site-admin-net VLAN 3 ipmi-net VLAN 2 esmaint-net VLAN 1 Primary CIMS Secondary CIMS failover-net eth3 eth3 CLFS MDS CLFS MDS CLFS OSS CLFS OSS IB switch Switch wlm-net eth1 CDL node wlm-net eth1 Cray SDB node Ethernet switch or VLAN Notes: 1. 4 CDL nodes can support 1 Cray internal login node 2. More than 2 OSS nodes per storage array requires a switch Cray system Cray LNET Cray LNET Cray LNET Cray LNET Cray LNET Cray LNET Internal login node Internal login node 1 2 CDL node ib-net (internal) esmaint-net (internal) ipmi-net (internal) site-user-net (external) site-admin-net (external) failover-net (internal) wlm-net (external) 1.6.1 CIMS Network Configuration The CIMS node requires the network interfaces shown in Figure 3 and listed in Table 2. Depending on the system-specific network configuration, additional interfaces may be required. The esmaint-net (eth0 interface) is a private network that connects all the managed devices in a DMP system and is defined as 10.141.x.x. The primary function of esmaint-net is to enable node provisioning and management. It may be helpful to follow the IP addressing scheme shown in Figure 3 to enable you to managed the various devices on the 10.141 network. The ipmi-net (eth2 interface) is a private network enables power control and remote console for all S–2327–B 25 Data Management Platform (DMP) Administrator's Guide of the slave nodes in the system. The site-admin-net (eth1 interface) is the site administration network. The CIMS IPMI interface (DRAC) may also be on this network to provide remote console and power management of the CIMS node. The failover-net (eth3 interface) is a private network used to direct heartbeats between CIMS nodes in an HA configuration. Figure 3. CIMS Network Interfaces and Default Addresses eth2 10.148.255.254 ipmi-net Primary CIMS eth3 10.50.0.1 failover-net eth1 aaa.bbb.ccc.ddd ipmi0 iDRAC site-admin-net Secondary CIMS eth0 10.141.255.253 esmaint-net eth0 10.141.255.254 esmaint-net eth2 10.148.255.253 ipmi-net eth1 site-admin-net ipmi0 iDRAC eth3 10.50.0.2 failover-net esmaint-net 10.141.0.0/16 10.141.0.x Server nodes 10.141.50.x ENet Switches 10.141.100.x RAID controllers 10.148.0.x Slave node BMC ports aaa.bbb.ccc.ddd Site admin node running Bright ipmi-net 10.148.0.0/16 eth1:0 alias - aaa.bbb.ccc.ddd (active CIMS in HA) eth0:0 alias - 10.141.255.250 (active CIMS in HA) 10.141.200.x IB Switches 10.141.150.x Managed FC/SAS Switches eth1:0 alias - aaa.bbb.ccc.ddd (active CIMS in HA) eth0:0 alias - 10.141.255.250 (active CIMS in HA) 26 S–2327–B Introduction [1] Table 2. CIMS Network Interfaces Interface Network Description eth0 esmaint-net Interface to the internal management network for maintenance and provisioning. The IP address for this network is set to 10.141.0.0/16. Other items on this network have the following IP addresses: 10.141.0.x Server node addresses. 10.141.50.x Managed Ethernet switch addresses. Note: The switch management IP assignments start at 10.141.50.1. 10.141.100.x Storage array controller addresses. 10.141.150.x Fibre Channel (FC) or serial attached SCSI (SAS) switches. 10.141.200.x InfiniBand® (IB) switches. On a system with two CIMS nodes in an HA configuration, the following IP addresses are used for eth0: 10.141.255.254 Primary CIMS 10.141.255.253 Secondary CIMS 10.141.255.250 eth0:0 on both CIMS servers in HA configuration. eth1 site-admin-net Interface to the administration network for the CIMS node (typically on the same network that the SMW is on). In an HA configuration, the alias eth1:0 is configured to connect to the active CIMS node. eth2 ipmi-net Interface to the remote console and power management network. Its IP address is set to 10.148.0.0/16. S–2327–B 27 Data Management Platform (DMP) Administrator's Guide Interface Network Description 10.148.255.254 Primary CIMS 10.148.255.253 Secondary CIMS eth3 failover-net Interface to the internal failover network for CIMS nodes in an HA configuration. The following IP addresses are assigned: 10.50.0.1 Primary CIMS 10.50.0.2 secondary CIMS ipmi0 (DRAC port) site-admin-net Remote console and power management of the CIMS (typically on the same network that the SMW is on). 1.6.2 CDL Network Configuration Note: BOOTIF is a special name for the eth0 interface. The node installer automatically translates BOOTIF into the name of the device (such as eth0), used for network booting. There can be only one interface configured as BOOTIF. Figure 4. CDL Network Interfaces and Default Addresses iDRAC 10.148.0.x eth0 10.141.0.x eth2 aaa.bbb.ccc.ddd site-user-net eth1 aaa.bbb.ccc.ddd ib0 10.149.0.x esmaint-net BOOTIF wlm-net ib-net CDL node Ethernet switch or VLAN IB switch VLAN 4 or site user gateway VLAN 2 or separate switch VLAN 1 or separate switch ipmi-net BMC ipmi0 1.6.3 CLFS Network Configuration Note: BOOTIF is a special name for the eth0 interface. The node installer automatically translates BOOTIF into the name of the device (such as eth0), used for network booting. There can be only one interface configured as BOOTIF. 28 S–2327–B Introduction [1] Figure 5. CLFS Network Interfaces and Default Addresses FC/IB RAID controller A RAID controller A 10.141.100.4 RAID controller B RAID controller B 10.141.100.3 RAID controller B 10.141.100.1 RAID controller A 10.141.100.2 OST OST FC/IB/SAS DRAC ipmi-net eth1 site-user-net DRAC ipmi-net eth1 site-user-net DRAC ipmi-net IB ib-net DRAC ipmi-net IB ib-net esmaint-net esmaint-net ipmi-net site-user-net ib-net esmaint-net CLFS OSS001 CLFS OSS002 FC or IB switch1 eth0 10.141.0.4 eth0 10.141.0.5 10.148.0.4 IB 10.149.0.2 IB 10.149.0.3 10.149.0.4 10.148.0.5 10.149.0.5 esmaint-net BOOTIF esmaint-net BOOTIF esmaint-net BOOTIF esmaint-net BOOTIF 10.141.150.1 CLFS MDS001 eth0 10.141.0.2 10.148.0.2 aaa.bbb.ccc.ddd CLFS MDS002 eth0 10.141.0.3 10.148.0.3 aaa.bbb.ccc.ddd A B MDT RAID controllers monitored by Bright SNMP and managed by 3rd-party software (NetApp SANtricity, DDN DirectMon) To CIMS Note: 1. More than 2 OSS nodes per storage array requires a switch S–2327–B 29 Data Management Platform (DMP) Administrator's Guide 1.7 Hardware Components A DMP system is comprised of specialized service nodes (see Figure 2), network switches, and storage arrays. These devices include, but are not limited to: • 1U or 2U rack-mounted servers configured with the necessary hardware and software to perform a specific role in the system, such as a software development platform or file server node • Ethernet switches to provide connectivity, maintenance access, and zones (VLANs) for each of the networks in the system • InfiniBand® (IB) switches for high-speed network connectivity • Fibre Channel (FC) or serial-attached SCSI (SAS) switches typically used for storage networks • Lustre-based storage arrays • Uninterruptible power sources or other power management equipment 1.7.1 Cray Management Server (CIMS) Hardware Overview The CIMS runs the Bright software and provides a centralized platform to manage the system hardware and software. Nodes that are managed by the CIMS are called slave nodes. Figure 6 show a generic representation of a 2U CIMS node. A DVD drive provides a method for installing software from DVD release media. All of the Cray Data Management Platform (DMP) system management software, slave node software, and software updates are installed from the CIMS DVD drive. 30 S–2327–B Introduction [1] A CIMS node is typically configured with six 1-TB disks, configured as two RAID-5 virtual disks (/dev/sda and /dev/sdb). PCIe slots can be used to add IB, FC, SAS, or GigE connectivity to the CIMS. All slaves nodes in a DMP system are managed, monitored, and provisioned by the CIMS over the esmaint-net network. An Intelligent Platform Management Interface (IPMI) network (ipmi-net) is used to control power and monitor hardware using simple network monitoring protocol (SNMP). Figure 2 shows how the CIMS is connected to other nodes in the system either through a GigE switch divided into VLANs, or via separate GigE switches. Figure 6. CIMS Node Hardware Overview iDRAC DVD PCIe slots /dev/sda /dev/sdb 6 TB RAID-5 eth3 failover-net eth1 site-admin-net eth2 ipmi-net eth0 esmaint-net GigE NICs VLAN 3 site-admin-net VLAN 3 site-admin-net VLAN 2 ipmi-net failover-net VLAN 1 esmaint-net PCIe connectivity FC8 XC-N2FC8-2P GigE XC-N2GE-4CU 10GbE XC-N2GT-2xx IB FDR XC-N3BF14-2P 6Gb SAS XC-N2S6-2P ESM XX-n.n.n ESL XC-n.n.n ESL XE-n.n.n ESF XX-n.n.n Bright Cluster Manager SLES-11 CentOS 6 etc... ISOs USB keyboard USB mouse Monitor CIMS node The CIMS disk partitioning is configured differently for a stand-alone CIMS or a high-availability (HA) CIMS during the initial installation procedure of the Bright and ESM software. The HA CIMS configurations make use of Distributed Replicated Block Device (DRBD) shared storage. S–2327–B 31 Data Management Platform (DMP) Administrator's Guide 1.7.2 Cray Development and Login (CDL) Node Hardware Overview User development and login (CDL) nodes provide the same programming environment as an internal login node on a Cray system. Each CDL node operates independently of the Cray system and are capable of accessing the same shared file system. The site-user-network provides user authentication and user access. The esmaint-net network enables the CDL node to use the pre-boot execution environment (PXE), or PXE boot, and together with the ipmi-net network, provides the means to manage and control the node. Figure 7. CDL Node Hardware Overview iDRAC PCIe slots /dev/sda 900-GB RAID-1 eth1 wlm-net eth2 site-user-net eth0 esmaint-net GigE NICs Cray internal login/MOM node VLAN 2 ipmi-net VLAN 4 site-user-net VLAN 1 esmaint-net PCIe connectivity FC8 XC-N2FC8-2P GigE XC-N2GE-4CU 10GbE XC-N2GT-2xx IB FDR XC-N3BF14-2P 6Gb SAS XC-N2S6-2P PXE boot IB IB switch to shared storage CDL node DVD The iDRAC port (ipmi-net) enables remote console and power management control from the CIMS node. There are two 900-GB disk drives in a CDL node that are configured as a RAID1. Two file systems master:/cm/shared and master:/home are NFS® mounted from the CIMS node (master). The master device can be used as an alias to designate the primary CIMS node. 1.7.3 Lustre® File Server (CLFS) Node Hardware Overview File system nodes (CLFS) provide a high-performance Lustre shared file system that operates independently of the Cray system. A DMP file system is comprised of metadata server nodes (MDS), object storage servers (OSS) and object storage target (OST) devices from DataDirect™ Networks (DDN) or NetApp™ within an InfiniBand® interconnect. Each CLFS node file system requires its own set of servers and block storage and are always configured with full redundancy in controllers, servers, and cabling for high-availability. 32 S–2327–B Introduction [1] The CLFS OSS servers are typically connected to the DDN or NetApp block storage controllers. Figure 8. CLFS OSS Node Hardware Overview iDRAC PCIe Gen3 slots /dev/sda 600-GB RAID-1 GigE NICs CLFS OSS node eth0 esmaint-net esmaint-net VLAN 2 ipmi-net VLAN 1 esmaint-net PCIe connectivity FC8 XC-N2FC8-2P GigE XC-N2GE-4CU 10GbE XC-N2GT-2xx IB FDR XC-N3BF14-2P 6Gb SAS XC-N2S6-2P PXE boot To LNET router on Cray system ib0 IB switch (LNET) SFA12K Controller 0 SFA12K Controller 1 ib3 OST OST ib2 DVD CLFS MDS servers are typically connected to the NetApp block storage controllers via InfiniBand® (IB) and are configured with a Mellanox dual-port InfiniBand (IB) host bus adapters (HBAs) installed in the CLFS node's PCIe GEN3 slot. CLFS MDS servers also connect to Cray XC30 systems through fourteen data rate (FDR) IB connection, and thus, include a Mellanox dual-port FDR IB HCA installed in a PCIe GEN3 slot. S–2327–B 33 Data Management Platform (DMP) Administrator's Guide Figure 9. CLFS MDS Hardware Overview iDRAC PCIe Gen3 slots /dev/sda 600-GB RAID-1 GigE NICs CLFS MDS node eth2 site-user-net VLAN 4 site-user-net (site LDAP) eth0 esmaint-net Cntrl B Cntrl A VLAN 2 ipmi-net VLAN 1 esmaint-net PCIe connectivity FC8 XC-N2FC8-2P GigE XC-N2GE-4CU 10GbE XC-N2GT-2xx IB FDR XC-N3BF14-2P 6Gb SAS XC-N2S6-2P PXE boot To LNET router on Cray system IB IB switch (LNET) SAS NetApp 2600 Controller A NetApp 2600 Controller B MDT DVD 1.7.4 Switches and PDUs Ethernet, InfiniBand, and Fibre Channel switches communicate with Bright using Simple Network Management Protocol (SNMP) using the device management port, which typically an Ethernet connection to esmaint-net. Other devices such as power distribution units (PDUs) can also be configured and managed by Bright. The SNMP community strings should be configured to public read, and private write access. Other device configuration settings such as administrative password, host name, and IP address should be configured using the device console configuration commands. Uplink ports (switch ports that are connected to other switches) must be configured in Bright. The CMDaemon (cmd) must be told about any switch ports that are uplink ports, or the traffic passing through an uplink port will lead to mistakes in what cmd knows about port and MAC correspondence. Uplink ports are thus ports that cmd is told to ignore. 34 S–2327–B Bright Cluster Manager® [2] 2.1 Managing a System with Bright Refer to the Bright Cluster Manager 6.0 Administrator Manual for detailed information about Bright software management. PDF files for the Bright manuals are stored on the CIMS in /cm/shared/docs/cm, and linked to from the /root directory. This chapter provides and overview of how Cray implements Bright to manage a DMP system. A CIMS running Bright provides a management interface that is used by: • Cluster management shell (cmsh) — A command line interface to manage and control the system. • Cluster management user interface (cmgui) — A graphical user interface (GUI) to manage and control the system. • Cluster management daemon (cmd) — CMDaemon runs on all nodes in the DMP system. Either the cmgui or cmsh can be used to manage the system, and you may find the certain tasks are more easily visualized using cmgui, and other tasks are more efficient using the cmsh. General information about how to use the cmgui are described in Bright GUI on page 49. 2.1.1 Devices and Device Names in Bright A device in a DMP system represents a physical hardware component. A device can be any of the following types: • Head nodes — Typically named esms1, esms2. • Ethernet switches — Typically named according to their function in the cluster such as switch-esfs1, switch-esmaint-net, switch-eslogin. • InfiniBand® switches — Typically named switch-ib1-scratch, switch-ib2-scratch. • Fibre Channel switches — Typically named fc-switch1, fc-switch2. • Slave nodes — Typically named according to their function in the cluster, such as eslogin-xe-001 (login node), esfs-mds001 (metadata server), esfs-oss001 (object storage server), esfs-ost001, (object storage targets). S–2327–B 35 Data Management Platform (DMP) Administrator's Guide • Storage array RAID controllers — Storage controllers are added as a generic device in cmsh or under the Other resource in the cmgui. Typically named by manufacturer model number, rack location or purpose, netapp5400-cntlA, netapp3992-cntlA, rack1-ddnsfa12k-cntrl0. 2.1.2 Node Organization Cray DMP systems along with Bright software support the concept of node categories and node groups. Categories specify a number of parameters that are common to all members. Among these are the management network, as well as the software image and scripts that are run by the node-installer to customize each node's image during provisioning. Category parameters can be overridden on a per node basis, if desired. Slave nodes can belong to several different node groups, and there are no parameters associated with node groups. Node groups are typically used to invoke commands across several nodes simultaneously. The Bright software configures a separate interface for each node because the IP addresses that Bright uses are specific to each node. Software images are common across multiple nodes, so the Bright interface files must reside in the Bright database and be placed on other nodes at boot time. The node category defines which software image is provisioned to its member nodes and other management attributes. Note: A node can belong only to one node category. Table 3 lists Cray DMP node categories. Table 3. DMP Slave Node Categories Category Description esLogin-XE Configures the software image and settings for all AMD® platform CDL nodes that run the production CDL image. esLogin-XC Configures the software image and settings for all Intel® platform CDL nodes that run the production CDL image. esFS-MDS Configures the software image and settings for all the metadata server (MDS) nodes. This category is created by ESFinstall and is not used under esfsmon. 36 S–2327–B Bright Cluster Manager® [2] Category Description esFS-OSS Configures the software image and settings for all the object storage server (OSS) nodes. This category is created by ESFinstall and is not used under esfsmon. esfs-mds-fo-filesystem Contains the esfsmon secondary MDS node for the file system named by filesystem. esfs-mds-failed-filesystem Contains any esfsmon failed MDS node for the file system named by filesystem. esfs-oss-even- filesystem Contains the esfsmon even-numbered object storage server (OSS) nodes for the file system named by filesystem. esfs-oss-odd-filesystem Contains the esfsmon odd-numbered OSS nodes for the file system named by filesystem. esfs-oss-failed-filesystem Contains any esfsmon failed OSS node for the file system named by filesystem. Most importantly, the node category determines which software image a node runs. Node categories also provide control over several other parameters such as: revision Object revision bmcpassword Password used to send ipmi/ilo commands to nodes. The baseboard management controller (BMC or iDRAC) password is inherited from the base partition and typically not set for the node category. bmcusername User name used to send ipmi/ilo commands to nodes. Inherited from the base partition, and typically not set for the category. defaultgateway Default gateway for the category filesystemexports Configure the entries placed in /etc/exports S–2327–B 37 Data Management Platform (DMP) Administrator's Guide filesystemmounts Configure the entries placed in /etc/fstab installbootrecord Install boot record on local disk installmode installmode to be used by default, if none is specified in the node ipmipowerresetdelay Delay used for ipmi/ilo power reset, default is 0 managementnetwork Determines what network should be used for management traffic. If not set, partition setting is used. name Name of category nameservers List of name servers the category will use newnodeinstallmode Default install mode for new nodes roles Assign the roles the node should play searchdomain Search domains for the category services Manage operating system services softwareimage Software image the category will use timeservers List of time servers the category will use usernodelogin ALWAYS or NEVER allow a user to log in to the node disksetup Disk setup for nodes excludelistfullinstall Exclude list for full install 38 S–2327–B Bright Cluster Manager® [2] excludelistgrab Exclude list for grabbing to an existing image excludelistgrabnew Exclude list for grabbing to a new image excludelistsyncinstall Exclude list for sync install excludelistupdate Exclude list for update finalizescript Finalize script to be used for category initializescript Initialize script to be used for category notes Administrator notes Node groups simplify management and control activities and enable you to perform commands on a group of nodes simultaneously. Table 4 lists typical node groups: Table 4. Node Groups Node Group Description login All CDL nodes oss All OSS nodes esfs-filesystem All CLFS nodes for the file system named by filesystem 2.1.3 Software Image Management A software image is a blueprint for the contents of the local file systems on slave nodes. Software images reside in /cm/images on the CIMS and contain a full Linux™ file system. When a slave node boots, the node provisioning system configures the node with a copy of its assigned software image determined by the node category. After the node has booted, it is possible to instruct the node to resynchronize its local file systems with the software image. This procedure can be used to distribute changes to the software image without rebooting nodes. It is also possible to lock a software image so that no node is able to pick up the image until the software image is unlocked. Software images can be changed using Linux tools and commands such as rpm and chroot. S–2327–B 39 Data Management Platform (DMP) Administrator's Guide Important: The software images in /cm/images should be backed-up regularly, as they are needed in the event you would have to re-install the CIMS software and reload the system configuration. Refer to Save the System Configuration on page 126 for more information about saving the system configuration. Software images are typically named in a way to be easily identified, beginning with ESM, ESL, or ESF. To list the software images that are installed on the CIMS, use cmsh to switch to softwareimage mode, and enter list to display the images. Example 1. List Software Images esms1# cmsh [esms1]% softwareimage [esms1->softwareimage]% list Name (key) Path Kernel version -------------------------- ---------------------------------------- --------------------------- ESF-XX-2.0.0-201302121343 /cm/images/ESF-XX-2.0.0-201302121343 2.6.32-279.14.1.el6.x86_64 ESF-XX-2.0.0-201304181540 /cm/images/ESF-XX-2.0.0-201304181530 2.6.32-279.14.1.el6.x86_64 ESL-TEST-4102013 /cm/images/ESL-TEST-4102013 3.0.38-0.5-default ESL-XC-1.0.2-201302211318 /cm/images/ESL-XC-1.0.2-201302211318 3.0.38-0.5-default ESL-XE-1.1.1-201211150916 /cm/images/ESL-XE-1.1.1-201211150916 2.6.32.59-0.7-default ESL-XE-1.1.1-kdump /cm/images/ESL-XE-1.1.1-kdump 2.6.32.59-0.7-default ESL-XE-1.1.1_CLE4.1 /cm/images/ESL-XE-1.1.1_CLE4.1 2.6.32.59-0.7-default default-image /cm/images/default-image 3.0.58-0.6.6-default default-image.previous /cm/images/default-image.previous Note that default-image and default-image.previous are images created by the Cray installer software, ESMinstall. The ESLinstall and ESFinstall installers configure software images for Cray Development and Login (CDL) and Cray Lustre File System (CLFS) nodes. The Cray installer software adds the latest released software and updates to the /cm/images directory on the CIMS when the installation completes, and also configures the image in the Bright database. Installed images will use the naming conventions in Distribution Media on page 20. Important: The default image (named default-image) should never be modified. Always clone default-image or production image to create a new image for a node. Always keep a functioning image as a backup. When a node boots, the node provisioning system sets up the node with a copy of the software image that is configured for the node. Software images are assigned to a category in Bright. Nodes are also assigned to a category in Bright. This enables the administrator more control over what software image is used by a node. After you have installed and configured an image, use the Bright clone command to create an archive of the old image. All software images in /cm/images on the CIMS should be backed up routinely. 2.1.4 Node Provisioning Node provisioning is the process of how nodes obtain a software image and occurs during power-up or when updating a running node. 40 S–2327–B Bright Cluster Manager® [2] 2.1.4.1 Preboot Execution Environment (PXE) Booting By default, slave nodes boot from the esmaint-net network over the BOOTIF interface. Bright controls this network boot or Preboot Execution Environment (PXE) boot. PXE booting is configured in the slave node BIOS setting (Refer to Appendix A, Changing BIOS for a DELL™ R720 Managed CDL Node on page 253 through Appendix C, Changing BIOS for a DELL™ R720 Managed CLFS Node on page 269. The CIMS runs a tftpd server process from within xinetd, which supplies the boot loader from within the default software image offered to nodes. The boot loader runs on the node and displays a menu based on loading a menu module within a configuration file. The configuration file is located in the default-image software image in the /cm/images/default-image/boot/pxelinux.cfg/ directory, default file. Figure 10. PXE Boot Menu The MENU DEFAULT value in the software image /cm/images/default-image/boot/pxelinux.cfg/default file is loaded for every node using the software image. To override its application on a per-node basis, the value of PXE Label can be set for each node. For example, to set a node to use the MEMTEST PXE label using cmsh: Example 2. Set PXE label for a node esms1# cmsh [esms1]% device use eslogin001 [esms1->device[eslogin001]]% set pxelabel MEMTEST ; commit To use the MEMTEST label for all nodes the esLogin-XE category use: [esms1->device]% foreach -c esLogin-XE (set pxelabel MEMTEST) [esms1->device*]% commit S–2327–B 41 Data Management Platform (DMP) Administrator's Guide Refer to the Bright Cluster Manager 6.0 Administrator Manual for more information. 2.1.4.2 Booting From the Hard Drive In addition to PXE boot, a node can also be provisioned entirely from its local drive, just like a standalone server. Select Install Boot Record from the cmgui Node or Category resource, or set installbootrecord to yes from cmsh category mode. esms1# cmsh [exms1]% category use default [esms1->category[default]]% set installbootrecord yes [esms1->category*[default*]]% commit 2.1.4.3 The Boot Role The action of providing a PXE boot image via DHCP and TFTP is known as providing node booting. 2.2 Bright License Management Cray DMP systems are configured with a Bright Cluster Manager license file installed on the CIMS node. The license file includes: • A licensee attribute or the name of the organization, the condition in which the specified organization may use the software; a licensed nodes attribute specifies the maximum number of nodes that the cluster manager may manage. CIMS nodes are also regarded as nodes too for this attribute. • licensed nodes attribute specifies the maximum number of nodes that the cluster manager may manage. CIMS nodes are included. • An expiration date for the license A license file can only be used on the machine for which it has been generated and cannot be changed once it has been issued. The license file is the X509v3 certificate for the CIMS node and is used throughout system operations. 2.2.1 Displaying License Attributes The license file is installed in the following location on the CIMS: /cm/local/apps/cmd/etc/cert.pem The associated private key file is in: /cm/local/apps/cmd/etc/cert.key To verify that the attributes of the license have been assigned the correct values, use the GUI to select the CIMS node, License tab, to display license details. 42 S–2327–B Bright Cluster Manager® [2] Alternatively the cmsh main mode licenseinfo command displays: Example 3. Display license attributes in cmsh esms1:~ # cmsh [esms1]% main licenseinfo License Information ---------------------------- ------------------------------------------- Licensee /C=US/ST=Wisconsin/L=Chippewa Falls/O=Cray Training/OU=Training and Doc/CN=Training Serial Number 5510 Start Time Tue May 1 00:00:00 2012 End Time Fri Feb 8 23:59:00 2013 Version 6.0 Edition Advanced Pre-paid Nodes 512 Max Pay-as-you-go Nodes 1000 Node Count 6 MAC Address / Cloud ID 78:2B:CB:40:CE:CA The license in the example above enables 512 nodes. It is tied to a specific MAC address, and the Node Count field in the output of licenseinfo shows the current number of nodes used. 2.2.2 Verifying A License The verify-license utility determines whether a license is valid even when the cluster management daemon is not running. Example 4. Verifying a License esms1# /etc/init.d/cmd start Waiting for CMDaemon to start... CMDaemon failed to start please see log file. esms1# tail -1 /var/log/cmdaemon Dec 30 15:57:02 esms-1 CMDaemon: Fatal: License has expired 2.2.3 Installing the Bright License The initial installation of Bright is licensed only for three nodes, including the CIMS. A permanent license must be installed before configuring the system. • Use Procedure 1 on page 44 to install the license. • If you have added a second CIMS in a HA configuration, contact Cray to purchase another node license for the second CIMS. • If your existing license has expired and must reinstate your license. Certificate Sign Request (CSR) data is displayed and saved in the file /cm/local/apps/cmd/etc/cert.csr.new. The cert.csr.new file may be used with an internet-connected browser. S–2327–B 43 Data Management Platform (DMP) Administrator's Guide After using a product key with request-license, reboot the system using the pexec reboot command from the CIMS. If you are relicensing an existing system, a reboot is not required. Important: If licensing a secondary CIMS for an HA configuration, get the MAC address for eth0 of the secondary node. The MAC address for the secondary node is typically located on a label on the front panel of the node. Procedure 1. Installing the Bright license on a CIMS Obtain a product key from Cray which enables the administrator to obtain and activate a license. The product key looks like the following string: XXXXXX-XXXXXX-XXXXXX-XXXXXX-XXXXXX 1. Log in to the CIMS as root. remote% ssh root@esms1 2. Get the MAC address (HWaddr) for eth0 (BOOTIF) interface. esms1# /sbin/ifconfig eth0 eth0 Link encap:Ethernet HWaddr 78:2B:CB:40:CE:CA inet addr:10.141.255.254 Bcast:10.141.255.255 Mask:255.255.0.0 UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1 RX packets:11944661 errors:0 dropped:0 overruns:0 frame:0 TX packets:11018308 errors:0 dropped:0 overruns:0 carrier:0 collisions:0 txqueuelen:1000 RX bytes:2589377379 (2469.4 Mb) TX bytes:2060383201 (1964.9 Mb) Interrupt:36 Memory:d6000000-d6012800 3. Run the request-license command on the CIMS. Important: If you are configuring the primary CIMS, enter license information for the first CIMS only and do not enter the MAC address for the secondary CIMS in an HA configuration. You must run this procedure again when configuring the secondary CIMS and enter yes in step 4.c. esms1# request-license Product Key (XXXXXX-XXXXXX-XXXXXX-XXXXXX-XXXXXX): Note: If the license already exists, you are prompted to re-use the private key and settings from the existing license. 4. Enter the product key, then press Enter. Note: If you are licensing a secondary CIMS for an HA configuration you are prompted to re-use the private key settings from the existing license. Select yes at the following prompt and then proceed at step step 4.b. Re-use private key and settings from existing license? [Y/n] yes 44 S–2327–B Bright Cluster Manager® [2] a. Enter the Country Name 2-letter code, and other pertinent information when prompted: Country Name (2 letter code): US State or Province Name (full name): State Locality Name (e.g. city): City Organization Name (e.g. company): Company Organizational Unit Name (e.g. department): ACME Cluster Name: Training Private key data saved to /cm/local/apps/cmd/etc/cert.key.new b. Enter the MAC address of the primary CIMS for eth0 on (esmaint-net). If you are licensing a secondary CIMS in an HA configuration, enter the MAC address for eth0 for the secondary CIMS. MAC Address of primary head node (esms1) for eth0 []: FF:AE:FF:B5:E2:64 MAC Address of secondary head node (esms2) for eth0 []: FF:AE:FF:B5:E2:65 Note: Enter N or No in step 4.c if you are configuring the primary CIMS. If you have already configured the primary CIMS, and are now configuring the secondary CIMS in an HA configuration, enter y or yes and supply the required information to obtain an HA license certificate. c. If you are configuring the primary CIMS, enter no at the following prompt. If your are configuring an HA setup, enter yes at the following prompt. Will this cluster use a high-availability setup with 2 head nodes? [y/N]: no d. Answer no when you are prompted to submit the certificate request: Certificate request data saved to /cm/local/apps/cmd/etc/cert.csr.new Submit certificate request to http://support.brightcomputing.com/licensing/index.cgi ? [Y/n] no The CSR (Certificate Sign Request) data is displayed and saved in the file /cm/local/apps/cmd/etc/cert.csr.new on the CIMS. The cert.csr.new file may be used to obtain a license with an Internet-connected browser: Note: The license strings used below are fictitious. -----BEGIN CERTIFICATE REQUEST----- MIICBjCCAW8CAQAwgcUxCzAJBgNVBAYTAlVTMRIwEAYDVQQIEwlXaXNjb25zaW4x FzAVBgNVBAcTDkNoaXBwZXdhIEZhbGxzMRIwEAYDVQQKEwlDcmF5IEluYy4xDDAK BgNVBAsTA0JJUzESMBAGA1UEAxMJaHVzayBlc01TMSAwHgYJKoZIhvcNAQkCExFF MDpEQjo1NTowODpGRjpDMDExMC8GCSqGSIb3DQEJBhMiMDExNDI2LTUxMTQ3Mi0w MDI3NDYtODU3MzM2LTQxNTYyNDCBnzANBgkqhkiG9w0BAQEFAAOBjQAwgYkCgYEA scCs7/hIZF5ehPq0ZhGn/bVY8cO+e9KF8psJHu1cYVC1WCcFj04LMQztUVvftigI HWo+YZVJbuMHphvAc4BfXDhYxjLPVw+yxU9FBBBDyFZxuMJpCIhr8YAKxABVX0fS zKK6eE7Pj1G6Ho9vW6+sHOgzCF3jm4xG52NTTma+BQUCAwEAAaAAMA0GCSqGSIb3 DQEBBQUAA4GBAG6VBE0HRqSKP8CFAaJ3AwewtXEL7gotOYBAhfe2rMvl6/NWzFGD uCCju5psN5LpsgyhKQTWPDQwS7EbxRQ+jerHVcsI/ZEgnzVBozjvVgESVML8+yA0 6Dtba8hrqBFFtLXmm3KE+qQCt+vqGMUFs8g4D0GYOkThlg6auJFXFU3N -----END CERTIFICATE REQUEST----- S–2327–B 45 Data Management Platform (DMP) Administrator's Guide 5. On a system with Internet access, use a web browser to open http://support.brightcomputing.com/licensing/. 6. Copy CSR data obtained in step 4.a to obtain a license certificate. 7. Paste the CSR data (contents of cert.csr.new) into the web form, then select Submit. Figure 11. Bright License Request Form A signed license certificate is displayed (the following example is fictitious). -----BEGIN CERTIFICATE----- MIIDNzCCAh+gAwIBAgICFkwwDQYJKoZIhvcNAQEFBQAwbDELMAkGA1UEBhMCVVM CzAJBgNVBAgTAkNBMREwDwYDVQQHEwhTYW4gSm9zZTEfMB0GA1UEChMWQnJpZ2h IENvbXB1dGluZywgSW5jLjEcMBoGA1UEAxMTQnJpZ2h0IENvbXB1dGluZyBDQTA Fw0xMjA5MTIwMDAwMDBaFw0zODEyMzEyMzU5MDBaMGYxCzAJBgNVBAYTAlVTMRQ EgYDVQQIEwtNaXNzaXNzaXBwaTESMBAGA1UEBxMJVmlja3NidXJnMREwDwYDVQQ EwhDcmF5IEluYzENMAsGA1UECxMEc3RjbzELMAkGA1UEAxMCbmEwgZ8wDQYJKoZ hvcNAQEBBQADgY0AMIGJAoGBAN8lCM52TEnZ63yvxPvpe4WbBTPsFKUWOpIOHT8 tbctYf54E4K4A1A0ahX48OYdhafhTb7AO0gGSv/Vp+QxZQrkrSi6A8zwlNkrz4j yDDYFNb4sBkJUbexHmEdR5Bhp/xfEx4X7EkFb5vdnhIyiwvn6Zs5tZ/Sgt+CJcH KJFtAgMBAAGjbTBrMA8GA1UdEwEB/wQFMAMBAf8wHgYHKwYBBKFkIQQTFhE4NDo Rjo2OTpFMzo1Qjo1ODAPBgcrBgEEoWQiBAQWAjE2MBAGBysGAQShZCMEBRYDNS4 MBUGBysGAQShZCQEChYIQWR2YW5jZWQwDQYJKoZIhvcNAQEFBQADggEBAF1pRtg vjMr9TlihmEcO23raTLp308zkVFpW7vJ0T8KqEirwkzzrD83igtJNd2q6jtrpRL 3kxQ2sB0gGmuptHkrYecwZtVm5FhGOuPeDUG3ww4W+GyCkczCRtPkVTXul250Z9 LZqrK0zzPRKhMNEraTXPDEgHSEgEeykro30EGHpcuCoGCBJNvhi0bCIxgJoW5DV ykGbeE4DKBlyW0r4NRqMBR+0BH2d1gCXBJtYhHMfDUWw6JOMmsCcoAY7Yhp4N9k AJb++bi/pO2fQeDJopfxvlU2WsEEMcEItNklknaHlOYfQ3IcloH9w464MtGFakt xGjtfqxcU -----END CERTIFICATE----- 8. Copy the license text received and save it to a plain text file named signedlicensefile on the CIMS. 46 S–2327–B Bright Cluster Manager® [2] 9. Enter the following command to install the license and answer each prompt. esms1 # install-license signedlicensefile ========= Certificate Information ======== Version: 6.0 Edition: Advanced Common name: na Organization: ACME Organizational unit: Training Locality: City State: State Country: US Serial: 3728 Starting date: 12 Sep 2012 Expiration date: 31 Dec 2038 MAC address: 84:8F:E4:E3:5B:64 Pre-paid nodes: 16 Max Pay-as-you-go Nodes: N/A ========================================== Is the license information correct ? [Y/n] Y Backup directory of old license: /var/spool/cmd/backup/certificates/2013-01-28_08.56.22 In order to authenticate to the cluster using the Cluster Management GUI (cmgui), one must hold a valid certificate and a corresponding key. The certificate and key are stored together in a password-protected PFX (a.k.a. PKCS#12)file. Provide a password to protect the admin.pfx file holding the administrator certificate. Please provide a password that will be used to password-protect the PFX file holding the administrator certificate (/root/.cm/cmgui/admin.pfx). Password: Verify password: Installed new license Waiting for CMDaemon to stop: OK Installing admin certificates Waiting for CMDaemon to start: OK New license was installed. In order to allow nodes to obtain a new node certificate, all nodes must be rebooted. Please issue the following command to reboot all nodes: pexec reboot Note: If the license process fails, check /var/log/cmdaemon for failure information. Important: Refer to Managing Bright admin.pfx Certificates on page 62 for important information about how to manage the Bright admin.pfx file holding the administrator certificate. S–2327–B 47 Data Management Platform (DMP) Administrator's Guide 2.2.4 Reinstating an Expired License Procedure 2. Reinstating an expired license 1. Log in to the CIMS as root. # ssh root@cray-esms1 2. Run the request-license command on the CIMS. cray-esms1:~ # request-license Product Key (XXXXXX-XXXXXX-XXXXXX-XXXXXX-XXXXXX): 3. Enter the product key, then press Enter (the example is not a valid key). 714354-916786-132324-207440-186713 4. Answer each prompt to re-install the license. Existing license was found: ... Re-use private key and settings from existing license? [Y/n] y Will this cluster use a high-availability setup with 2 head nodes? [y/N] n MAC Address of primary head node for eth0 []: 78:2B:CB:40:CE:CA Certificate request data saved to /cm/local/apps/cmd/etc/cert.csr.new Submit certificate request to http://support.brightcomputing.com/licensing/index.cgi ? [Y/n] y Contacting http://support.brightcomputing.com/licensing/index.cgi... License granted. License data was saved to /cm/local/apps/cmd/etc/cert.pem.new Install license? [Y/n] y ========= Certificate Information ======== Version: 6.0 Edition: Advanced Common name: Training Organization: ACME Training Organizational unit: Training and Doc Locality: Chippewa Falls State: Wisconsin Country: US Serial: 5846 Starting date: 01 May 2012 Expiration date: 13 Mar 2013 MAC address: 78:2B:CB:40:CE:CA Pre-paid nodes: 512 Max Pay-as-you-go Nodes: 1000 ========================================== Is the license information correct ? [Y/n] y Backup directory of old license: /var/spool/cmd/backup/certificates/2013-02-10_11.34.53 Is this host the cluster's head node? [Y/n] y Installed new license Restarting Cluster Manager Daemon to use new license: OK 2.2.5 Reboot After Installing License After using a product key with request-license, reboot the system using the pexec reboot command from the CIMS. 48 S–2327–B Bright Cluster Manager® [2] 2.3 Bright GUI Refer to Installing and Running cmgui on a Remote System on page 53 for procedures to install the cmgui. To connect to a system, use the procedures in Running cmgui and Connecting to a DMP System on page 54. The Bright cmgui window provides a resource tree down the left side that lists all of the components in a system. Selecting a resource opens an associated tabbed pane on the right side of the window that allows tab-related parameters to be viewed and managed. The number of tabs displayed and their contents depend on the resource selected. When learning to use Bright software, the cmgui window may be easier to learn and understand as opposed to the shell (cmsh), conversely, the cmsh shell environment may be easier and more efficient to use for some tasks. Both user interfaces (cmsh and cmgui) provide the same administrative capabilities. Refer to the Bright Cluster Manager 6.0 Administrator Manual, is stored on the CIMS as a PDF file in /cm/shared/docs/cm. Figure 12. Bright cmgui Window 2.4 The Command Shell The cluster management shell (cmsh) provides a command-line interface to the system. The cmsh and the cmgui each provide the same capability. The command-line shell (cmsh) is invoked from an interactive session (through ssh) on the CIMS node, but cmsh can also be used to manage a cluster remotely. This section introduces the cmsh and provides examples of common tasks. S–2327–B 49 Data Management Platform (DMP) Administrator's Guide Enter the following command as the root user to start the cmsh on the CIMS node: esms1# cmsh [esms1]% When you run the cmsh from a UNIX® shell without arguments, it starts an interactive session. To return to the UNIX shell, enter the quit command: [esms1]% quit esms1# The cmsh can be used in batch mode by specifying a command using the -c flag. Commands can be separated using semi-colons (;). esms1# cmsh -c "main showprofile; device status eslogin01" admin eslogin-01 ............... [ UP ] esms1# The syntax for the cmsh is listed below: cmsh [options] ................ Connect to localhost using default port cmsh [options] <--certificate|-i certfile> <--key|-k keyfile> <host[:port]> Connect to a cluster using certificate and key in PEM format cmsh [options] <--certificate|-i certfile> [-password|-p password] <uri[:port]> Connect to a cluster using certificate in PFX format Valid options: --help|-h ..................... Display this help --noconnect|-u ................ Start unconnected --controlflag|-z .............. ETX in non-interactive mode --nossl|-s .................... Do not use SSL --norc|-n ..................... Do not load cmshrc file on start-up --command|-c <"c1; c2; ..."> .. Execute commands and exit --file|-f .......... Execute commands in file and exit --echo|-x ..................... Echo all commands --quit|-q ..................... Exit immediately after error Alternatively, commands can be piped to the cmsh from the UNIX® command line as root user: esms1# echo device status | cmsh eslogin01 ............... [ UP ] mycluster ............... [ UP ] oss001 .................. [ UP ] oss002 .................. [ UP ] switch01 ................ [ UP ] esms1# 50 S–2327–B Bright Cluster Manager® [2] The cluster management functions are grouped in separate cmsh modes. The first thing you must do when performing a cluster management operation is switch to the appropriate mode. The cmsh modes are listed below: disconnect .................... Disconnect from cluster connect ....................... Connect to cluster quit .......................... Quit shell exit .......................... Exit from current object or mode help .......................... Display this help run ........................... Execute cmsh commands from specified file alias ......................... Set aliases unalias ....................... Unset aliases modified ...................... List modified objects export ........................ Display list of aliases current list formats events ........................ Manage events list .......................... List state for all modes category ...................... Enter category mode cert .......................... Enter cert mode device ........................ Enter device mode jobqueue ...................... Enter jobqueue mode jobs .......................... Enter jobs mode main .......................... Enter main mode monitoring .................... Enter monitoring mode network ....................... Enter network mode nodegroup ..................... Enter nodegroup mode partition ..................... Enter partition mode process ....................... Enter process mode profile ....................... Enter profile mode session ....................... Enter session mode softwareimage ................. Enter softwareimage mode test .......................... Enter test mode user .......................... Enter user mode Type device at the cmsh prompt to enter device mode. [esms1]% device [esms1->device]% list Type Hostname MAC Ip -------------- --------------- -------------------- --------------- EthernetSwitch switch01 00:00:00:00:00:00 10.142.253.1 MasterNode mycluster 00:E0:81:34:9B:48 10.142.255.254 ossNode oss0 00:E0:81:2E:F7:96 10.142.0.1 ossNode oss2 00:30:48:5D:8B:C6 10.142.0.2 [esms1->device]% exit [esms1]% Most modes in cmsh require that you specify an object, for instance, device mode requires that you specify device objects such as esfs-mds1, or ib-switch-1, and network mode requires you to specify network objects such as esmaint-net or site-user-net. The commands that can be used for controlling objects are the same in all modes. Table 5 lists the commands that may be used to act on objects in a particular mode. S–2327–B 51 Data Management Platform (DMP) Administrator's Guide Table 5. Command Shell Object Descriptions Command Description use Make the specified object the current object add Create an object and make it the current object clone Clone an object and make it the current object remove Remove an object commit Commit local changes to an object to the cluster management infrastructure refresh Undo local changes to an object list List all objects format Set formatting preferences for list output show Display all properties of an object get Display a particular property of an object set Set a particular property of an object clear Set empty value for a particular property of an object append Append a value to a particular list-property of an object removefrom Remove a given value from a particular list-property of an object modified Lists objects with uncommitted local changes usedby Lists objects that depend on a particular object validate Perform validation-check on the properties of an object 2.4.1 Mixing cmsh and UNIX Shell Commands You can execute UNIX commands while you perform cluster management. The cmsh enables users to execute UNIX commands by prefixing the command with a ! character. Example 5. Mixing cmsh and UNIX Commands esms1# cmsh [esms1]% !hostname -f esms1.cm.cluster [esms1]% Executing the ! command only starts an interactive login sub-shell. When you exit the sub-shell, you return to the cmsh prompt. It is also possible to use the output of UNIX shell commands as part of a cmsh command by using the "backtick" syntax" that is available in most UNIX shells as shown in Example 6. 52 S–2327–B Bright Cluster Manager® [2] Example 6. Using UNIX output in cmsh commands [esms1->device]% device use h` ostname ;` status cf-esms01 ........... [ UP ] [esms1->device]% Similar to UNIX shells, cmsh also supports output redirection through common operators such as >, >> and |. While looping over objects it may be helpful to execute a cmsh command for several objects simultaneously. The foreach can be used in several cmsh modes which enables you to loop over a list of objects. A foreach command takes a list of object names separated by spaces, and a list of commands that must be enclosed by ( and ) characters. The foreach command iterates over the specified objects and executes commands for each loop iteration. Example 7 shows an example of the foreach command syntax: Example 7. Using a foreach loop to invoke commands [esms1->device]% foreach Object...Object ( Command; Command; ) [esms1->device]% foreach oss001 oss002 (get hostname; status) oss001 oss001 ............. [ UP ] oss002 oss002 ............. [ UP ] [esms1->device]% Refer to the Bright Cluster Manager 6.0 Administrator Manual for more information. 2.5 Installing and Running cmgui on a Remote System The Linux™ or Windows® installation software for the cluster management GUI (cmgui) is located in /cm/shared/apps/cmgui/dist on the CIMS. Procedure 3. Installing and running cmgui Important: Whenever you update a CIMS with the latest ESM software, always re-install the cmgui software on the remote systems so that software updates are applied. 1. Copy the Windows .exe file, which is similar to this filename: (install.cmgui.6.0.r4356.exe). Or copy the Linux compressed TAR file (cmgui-6.0-r4101-src.tar.bz2) from the /cm/shared/apps/cmgui/dist directory on the slave node to a tmp directory on the remote system. remote% scp root@esms1:/cm/shared/apps/cmgui/dist/* /tmp 2. Copy the PFX certificate file from the root directory of the DMP system to a secure location on the remote system so that it can be used for authentication purposes. Rename the file so that you can identify which system it authorizes (esSystem-admin.pfx for example). S–2327–B 53 Data Management Platform (DMP) Administrator's Guide Important: Refer to Managing Bright admin.pfx Certificates on page 62 for important information about how to manage the Bright admin.pfx file holding the administrator certificate. remote% scp root@esms1:/root/admin.pfx /securelocation/esSystem-admin.pfx 3. Install the software. a. On Windows, execute the installer .exe file and follow the instructions. b. On Linux, extract the files using the tar command: Remote% tar -xvjf cmgui-6.0-xxxxx-src.tar.bz2 4. Start the cmgui and select the power plug icon and enter the PFX certificate password to connect to the DMP system. See Figure 14. 5. Run the executable and connect to the DMP system using the procedure in Running cmgui and Connecting to a DMP System on page 54. 2.6 Running cmgui and Connecting to a DMP System Before making the initial connection from a desktop computer running cmgui, a PFX file containing both the certificate and private key must be copied from the CIMS (/root/.cm/cmgui/admin.pfx) on the DMP system and stored in a secure location on the remote system. Important: Refer to Managing Bright admin.pfx Certificates on page 62 for important information about how to manage the Bright admin.pfx file holding the administrator certificate. If you need to manage more than one DMP system, rename the admin.pfx file on your local system appropriately. You only need to select and validate the admin.pdx file once. Procedure 4. Starting the cmgui and connecting to the system 1. Copy the admin.pfx file to the remote computer. Rename the file to something specific, such as cray_es_admin.pfx. remote# mkdir ~/cmgui-keys remote# chmod 700 ~/cmgui-keys remote# scp root@esms1:/root/.cm/cmgui/admin.pfx ~/cmgui-keys/cray_es_admin.pfx 2. Run the cmgui executable. 3. To connect cmgui to a DMP system, select the + button. When you run cmgui and add a new system, you must enter the host name of the CIMS, the location of the certificate file, and system password you configured during ESM installation and click OK. 54 S–2327–B Bright Cluster Manager® [2] Figure 13. Connecting cmgui to a DMP System 4. The GUI window power plug icon enables you to connect to the system. Figure 14. Connecting cmgui to a DMP System S–2327–B 55 Data Management Platform (DMP) Administrator's Guide 2.7 Running cmgui From the CIMS The cluster management GUI (cmgui) is used to manage the system after you have installed and licensed the Bright software. The cmgui program may be run on the CIMS node and displayed to a remote X Window System™ running on a Linux™ desktop or other platform. The cmgui program may also be installed on a Linux or Windows® platform and supports a virtual network computing (VNC®) server for remote connections. Note: Communication between the remote computer and the CIMS node should be encrypted. For Procedure 5 on page 56, we recommend using SSH port forwarding or SSH tunneling. When running the cmgui program from the remote computer, cmgui connects to the CIMS node using SSL. Cray recommends using SSH port forwarding when using VNC. Procedure 5. Install and run cmgui from the CIMS 1. On a remote system such as a Linux desktop or PC, start an X-server application such as Xming or Cygwin/X. 2. Enter the following command to log in to the CIMS (in this example, esms1) with SSH X-forwarding. remote% ssh -X root@esms1 esms1 # 3. Start the cmgui program. esms1# /cm/shared/apps/cmgui/cmgui & Figure 15. cmgui Splash Screen 56 S–2327–B Bright Cluster Manager® [2] 4. Select Add a new cluster. 5. Select the power plug icon and enter the system password to connect to the system. Figure 16. cmgui Window The cmgui window displays the DMP system configuration. Figure 17. cmgui Connect-to-Cluster S–2327–B 57 Data Management Platform (DMP) Administrator's Guide 58 S–2327–B CIMS Configuration Tasks [3] 3.1 Software Installation Refer to the Installing Cray Integrated Management Services (CIMS) Software (S–2522) for software installation procedures. 3.1.1 Updating Slave Node RPMs from ESM Media If you update ESM software, without updating CDL (ESL) or CLFS (ESF) software, you must run the ESMupdateimage command for all slave node production images to update the RPMs delivered via the ESM media which need to be installed on the CDL and CLFS nodes. Note: Always reboot the slave nodes that are running the updated image, or push the updated image to the running slave node using cmgui Update Node command or cmsh imageupdate command from device mode. Note: For CLFS nodes, the lustre_control command must be the same on the CIMS node and all CLFS nodes to control the Lustre® file system for actions such as start, stop, and status. This RPM is provided on the ESM media. Procedure 6. Updating slave node RPMs from ESM media 1. Log in to the CIMS as root. 2. Run ESMupdateimage for each software image. The ESMupdateimage command verifies that the software image exists in /cm/images and in the Bright database (via cmsh). If it is not a valid software image, the command aborts. esms1# ESMupdateimage -s softwareimage Note: If the software image is valid, the command determines whether the software image has ESF or ESL software installed (/etc/opt/cray/release/ESFrelease or /etc/opt/cray/release/ESLrelease) and then installs the proper RPMs for CLFS or CDL nodes. If neither of these files is present, then the generic set of slave node RPMs is installed. S–2327–B 59 Data Management Platform (DMP) Administrator's Guide 3. Update the slave node from the software image. If possible, reboot the slave node. When the node reboots it will get all of the changes to the updated software image. esms1# cmsh [esms1]% device [esms1->device]% reboot -n slavenode If not possible to reboot the node, then push the changes in the updated software image to the node. esms1# cmsh [esms1]% device [esms1->device]% imageupdate -w -n slavenode Use synclog to display the provisioning sync log for the node. [esms1->device]% synclog -p slavenode 3.2 Administrative Passwords There are several administrative passwords for a Cray Data Management Platform (DMP) system. Each password is described below: • CIMS - The root password for the primary CIMS and secondary CIMS (the same). • The root password of the software images: This allows a root log in to a slave node, and is stored in the software image. • The root password of the node-installer: This allows a root log in to the node when the node-installer, a stripped-down operating system, is running. The node-installer stage prepares the node for the final operating system when the node is booting up. • The root password of MySQL®: This allows a root log in to the MySQL server. • The administrator certificate password: This decrypts the /root/admin.pfx file on the CIMS so that the administrator certificate can be submitted to CMDaemon for administrative tasks. See Managing Bright admin.pfx Certificates on page 62. • The baseboard management controller (BMC or iDRAC) password for the CIMS (changed using the cmsh shell, and not the cm-change-passwd script). See Changing the Password for the Baseboard Management Controller (BMC or iDRAC) on page 64. • Switch administrative passwords should be managed using the device's console configuration commands. 60 S–2327–B CIMS Configuration Tasks [3] Procedure 7. Changing DMP system passwords 1. Log in to the CIMS as root. 2. Enter cm-change-passwd and follow the prompts to change each password on the system. esms1# cm-change-passwd With this utility you can easily change the following passwords: * root password of head node * root password of slave images * root password of node-installer * root password of mysql * administrator certificate for use with cmgui (/root/admin.pfx) Note: if this cluster has a high-availability setup with 2 head nodes, be sure to run this script on both head nodes. Change password for root on head node? [y/N]: y Changing password for root on head node. Changing password for user root. New UNIX password: newrootpassword Retype new UNIX password: newrootpassword passwd: all authentication tokens updated successfully. Change password for root in default-image [y/N]: y Changing password for root in default-image. Changing password for user root. New UNIX password: newdefaultimagepassword Retype new UNIX password: newdefaultimagepassword passwd: all authentication tokens updated successfully. Change password for root in node-installer? [y/N]: y Changing password for root in node-installer. Changing password for user root. New UNIX password: newnode-installerpassword Retype new UNIX password: newnode-installerpassword passwd: all authentication tokens updated successfully. Change password for MYSQL root user? [y/N]: y Changing password for MYSQL root user. Old password: oldMYSQLpassword New password: newMYSQLpassword Re-enter new password: newMYSQLpassword Change password for admin certificate file? [y/N]: y Enter old password: oldcertificatepassword Enter new password: newcertificatepassword Verify new password: newcertificatepassword Password updated Important: See Managing Bright admin.pfx Certificates on page 62 for more information about changing passwords for the admin.pfx certificate. S–2327–B 61 Data Management Platform (DMP) Administrator's Guide 3. Use cmsh to change the CIMS BMC (iDRAC port) password. esms1# cmsh [esms1]% partition use base [esms1->partition[base]]% show Parameter Value ------------------------------ ------------------------------- Administrator e-mail BMC Password ********* BMC User ID 2 BMC User name root Burn configs Cluster name Training Default burn configuration Default category default Default software image default-image External network site-admin-net Externally visible IP Failover not defined Management network esmaint-net Masternode esms1 Name base Name servers aaa.bbb.ccc.ddd aaa.bbb.ccc.ddd Node basename node Node digits 3 Notes Revision Search domains your.domain.com Time servers timeserver1.com timerserver2.com Time zone America/Chicago [esms1->partition[base]]% set bmcpassword newbmcpassword [esms1->partition*[base*]]% commit 3.2.1 Managing Bright admin.pfx Certificates Important: When an administrator changes the password on the system certificate, the certificate is re-encrypted with the new password. If an administrator has an old copy of a valid certificate that is encrypted with the old password, this administrator can continue to access the CMDaemon unless you revoke the old certificate. Old certificates must be revoked by using the cmsh cert mode revokecertificate command (refer to Procedure 8 on page 64), or by using the Authentication resource from the cmgui resource tree (see Figure 18). The Bright Cluster Manager® (Bright) infrastructure (CMDaemon or cmd) requires public key authentication using X.509v3. X.509 is an ITU-T standard for a public key infrastructure (PKI) for single sign-on (SSO) and Privilege Management Infrastructure (PMI). The X.509 standard specifies, amongst other things, standard formats for public key certificates, certificate revocation lists, attribute certificates, and a certification path validation algorithm. This means in practice, a person authenticating to the cluster management infrastructure must present his/her certificate (i.e. the public key) and in addition must have access to the private key that corresponds to the certificate. A certificate includes a profile that determines which cluster management operations the holder of the certificate may perform. 62 S–2327–B CIMS Configuration Tasks [3] The administrator password provided during Bright installation encrypts the admin.pfx file generated as part of the installation. The same password is also used as the initial root password for all nodes. The administrator certificate is required to enable the CMDaemon and cmsh shell. Typically, administrators copy the admin.pfx file to their local laptop or workstation and use the Bright GUI (cmgui) to manage the system. When the /root/admin.pfx file is updated with a new licence or password, the previous copy of the admin.pfx file continues to enable administrators to access the CMDaemon. The password defined for the administrator certificate is used to decrypt the admin.pfx file, so that the administrator certificate can be presented to CMDaemon. When the password for the admin.pfx file changes, the administrator must distribute the admin.pfx file and password to other administrators, and revoke older certificates to prevent administrators access with the old system certificate. Figure 18. cmgui Authentication Menu S–2327–B 63 Data Management Platform (DMP) Administrator's Guide Procedure 8. Revoking administration certificates 1. Log in to the CIMS as root and start cmsh. esms1# cmsh 2. Switch to cert mode. [esms1]% cert [esms1->cert]% listcertificates 3. List the certificates. The Name column shows the MAC address of the node's esmaint-net network adapter. [esms1->cert]% listcertificates Serial num Days left Profile Country Name Revoked ------------ ---------- ---------------- ---------------- ------------------ -------- 1 36481 node US 84-2b-2b-61-b0-04 No 2 36495 node US 78-2b-cb-38-12-fe No 4. To revoke a certificate, specify the Serial number. [esms1->cert]% revokecertificate 1 Certificate revoked. [esms1->cert]% listcertificates Serial num Days left Profile Country Name Revoked ------------ ---------- ---------------- ---------------- ------------------ -------- 1 36481 node US 84-2b-2b-61-b0-04 Yes 2 36495 node US 78-2b-cb-38-12-fe No 3.2.2 Changing the Password for the Baseboard Management Controller (BMC or iDRAC) Procedure 9. Changing the password on the BMC (iDRAC) 1. Log in to the CIMS as root and start cmsh. esms1# cmsh 2. Switch to partition mode. Use the base partition to change the BMC password. [esms1]% partition use base [esms1->partition[base]]% 3. Get the BMC user name. [esms1->partition[base]]% get bmcusername root 4. Get the BMC password (set during installation). [esms1->partition[base]]% get bmcpassword bmcpassword 64 S–2327–B CIMS Configuration Tasks [3] 5. Change and commit the BMC password. [esms1->partition[base]]% set bmcpassword enter new password: NewPassWord retype new password: NewPassWord [esms1->partition[base*]]% commit [esms1->partition[base]]% 3.3 Changing CIMS Configuration Settings You can modify the CIMS configuration settings such as baseboard management controller (BMC) password, name servers, search domains, and time servers using cmsh partition mode commands. The following example lists the CIMS configuration settings for the base partition. Use the set command from partition mode to set specific properties for the CIMS. Example 8. CIMS configuration settings esms1# cmsh [esms1]% partition use base [esms1->partition[base]]% show Parameter Value ------------------------------ ----------------------------- Administrator e-mail BMC Password ********* BMC User ID 2 BMC User name root Burn configs Cluster name Cray Training Default burn configuration Default category default Default software image default-image External network site-admin-net Externally visible IP Failover not defined Management network esmaint-net Masternode esms1 Name base Name servers aaa.bbb.ccc.ddd aaa.bbb.ccc.ddd Node basename node Node digits 3 Notes Revision Search domains your.domain.com Time servers timeserver1.com timerserver2.com Time zone America/Chicago S–2327–B 65 Data Management Platform (DMP) Administrator's Guide 3.4 Configuring the RAID Virtual Disks A CIMS node has six physical disks. You must reconfigure the CIMS node disks into two RAID-5 virtual disks, /dev/sda and /dev/sdb. The Bright software creates the required disk partitions during installation. Note: If you configure partitions for a single CIMS, then later add a second CIMS, you must resize the /cm partition for the HA configuration. Procedure 10. Setting up RAID virtual disks Note: This procedure includes detailed steps for the DELL™ R720 server using the PERC H710P Mini BIOS Configuration Utility 4.00-0014. Depending on your server model and version of RAID configuration utility, there could be minor differences in the steps to configure your system. For more information, refer to the documentation for your DELL™ PERC controller or server RAID controller software. 1. Connect a keyboard, monitor, and mouse to the front panel USB and monitor connectors on the CIMS. 2. Power up the CIMS. As the CIMS node reboots, enter the RAID controller configuration utility by pressing Ctrl-R when prompted. Note: Cray recommends using the RAID configuration utility (via Ctrl-R) to configure the RAID virtual disks instead of the System Setup Device Settings menu. In this utility, use the up-arrow or down-arrow key to select (highlight) an item in a list. Press Enter to select items. To display a menu of options for an item, press the F2 key. Use the right-arrow, left-arrow, or Tab key to change between the Yes and No buttons in a confirmation window. 3. Clear the default disk configuration, if necessary. a. If any disk groups are currently defined, select Disk Group 0, then press the F2 key. b. Select Delete Disk Group, then press Enter. c. In the pop-up confirmation window, select Yes to confirm your changes. 4. Create a new virtual disk for /dev/sda. In this step, you will configure /dev/sda as a RAID-5 virtual disk with a capacity of 256 GB. a. Select No Configuration Present, then press the F2 key. b. Select Create New VD, then press Enter. The Create New VD screen opens. c. Change the RAID level to RAID-5. 1) Select RAID Level, then press Enter to display the available options. 66 S–2327–B CIMS Configuration Tasks [3] 2) Select RAID-5. 3) Press Enter to return to the main screen. d. Select all physical disks for this RAID-5 disk group. 1) Press Tab to move to the Physical Disks area. 2) Press Enter to check the box for a physical disk. This action also advances the selection to the next disk. 3) Repeat the previous step for each physical disk. e. Press Tab to move to VD Size, then enter 256. Note: Be sure to specific GB and not MB. The PERC controller software automatically adjusts this value to 255.9. f. Press Tab to move to VD Name, then enter sda. g. Enable disk initialization. 1) Press Tab to move to the Advanced Settings area. 2) Press Enter to check the Advanced Settings box so that you can make changes. 3) Select Initialize, then press Enter to check the box. h. To confirm your changes, press the Tab key to select OK, then press Enter. i. A message appears to let you know that initialization will destroy data on the virtual disk. Select OK to continue, then press Enter. j. An "Initialization complete" message appears. Select OK to continue, then press Enter. 5. Add a new virtual disk for /dev/sdb. In this step, you will configure /dev/sdb as a RAID-5 virtual disk with the remainder of the available space. a. Select Disk Group: 0, RAID-5, then press the F2 key. b. Select Add New VD, then press Enter. The Add VD in Disk Group 0 screen opens. c. Keep VD Size as presented (the remainder of the disks). d. Press Tab to move to VD Name, then enter sdb. e. Enable disk initialization. 1) Press Tab to move to the Advanced Settings area. 2) Press Enter to check the Advanced Settings box so that you can make changes. S–2327–B 67 Data Management Platform (DMP) Administrator's Guide 3) Select Initialize, then press Enter to check the box. f. To confirm your changes, press the Tab key to select OK, then press Enter. g. A message appears to let you know that initialization will destroy data on the virtual disk. Select OK to continue. h. An "Initialization complete" message appears. Select OK to continue, then press Enter. 6. Verify the virtual disk changes. Compare your settings with those shown in Figure 19. Figure 19. Final RAID Configuration Settings 7. To exit the RAID configuration utility, press the Escape key. 8. To confirm, press OK, then press Enter. Note: Disk initialization is performed in the background, and takes about 2 hours to complete. You can exit this utility and continue to the next procedure. 9. A message appears prompting you to reboot. Press Ctrl-Alt-Delete. The server will restart the boot process and will not interrupt RAID initialization. Note: During the system reboot, be prepared to type F2, when prompted, to change the system setup. Refer to Configuring the LSI® MegaCLI™ RAID Utility on page 69 for a procedure to configure Bright healthchecks for local RAID devices. 68 S–2327–B CIMS Configuration Tasks [3] 3.5 Configuring the LSI® MegaCLI™ RAID Utility CIMS nodes that use LSI® Inc. MegaRAID™ controllers and the megaraid_sas kernel module use PERC710P, PERC 6/i RAID or other hardware modules. To manage, monitor, and configure the local RAID systems, install the LSI MegaCLI RAID utility on the CIMS. The MegaCLI utility also enables you to configure monitoring metrics, healthchecks, and administrator alerts in Bright that monitor the CIMS local RAID systems. Use Procedure 11 on page 69 to install and run the MegaCLI utility. The utility installs in /opt/MegaRAID/MegaCli. Verify the utility is installed and running correctly, then use Procedure 13 on page 72 to configure the Bright healthcheck feature to monitor CIMS node RAID devices. Use Procedure 12 on page 70 to configure the MegaCLI utility for a slave node. Procedure 11. Installing the MegaCLI utility on the CIMS Note: This procedure configures the MegaCLI utility on the CIMS node. 1. Open a web browser and access the license agreement at http://www.lsi.com/Pages/user/eula.aspx?file=http://www.lsi.com/. Click Accept to accept the software license agreement. 2. Navigate to the storage downloads area of the LSI website and search for "MegaCLI". 3. Download the latest MegaCLI archive for Linux® (for example, MegaCli_Linux.zip) from the Downloads area of the www.lsi.com website. a. Log in to CIMS as root, copy the downloaded MegaCLI archive to the CIMS, and decompress the archive. esms1# mkdir /root/MegaCLI esms1# cd /root/MegaCLI esms1# scp user@remotesystem:/user/MegaCli_Linux.zip /root/MegaCLI esms1# unzip MegaCli_Linux.zip Archive: ./MegaCli_Linux.zip creating: MegaCli_Linux/ inflating: MegaCli_Linux/MegaCli-8.07.08-1.i386.rpm inflating: MegaCli_Linux/megacli_8.07.08-1_all.deb b. Install the Linux MegaCLI RPM. esms1# pwd /root/MegaCLI esms1# cd MegaCli_Linux esms1# rpm -iMegaCli-8.07.08-1.i386.rpm S–2327–B 69 Data Management Platform (DMP) Administrator's Guide 4. Run the utility to verify it is functioning. esms1# cd /opt/MegaRAID/MegaCli esms1# ./MegaCli64 -AdpAllInfo -aAll Adapter #0 ======================================================== Versions ================ Product Name : PERC 6/i Integrated Serial No : 1122334455667788 FW Package Build: 6.0.2-0002 Mfg. Data ================ Mfg. Date : 06/08/07 Rework Date : 06/08/07 Revision No : Battery FRU : N/A Image Versions in Flash: ================ FW Version : 1.11.52-0396 BIOS Version : NT13-2 WebBIOS Version : 1.1-32-e_11-Rel Ctrl-R Version : 1.01-010B Boot Block Version : 1.00.00.01-0008 ... Procedure 12. Installing the MegaCLI utility on slave node 1. Open a web browser and access the license agreement at http://www.lsi.com/Pages/user/eula.aspx?file=http://www.lsi.com/. Click Accept to accept the software license agreement. 2. Navigate to the storage downloads area of the LSI website and search for "MegaCLI". 3. Download the latest MegaCLI archive for Linux® (for example, MegaCli_Linux.zip) from the Downloads area of the www.lsi.com website. 4. Log in to CIMS as root, copy the downloaded MegaCLI archive to the CIMS, and decompress the archive. esms1# mkdir /root/MegaCLI esms1# cd /root/MegaCLI esms1# scp user@remotesystem:/user/MegaCli_Linux.zip /root/MegaCLI esms1# unzip MegaCli_Linux.zip Archive: ./MegaCli_Linux.zip creating: MegaCli_Linux/ inflating: MegaCli_Linux/MegaCli-8.07.08-1.i386.rpm inflating: MegaCli_Linux/megacli_8.07.08-1_all.deb 70 S–2327–B CIMS Configuration Tasks [3] 5. Clone the current working slave node software image. Choose a unique name to identify the MegaCLI utility image. Note: Copy the image from the UNIX™ prompt and wait for the copy to complete before cloning the image in Bright. This ensures the clone operation is complete before you continue. esms1# cp -pr /cm/images/SlaveImage /cm/images/megacli-image 6. Start cmsh, clone the slave node software image to megacli-image, setup a node category for the MegaCLI utility, and assign the megacli-image to the new category (megacli-category). esms1# cmsh [esms1]% softwareimage [esms1->softwareimage]% clone SlaveImage megacli-image [esms1->softwareimage*]% commit [esms1->softwareimage*]% category [esms1->category]% clone SlaveCategory megacli-category [esms1->category*[megacli-category*]% commit [esms1->category[megacli-category]% set softwareimage megacli-image [esms1->category*[megacli-category*]% commit [esms1->category[megacli-category]% quit esms1# 7. Bind mount /root/MegaCLI_Linux to /cm/images/megacli-image/tmp/MegaCLI or copy the RPM to a directory in the megacli-image software image. This example shows the bind mount method. esms1# mkdir -p /cm/images/megacli-image/tmp/MegaCLI esms1# mount --bind /root/MegaCLI_Linux /cm/images/megacli-image/tmp/MegaCLI 8. Install the MegaCLI software on the slave node software image (megacli-image). esms1# chroot /cm/images/megacli-image/ esms1:> rpm -ivh /tmp/MegaCLI/MegaCli-8.07.08-1.i386.rpm Preparing... ########################################### [100%] 1:MegaCli ########################################### [100%] exitesms1:> exit Important: You must remove the bind mount. 9. Remove the bind mount. esms1# umount /cm/images/megacli-image/tmp/MegaCLI esms1# rm -f /cm/images/megacli-image/tmp/MegaCLI 10. Use cmsh to assign a slave node (in this example, esfs-mds1) to the megacli-category. esms1# cmsh [esms1]% device [esms1->device]% use esfs-mds1 [esms1->device[esfs-mds1]]% set category megacli-category [esms1->device*[esfs-mds1*]]% commit S–2327–B 71 Data Management Platform (DMP) Administrator's Guide 11. Reboot the esfs-mds1 node (or fromcmsh use imageupdate ) to test the megacli-image and exit cmsh. [esms1->device[esfs-mds1]]% reboot esfs-mds1 [esms1->device[esfs-mds1]]% quit Open a remote console to the esfs-mds1 node and verify that the megacli-image software image boots without errors. 12. SSH to esfs-mds1 and run the utility to verify it is functioning on the slave node. esms1# ssh esfs-mds01 Last login: Thu Nov 7 12:56:51 2013 from esms1.cm.cluster [root@esfs-mds1 ~]# cd /opt/MegaRAID/MegaCli [root@esfs-mds1 ~]# ./MegaCli64 -AdpAllInfo -aAll Adapter #0 ======================================================== Versions ================ Product Name : PERC 6/i Integrated Serial No : 1122334455667788 FW Package Build: 6.0.2-0002 Mfg. Data ================ Mfg. Date : 06/08/07 Rework Date : 06/08/07 Revision No : Battery FRU : N/A Image Versions in Flash: ================ FW Version : 1.11.52-0396 BIOS Version : NT13-2 WebBIOS Version : 1.1-32-e_11-Rel Ctrl-R Version : 1.01-010B Boot Block Version : 1.00.00.01-0008 ... 13. Configure the megaraid healthcheck in Bright using Procedure 13 on page 72. 14. Assign other slave nodes to the megacli-category. Procedure 13. Configuring the megaraid healthcheck in Bright 1. Log in to the CIMS as root and start cmsh. esms1: # cmsh [esms1]% 72 S–2327–B CIMS Configuration Tasks [3] 2. Switch to monitoring healthchecks mode. [esms1]% monitoring healthchecks [esms1->monitoring->healthchecks]% list Name (key) Command ---------------------------- ------------------------------------------------ DeviceIsUp ManagedServicesOk chrootprocess /cm/local/apps/cmd/scripts/healthchecks/chrootp+ cmsh /cm/local/apps/cmd/scripts/healthchecks/cmsh diskspace /cm/local/apps/cmd/scripts/healthchecks/diskspa+ exports /cm/local/apps/cmd/scripts/healthchecks/exports failedprejob /cm/local/apps/cmd/scripts/healthchecks/failedp+ failover /cm/local/apps/cmd/scripts/healthchecks/failover hardware-profile /cm/local/apps/cmd/scripts/healthchecks/node-ha+ hpraid /cm/local/apps/cmd/scripts/healthchecks/hpraid interfaces /cm/local/apps/cmd/scripts/healthchecks/interfa+ ipmihealth /cm/local/apps/cmd/scripts/metrics/sample_ipmi ldap /cm/local/apps/cmd/scripts/healthchecks/ldap lustre /cm/local/apps/cmd/scripts/healthchecks/lustre mounts /cm/local/apps/cmd/scripts/healthchecks/mounts mysql /cm/local/apps/cmd/scripts/healthchecks/mysql ntp /cm/local/apps/cmd/scripts/healthchecks/ntp oomkiller /cm/local/apps/cmd/scripts/healthchecks/oomkill+ portchecker /cm/local/apps/cmd/scripts/healthchecks/portche+ rogueprocess /cm/local/apps/cmd/scripts/healthchecks/roguepr+ schedulers /cm/local/apps/cmd/scripts/healthchecks/schedul+ smart /cm/local/apps/cmd/scripts/healthchecks/smart ssh2node /cm/local/apps/cmd/scripts/healthchecks/ssh2node swraid /cm/local/apps/cmd/scripts/healthchecks/swraid testhealthcheck /cm/local/apps/cmd/scripts/healthchecks/testhea+ 3. Add the megaraid healthcheck. [esms1->monitoring->healthchecks]% add megaraid [esms1->monitoring->healthchecks*[megaraid*]]% show Parameter Value ------------------------------ ------------------------------------------------ Class of healthcheck misc Command Description Disabled no Extended environment no Name megaraid Notes Only when idle no Parameter permissions optional Revision Sampling method samplingonnode State flapping count 7 Timeout 5 Valid for node,headnode 4. Configure the megaraid healthcheck and commit the settings. [esms1->monitoring->healthchecks*[megaraid*]]% set classofhealthcheck disk [esms1->monitoring->healthchecks*[megaraid*]]% set command /cm/local/apps/cmd/scripts/healthchecks/megaraid [esms1->monitoring->healthchecks*[megaraid*]]% commit [esms1->monitoring->healthchecks[megaraid]]% S–2327–B 73 Data Management Platform (DMP) Administrator's Guide 5. Configure the megaraid healthcheck for the CIMS. Note: To configure the healthcheck on a slave node, specify its category (in this example, megacli-category) using: monitoring setup healthconf megacli-category [esms1->monitoring->healthchecks[megaraid]]% monitoring setup healthconf headnode [esms1->monitoring->setup[HeadNode]->healthconf]% add megaraid [esms1->monitoring->setup*[HeadNode*]->healthconf*[megaraid*]]% show Parameter Value ------------------------------ ------------------------------------------------ Check Interval 120 Disabled no Fail Actions Fail severity 10 GapThreshold 2 HealthCheck megaraid HealthCheckParam LogLength 3000 Only when idle no Pass Actions Revision Stateflapping Actions Store yes ThresholdDuration 1 Unknown Actions Unknown severity 10 6. Commit the changes. [esms1->monitoring->setup*[HeadNode*]->healthconf*[megaraid*]]% commit [esms1->monitoring->setup[HeadNode]->healthconf[megaraid]]% 74 S–2327–B CIMS Configuration Tasks [3] 7. Go into device mode and show the health data for the CIMS or slave node. Verify the megaraid health data shows PASS. Note: To show the health data for a slave node (esfs-mds1), type the command device use esfs-mds1. [esms1->monitoring->setup[HeadNode]->healthconf[megaraid]]% device use esms1 [esms1->device[esms1]]% latesthealthdata Health Check Severity Value Age (sec.) Info Message ---------------------------- -------- ---------------- ---------- ------------------------ DeviceIsUp 0 PASS 1 ManagedServicesOk 0 PASS 19 mounts 0 PASS 19 exports 0 PASS 19 smart 0 PASS 19 sda: Smart command failed ldap 0 PASS 19 failover 0 PASS 19 interfaces 40 FAIL 19 eth4 not up oomkiller 0 PASS 19 cmsh 0 PASS 19 mysql 0 PASS 19 failedprejob 0 PASS 19 diskspace:2% 10% 20% 0 PASS 19 ntp 0 PASS 19 schedulers 0 PASS 19 chrootprocess 0 PASS 19 megaraid 0 PASS 19 Note: Use the following command to run the megaraid healthcheck from a command line. [esms1->device[esms1]]% quit esms1# /cm/local/apps/cmd/scripts/healthchecks/megaraid PASS esms1# /cm/local/apps/cmd/scripts/healthchecks/megaraid -d 3>&1 megacli command path: /opt/MegaRAID/MegaCli/MegaCli64 cmd: /opt/MegaRAID/MegaCli/MegaCli64 -LdPdInfo -aALL -NoLog line: Adapter #0 adapter: 0 line: Virtual Drive: 0 (Target Id: 0) vdrive: 0 line: State : Optimal vstate: Optimal line: Span: 0 - Number of PDs: 3 span: 0 line: PD: 0 Information pdisk: 0 line: Enclosure Device ID: 32 enc: 32 line: Slot Number: 0 encslot: 0 line: Firmware state: Online, Spun Up pstate: Online, Spun Up line: Drive has flagged a S.M.A.R.T alert : No psmart: No line: PD: 1 Information ... S–2327–B 75 Data Management Platform (DMP) Administrator's Guide 3.6 Adding a New or Modified Disk Setup XML File to the Bright Database Important: If the default disk setup XML files are updated in a ESM release and the site disk setup XML files have been customized, system administrators must compare the newly released disk setup XML files with the current production disk setup XML files, and merge the changes manually. After the changes have been merged, you must load the updated disk setup file into the Bright database for the node category and reboot the node. An ESM software update may modify the default disk setup XML files for the default esFS-MDS and esFS-OSS categories or default CDL categories. New disk setup files may also be added to the /opt/cray/esms/cray-es-diskpartitions-XX/default/ directory, which then must be added to the Bright database manually. Use this procedure to add a new disk setup XML file to an existing node category (in this example, the esFS-MDS category). Procedure 14. Changing the disk setup XML file for a category 1. Log in to the CIMS as root and start cmsh. esms1# cmsh [esms1]% 2. Switch to category mode and select the esFS-MDS category. [esms1]% category [esms1->category]% use esFS-MDS 3. Get the current disk setup for the esFS-MDS category. [esms1->category[esFS-MDS]]% get disksetup 4. Set your disk setup to either the full disk setup using 1TB capacity disks (esfs-diskfull.xml), or if the system contains smaller capacity disks, choose (esfs-small-diskfull.xml), depending on the hardware configuration. [esms1->category[esFS-MDS]]% set disksetup /opt/cray/esms/cray-es-diskpartitions-XX/default/esfs-diskfull.xml or [esms1->category[esFS-MDS]]% set disksetup /opt/cray/esms/cray-es-diskpartitions-XX/default/esfs-small-diskfull.xml 5. Commit the change. [esms1->category[esFS-MDS*]]% commit 76 S–2327–B CIMS Configuration Tasks [3] 6. Reboot the nodes in the esFS-MDS category. [esms1->category[esFS-MDS]]% device [esms1->device]% reboot -c esFS-MDS esfs-mds1: Reboot in progress ... esfs-mds2: Reboot in progress ... 7. Repeat this procedure, for the esFS-OSS category, or other default node categories that use the new disk setup files. 3.7 Changing BIOS for a DELL™ R720 CIMS Node This procedure describes how to change the system setup for the CIMS: the network connections, remote power control, and the remote console. Procedure 15. Changing the system setup for the CIMS Note: This procedure includes detailed steps for the Dell R720 server. Depending on your server model and version of BIOS configuration utility, there could be minor differences in the steps to configure your system. For more information, refer to the documentation for your Dell server. Note: When configuring a secondary CIMS, do not disable the embedded NIC in the BIOS settings on the secondary CIMS. The secondary CIMS needs to initially PXE boot from the primary CIMS to perform the cloning operation. Also, the secondary CIMS should have Boot Sequence set to Integrated Nic Hard drive C:, and then the DVD/Optical drive. 1. Watch as the system reboots. When the BIOS power-on self-test (POST) process begins, quickly press the F2 key after the following messages appear in the upper-right of the screen. F2 = System Setup F10 = System Services F11 = BIOS Boot Manager F12 = PXE Boot When the F2 keypress is recognized, the F2 = System Setup line changes to Entering System Setup. After the POST process completes and all disk and network controllers have been initialized, the Dell System Setup screen appears. The following submenus are available on the System Setup Main Menu: S–2327–B 77 Data Management Platform (DMP) Administrator's Guide Figure 20. Dell 720 BIOS Menu Note: In this utility, use the Tab key to move to different areas on the screen. To select an item, use the up-arrow and down-arrow keys to highlight the item, then press the Enter key. Press the Escape key to exit a submenu and return to the previous screen. 2. Change the System BIOS settings. Figure 21. Dell 720 BIOS Boot Settings a. Select System BIOS, then press Enter. b. Select Boot Settings, then press Enter. c. Select BIOS Boot Settings, then press Enter. 78 S–2327–B CIMS Configuration Tasks [3] d. Select Boot Sequence, then press Enter to view the boot settings. Figure 22. Dell 720 BIOS Boot Sequence e. If creating a stand-alone CIMS, change the boot order in the pop-up window so that the optical drive appears first, then hard drive, then integrated NIC last. Note: If creating an HA CIMS, do not disable the embedded NIC in the BIOS settings for the secondary CIMS. The secondary CIMS must initially PXE boot from the primary CIMS to perform the cloning operation. Set the secondary CIMS boot sequence to boot from the integrated NIC first, then hard drive, then DVD/Optical drive last. Tip: Use the up-arrow or down-arrow key to highlight an item, then use the + and - keys to move the item up or down. f. Enable Hard drive C: under the Boot Option/Enable/Disable section. S–2327–B 79 Data Management Platform (DMP) Administrator's Guide Figure 23. Dell 720 BIOS Boot Settings g. Press Enter to return to the BIOS Boot Settings screen. h. Press Escape to exit BIOS Boot Settings. i. Press Escape to exit Boot Settings and return to the System BIOS Settings screen. 3. Change the serial communication settings. a. On the System BIOS Settings screen, select Serial Communication. Figure 24. Dell 720 BIOS Serial Communication Settings b. On the Serial Communication screen, select Serial Communication and press Enter. A pop-up window displays the available options. 80 S–2327–B CIMS Configuration Tasks [3] c. Select On with Console Redirection via COM2, then press Enter. d. Select Serial Port Address, then select Serial Device1=COM1, Serial Device2=COM2, and press Enter. Figure 25. Dell 720 BIOS Serial Port Address Settings e. Select External Serial Connector, then pressEnter. A pop-up window displays the available options. f. In the pop-up window, select Remote Access Device, then press Enter to return to the previous screen. g. Select Failsafe Baud Rate, then press Enter. A pop-up window displays the available options. h. In the pop-up window, select 115200, then press Enter to return to the previous screen. i. Press the Escape key to exit the Serial Communication screen. j. Press the Escape key to exit the System BIOS Settings screen. k. A "Settings have changed" message appears. Select Yes to save your changes. l. A "Settings saved successfully" message appears. Select Ok. 4. On the System Setup Main Menu, select iDRAC Settings, then press Enter. S–2327–B 81 Data Management Platform (DMP) Administrator's Guide Figure 26. Dell 720 BIOS iDRAC Settings 5. Select Network, then press Enter. A long list of network settings is displayed. 6. Use the down-arrow key to scroll to DNS DRAC Name and press Enter. 7. Enter an iDRAC host name that is similar to the CIMS node host name. For example, esms1-idrac. Figure 27. Dell 720 BIOS iDRAC Name 8. Change the IPv4 settings. a. Use the down-arrow key to scroll to the IPV4 SETTINGS list. b. Ensure that IPv4 is enabled. 1) If necessary, select Enable IPV4 and press Enter. 82 S–2327–B CIMS Configuration Tasks [3] 2) In the pop-up window, select . 3) Press Enter to return to the previous screen. c. Ensure that DHCP is disabled. 1) If necessary, select Enable DHCP and press Enter. 2) In the pop-up window, select . 3) Press Enter to return to the previous screen. d. Change the IP address. 1) Select IP Address. A pop-up window opens for entering the new data. 2) In the pop-up window, enter the IP address of the iDRAC interface (ipmi0) for site-admin-net on the CIMS. 3) Press Enter to return to the previous screen. e. Change the gateway. 1) Select Gateway. A pop-up window opens for entering the new data. 2) In the pop-up window, enter the appropriate value for the gateway of the site-admin-net network. 3) Press Enter to return to the previous screen. f. Change the subnet mask. 1) Select Subnet Mask. A pop-up window opens for entering the new data. 2) In the pop-up window, enter the subnet mask for site-admin-net (such as 255.255.255.0). 3) Press Enter to return to the previous screen. g. Change the DNS server settings. 1) Select Preferred DNS Server. A pop-up window opens for entering the new data. 2) In the pop-up window, enter the IP address of the primary DNS server. 3) Press Enter to return to the previous screen. 4) Select Alternate DNS Server. A pop-up window opens for entering the new data. 5) In the pop-up window, enter the IP address of the alternate DNS server. 6) Press Enter to return to the previous screen. 9. Change the IPMI settings to enable the Serial Over LAN (SOL) console. S–2327–B 83 Data Management Platform (DMP) Administrator's Guide a. Use the down-arrow key to scroll to the IPMI SETTINGS list. b. Ensure that IPMI over LAN is enabled. 1) If necessary, select Enable IPMI over LAN, then press Enter. 2) In the pop-up window, select . Figure 28. Dell 720 BIOS Enable IPMI over LAN (SOL) 3) Press Enter to return to the previous screen. c. Verify that Channel Privilege Level Limit is set to Administrator. 1) If necessary, select Channel Privilege Level Limit to display the pop-up window. 2) In the pop-up window, select Administrator. 3) Press Enter to return to the previous screen. d. Press the Escape key to exit the Network screen and return to the iDRAC Settings screen. 10. On the iDRAC Settings screen, change the user configuration settings. a. Use the down-arrow key to highlight User Configuration, then press Enter. b. Confirm that User Name is root. 1) If necessary, select User Name. A pop-up window opens to let you enter the user name. 2) In the pop-up window, enter root. 84 S–2327–B CIMS Configuration Tasks [3] 3) Press Enter to return to the previous screen. c. Select Change Password. A pop-up window opens to let you create a new password. d. In the pop-up window, enter a new password. e. In the next pop-up window, re-enter the new password to confirm it. f. Press the Escape key to exit the User Configuration screen. 11. Change the LCD configuration to show the host name in the LCD display. a. On the iDRAC Settings screen, use the down-arrow key to scroll down and highlight LCD, then press Enter. b. Select Set LCD message. A pop-up window opens (line 1). c. In the pop-up window, select User-Defined String, then press Enter. d. Select User-Defined String , then press Enter. A text pop-up window opens (line 2) for entering the new string. e. In the text pop-up window, enter the CIMS host name (such as esms1), then press Enter. Figure 29. Dell 720 BIOS iDRAC LCD Settings f. Press the Escape key to exit the LCD screen. g. Press the Escape key to exit the iDRAC Settings screen. h. A "Settings have changed" message appears. Select Yes, then press Enter to save your changes. S–2327–B 85 Data Management Platform (DMP) Administrator's Guide i. A "Settings saved successfully" message appears. Select Ok, then press Enter. The main screen (System Setup Main Menu) appears. 12. Disable the integrated NIC device by changing the setting for the integrated NIC on port 1 from PXE to None. Note: If you are configuring a secondary CIMS in an HA configuration, set the integrated NIC port (on esmaint-net) to PXE so that it PXE boots from the primary CIMS. a. On the System Setup Main Menu, select Device Settings, then press Enter. b. On the Device Settings screen, select Integrated NIC 1 Port 1: ..., then press Enter. c. On the Main Configuration Page screen, select MBA Configuration Menu, then press Enter. Figure 30. Dell 720 BIOS MBA Configuration Settings d. On the MBA Configuration Menu screen, select Legacy Boot Protocol, then press Enter. A pop-up window displays the available options. e. In the pop-up window, use the down-arrow key to highlight None, then press Enter. f. Press the Escape key to exit the MBA Configuration Menu screen. g. Press the Escape key to exit the Main Configuration Page screen. h. Press the Escape key to exit the Device Settings screen. i. A "Settings have changed" message appears. Select Yes, then press Enter to save your changes. 86 S–2327–B CIMS Configuration Tasks [3] j. A "Settings saved successfully" message appears. Select Ok, then press Enter. The main screen (System Setup Main Menu) appears. 13. Insert the Bright software media in the optical (DVD) drive of the CIMS node. Important: The CIMS must boot from the Bright software media to configure and install the Bright software. While the memory-resident operating system is loaded, you must copy the XML configuration file from the Cray ESM software media and save it on the CIMS, then edit and save the file to /root/cm with your site-specific configuration settings. Be aware that changes to the XML configuration file will be lost if the file is not saved to a remote system before the CIMS is rebooted. 14. While viewing the System Setup Main Menu, press the Escape key to exit the Dell System Setup utility. 15. A message appears asking if you want to exit and reboot. Select Yes. The server will restart the boot process. Figure 31. Dell 720 System BIOS Settings 3.8 Power Control The Bright software, IPMI, and the DELL™ Remote Access Controller (iDRAC) enable you to monitor and control power remotely. (If the system includes intelligent PDUs, these too can be controlled and monitored from Bright.) Refer to iDRAC Remote Console on page 120 for more information about the iDRAC. S–2327–B 87 Data Management Platform (DMP) Administrator's Guide The Bright cmgui Overview tab of a device can be used to check its power status information. Right-clicking a node in the resource tree also displays power control commands. The Task tab, enables you to select: • Power on • Power off • Reset - powers off a device and powers it on again after a brief delay When doing a power operation on multiple devices, CMDaemon inserts a 1 second delay between successive devices, to avoid power surges on the infrastructure. The delay period may be altered using cmsh -d | --delay option. The following power control examples can be used from cmsh device mode. Log in to the CIMS as root, start cmsh, and switch to device mode. esms1# cmsh [esms1]% device [esms1-<device]% power -n eslogin01 on Powers up an individual node such as, eslogin01 power -n eslogin01..eslogin04,eslogin06 off Powers off a list of nodes, such as eslogin01 to eslogin04 and eslogin06 power -c eslogin -d 10 reset Power cycles all nodes in the eslogin category, with a 10 second delay between each node power reset power -g es-datamover Power on all nodes in the es-datamover node group 88 S–2327–B CIMS Configuration Tasks [3] power -g esfs-oss status Check power status of all nodes in the esfs-oss node group. ON Power is ON OFF Power is OFF RESET Displays during the short time the power is off during a power reset. The reset is a hard power off for PDUs, but can be a soft or hard reset for other power control devices. FAILED Power status communication failure FAILED Power status communication failure UNKNOWN Power status script timeout pexec power off Powers off all nodes Bright software also supports power saving features through resource managers such as Simple Linux Utility for Resource Management (SLURM) or other workload management software. Refer to the Bright Cluster Manager 6.0 Administrator Manual for more information. 3.9 Rebooting Slave Nodes Procedure 16. Rebooting slave nodes 1. Log in to the CIMS as root and start cmsh. esms1# cmsh [esms1]% 2. Switch to device mode and launch a remote console (rconsole) on the slave node (in this example eslogin1). [esms1]% device use eslogin1 [esms1->device[eslogin1]]% rconsole 3. In a separate CIMS window, login as root and reboot the slave node using cmsh or use the Reboot button from the cmgui. esms1# cmsh [esms1]% device use eslogin1 [esms1->device[eslogin1]]% reboot The following reboot examples can be used from cmsh device mode S–2327–B 89 Data Management Platform (DMP) Administrator's Guide Log in to CIMS as root, start cmsh, and switch to device mode. esms1# cmsh [esms1]% device [esms1->device]% reboot -n eslogin01 Reboot an individual node reboot -n eslogin01..eslogin04,eslogin06 To reboot a list of nodes reboot -c esLogin-XC Reboot all nodes in the eslogin category reboot -g Login Reboot all nodes in the Login node group pexec reboot Reboot all nodes. reboot esfs-mds[1,2],esfs-oss[1,2,3,4] Specifies a range of nodes The parallel shell execution command, pexec, can be run from within the OS shell (bash by default), or from within CMDaemon (cmsh or cmgui). The OS shell pexec commands run on the nodes sequentially by default, and wait for the output from one node before moving on to the next. If the OS shell pexec is run with the background execution option (-b), then the bash commands are executed in parallel. Running in parallel is not done by default, because it could be risky for some commands, such as power-cycling nodes with a reboot, which may put unacceptable surge demands on the power supplies. For example: Within cmsh or cmgui, the execution of a power reset command from device mode to power cycle a properly-configured group of nodes is safe, due to safeguards in CMDaemon to prevent nodes powering up too soon after each other. 3.10 Shutting Down Slave Nodes This procedure shows you how to perform an orderly shutdown and power off a node or nodes. You can shutdown nodes individually, by list, range, rack, and by category or node group. 90 S–2327–B CIMS Configuration Tasks [3] The following reboot examples can be used from cmsh device mode. Log in to the CIMS as root, start cmsh, and switch to device mode. esms1# cmsh [esms1]% device [esms1->device]% shutdown -n eslogin01 Shutdown an individual node shutdown -n eslogin01..eslogin04,eslogin06 Shutdown a list of nodes shutdown -c eslogin Shutdown all nodes in the eslogin category shutdown -g es-datamover Shutdown all nodes in the es-datamover node group 3.11 Network Settings Bright Cluster Manager® (Bright) configures three default network objects. These are internalnet, externalnet, and globalnet. These are renamed for a DMP system, but used throughout the GUI menus as a way to classify the different networks. Internal The internal system network, or management network (these networks are renamed to esmaint-net, ib-net, ipmi-net, wlm-net). External The network connecting the DMP system to the outside world (site-user-net, site-admin-net typically a user or campus network). Global A special network used to set the domain name for nodes so that they can be resolved (not used in a Cray DMP system). Figure 2 shows an example of the networks in a Cray DMP system from the Bright GUI. Select the Networks object in the Resources tree to view all the networks defined in the DMP system. You can sort on each of the columns, S–2327–B 91 Data Management Platform (DMP) Administrator's Guide Figure 32. Bright Network Configuration GUI The network mode of the cmsh command can be used to modify network parameters for each of the networks defined in the system. A DMP system requires the following networks: esmaint-net Internal management network that connects the CIMS server(s) with the slave nodes. This network enables Bright to manage and provision the slave nodes and other devices in the DMP system. ipmi-net Internal IPMI/DRAC (Dell Remote Access Controller) network that provides remote console and power management of the slave nodes from the CIMS. site-admin-net External administration network used by site administrators to log in to the CIMS server (typically on the same network that the SMW is on). The name and IP address of this network are customized during installation. Note: The CIMS IPMI interface (BMC or iDRAC) may also be on this network (instead of ipmi-net) to provide remote console and power management of the CIMS server. 92 S–2327–B CIMS Configuration Tasks [3] site-user-net External user network used by the slave nodes. On CDL nodes, this network provides user access and authentication services such as LDAP. On CLFS nodes, this network connects to the site LDAP for file ownership authentication. The name and IP address of this network are customized during installation. Note: Connections to additional site-specific networks are optional. ib-net InfiniBand® network used by the slave nodes for Lustre LNET traffic. failover-net Internal failover network used between two CIMS servers in an HA configuration for heartbeats between the active/passive CIMS nodes. This network does not connect to a managed switch. Note: Depending on system configuration, additional network configuration may be required. The network parameters can be modified using Bright Network→Settings tab from the cmgui. See Figure 33. Table 6 lists and describes the network parameters you can modify either from the cmgui or from the cmsh. Figure 33. Network Settings GUI S–2327–B 93 Data Management Platform (DMP) Administrator's Guide Table 6. Network Configuration Settings Setting Description Name Name of this network Domain name DNS domain associated with the network Management network Modify this setting if nodes are managed by the CIMS. External network Modify this setting if it is an external network. Base address Base address of the network (also known as the network address) Gateway Default route IP address Netmask bits Prefix-length, or number of bits in netmask. The part after the / in classless inter-domain routing (CIDR) notation. MTU Maximum Transmission Unit. The maximum size of an IP packet transmitted without fragmenting. Dynamic range start/end Start/end IP addresses of the DHCP range temporarily used by nodes during PXE boot on the internal network. Allow node booting Nodes set to boot from this network (useful in the case of nodes on multiple networks). Don’t allow nodes to boot from this network New nodes are not offered a PXE DHCP IP address from this network (DHCPD is locked down by default in /cm/local/apps/cmd/etc/cmd.conf). The lockdowndhcpd setting is can also be configured in cmsh network mode for a specific network. The example cmsh commands below show how to view or set the network parameters for a DMP system CIMS node (esms1) and esmaint-net network. IP address esms1# cmsh -c "device interfaces esms-1; get eth1 ip" esms1# cmsh -c "device interfaces esms-1; set eth1 ip address;commit" 94 S–2327–B CIMS Configuration Tasks [3] Base address esms1# cmsh -c "network get esmaint-net baseaddress" esms1# cmsh -c "network; set esmaint-net baseaddress address;commit" Broadcast address esms1# cmsh -c "network get esmaint-net broadcastaddress" esms1# cmsh -c "network; set esmaint-net broadcastaddress address;commit" Netmask bits esms1# cmsh -c "network get esmaint-net netmaskbits" esms1# cmsh -c "network; set esmaint-net netmaskbits bitsize;commit" Gateway esms1# cmsh -c "network get esmaint-net gateway" esms1# cmsh -c "network; set esmaint-net gateway address; commit" Name servers esms1# cmsh -c "partition get base nameservers" esms1# cmsh -c "partition; set base nameservers address; commit" Search domains esms1# cmsh -c "partition get base searchdomains" esms1# cmsh -c "partition; set base searchdomains hostname;commit" Time servers esms1# cmsh -c "partition get base timeservers" esms1# cmsh -c "partition; set base timeservers hostname; commit" 3.11.1 The sipcalc Utility The sipcalc utility installed on the CIMS node is a useful tool for calculating or checking such IP subnet values (see the man page on sipcalc or see sipcalc -h for help on this utility). Example 9. The sipcalc Utility esms1# sipcalc 192.168.0.1/28 -[ipv4 : 192.168.0.1/28] - 0 [CIDR] Host address - 192.168.0.1 Host address (decimal) - 3232235521 Host address (hex) - C0A80001 Network address - 192.168.0.0 Network mask - 255.255.255.240 Network mask (bits) - 28 Network mask (hex) - FFFFFFF0 Broadcast address - 192.168.0.15 Cisco wildcard - 0.0.0.15 Addresses in network - 16 Network range - 192.168.0.0 - 192.168.0.15 Usable range - 192.168.0.1 - 192.168.0.14 S–2327–B 95 Data Management Platform (DMP) Administrator's Guide 3.11.2 DNS Domains Every network has an associated DNS domain which can be used to access a device through a particular network. For esmaint-net, the default DNS domain is set to esmaint-net.cluster, which means that the host name esms1.cm.cluster can be used to access device esms1 through the maintenance network. The InfiniBand® network domain is ib-net.cm.cluster. Internal DNS zones are generated automatically based on the network definitions and the defined nodes on these networks. For networks marked as external, no DNS zones are generated. 3.11.3 Adding a Network In cmsh, a new network can be added from the network mode using the add or clone commands. The default assignment of networks can be set from the GUI Management network and External network menus on the Settings tab of the top level DMP system object in the resource tree. Figure 34. Adding a Network In cmsh the assignment to Management network and External network is set or modified from the base object in partition mode: 96 S–2327–B CIMS Configuration Tasks [3] Example 10. Changing the Default Setting of a Network esms1# cmsh [esms1]% partition use base [esms1->partition[base]]% set managementnetwork esmaint-net; commit [esms1->partition[base]]% set externalnetwork site-user-net; commit 3.11.4 Changing Node Host Names The alias master may be used to reach the head node. The name can be changed in a similar manner for each, following the guidelines in Devices and Device Names in Bright on page 35. Procedure 17. Changing node hostnames 1. To change the host name of the head node (CIMS), the CIMS device object listed under Head Nodes must be modified (see Figure 35). 2. Using cmgui, select the device listed under Head Nodes in the resource tree, then select the Settings tab. Figure 35. Changing Node Host Names with cmgui 3. Modify the Hostname property (follow guides in Devices and Device Names in Bright on page 35), and click on the Save button. In cmsh, the host name of the head node is changed in device mode: S–2327–B 97 Data Management Platform (DMP) Administrator's Guide Example 11. Changing node host names with cmsh esms1# cmsh [esms1]% device use esms1 [esms1->device[esms1]]% set hostname esms2 [esms2->device*[esms2*]]% commit [esms2->device[esms2]]% quit esms1# sleep 30; hostname -f esms2.cm.cluster esms2.cm.cluster 3.11.5 Adding Hostname to an Internal Network Hostname can be added as name/value pairs to the /etc/hosts file(s) within the system, but it is recommended to let Bright manage host name resolution for devices on the esmaint-net through its DNS server on the esmaint-net interface. Multiple host names can be added as space-separated entries. The named service automatically restarts within about 20 seconds after committal, implementing the configuration changes. The system restarts automatically when there are changes made to service configurations by cmgui or cmsh. Note: The CIMS in Figure 36 is glacier. The cmgui can be used to add a host name to a network by selecting the CIMS head node (glacier) in the resource tree, then the Networks tab, and physical device for eth0 and the esmaint-net, and clicking Edit. See Figure 36. 98 S–2327–B CIMS Configuration Tasks [3] Figure 36. Adding a Hostname to an Internal Network In cmsh, the host names can be added to the additionalhostnames object, from within interfaces submode for the CIMS. In Example 12, the CIMS is glacier). The interfaces submode is accessible from the device mode. Thus, for the CIMS, with eth0 as the interface for esmaint-net: S–2327–B 99 Data Management Platform (DMP) Administrator's Guide Example 12. Adding host names to an internal network using cmsh glacier# cmsh [glacier]% device use glacier [glacier->device[glacier]]% interfaces [glacier->device[glacier]->interfaces]% list Type Network device name IP Network ------------ -------------------- ---------------- ---------------- bmc ipmi0 aaa.bbb.ccc.ddd site-admin-net physical eth0 [prov] 10.141.255.254 esmaint-net physical eth1 aaa.bbb.ccc.ddd site-admin-net physical eth2 10.148.255.254 ipmi-net [glacier->device[glacier]->interfaces]% use eth0 [glacier->device[glacier]->interfaces[eth0]]% set additionalhostnames test [glacier->device*[glacier*]->interfaces*[eth0*]]% commit [glacier->device[glacier]->interfaces[eth0]]% [glacier->device[glacier]->interfaces[eth0]]% Tue Jan 22 16:40:29 2013 [notice] glacier: Service named was restarted [glacier->device[glacier]->interfaces[eth0]]% !ping test PING test.cm.cluster (10.141.255.254) 56(84) bytes of data. 64 bytes from glacier.cm.cluster (10.141.255.254): icmp_seq=1 ttl=64 time=0.038 ms 64 bytes from glacier.cm.cluster (10.141.255.254): icmp_seq=2 ttl=64 time=0.033 ms Note: The ! symbol can be used to invoke Linux commands such as ping, when in the cmsh. 3.11.6 Changing External Network Parameters for the System Changing the network parameters of a DMP system (apart from the IP address of the system) requires making changes to the external network object (site-admin-net, site-user-net), and the system object network settings. 3.11.6.1 Changing the External Network Object Settings External network objects (site-admin-net or site-user-net) contain the network settings to enable connections to the external network, for example, a head node. Network settings are configured in the Settings tab of the Networks resource of cmgui. See Figure 37. 100 S–2327–B CIMS Configuration Tasks [3] Figure 37. Changing External Network Object Parameters The following external network parameters can be configured: • IP network parameters of the system (but not the IP address of the system): – Base address: the IP address of the external network. This is not to be confused with the IP address of the system. – Netmask bits: the netmask size, or prefix-length, of the external network, in bits. – Gateway: the default route for the external network. – Dynamic range start and Dynamic range end: Not used by the external network configuration. • Domain name: the network domain (LAN domain, i.e. what domain machines on the external network use as their domain) • Name: the network name such as, site-admin-net, site-user-net, site-net) • The External network checkbox: this is checked for a Type 1 cluster (nodes are connected on a private internal network) • MTU: size (the maximum value for a TCP/IP packet before it fragments on the external network the default value is 1500) S–2327–B 101 Data Management Platform (DMP) Administrator's Guide 3.11.6.2 Changing Network Settings for the CIMS The CIMS (head node object) contains other network settings used to connect to the outside. These are configured in the Settings tab of the head node object resource in cmgui. See Figure 38. These settings are e-mail address(es) for the administrator, the external name servers used by the system to resolve external host names, the DNS search domain (what the cluster uses as its domain), and NTP time servers (used to synchronize the time on the system with standard time) and time zone settings. Figure 38. Changing Network Settings for the CIMS The static IP address of the head node can also be changed using cmsh in the base object under partition mode. 102 S–2327–B CIMS Configuration Tasks [3] Example 13. Changing the Network Settings for the CIMS esms1# cmsh [esms1]% network use site-admin-net [esms1->network[site-admin-net]]% set baseaddress 192.168.1.0 [esms1->network*[site-admin-net*]]% set netmaskbits 24 [esms1->network*[site-admin-net*]]% set gateway 192.168.1.1 [esms1->network*[site-admin-net*]]% commit [esms1->network[site-admin-net]]% partition use base [esms1->partition[base]]% set nameservers 192.168.1.1 [esms1->partition*[base*]]% set searchdomains searchdomain1.com searchdomain2.com [esms1->partition*[base*]]% append timeservers ntp.timeserver1.com ntp.timeserver2.com [esms1->partition*[base*]]% commit [esms1->partition[base]]% device use esms1 [esms1->device[esms1]]% interfaces [esms1->device[esms1]->interfaces]% use eth1 [esms1->device[esms1]->interfaces[eth1]]% set ip 192.168.1.176 [esms1->device[esms1]->interfaces*[eth1*]]% commit [esms1->device[esms1]->interfaces[eth1]]% exit; exit; [esms1->device]% reboot Note: Reboot the CIMS to activate the changes. 3.11.7 Using DHCP to Supply Network Values for the External Interface Connecting the DMP system via DHCP on the external network is not recommended. This is because DHCP-related issues can complicate network troubleshooting when compared with using static assignments. 3.12 Isolate a Slave Node for Testing This procedure describes how to isolate a slave node (in this example, eslogin1) for testing. Procedure 18. Isolating slave node for testing 1. Log in to the CIMS node as root. 2. Use Bright cmsh to create a test image. S–2327–B 103 Data Management Platform (DMP) Administrator's Guide a. Copy (clone) the current working software image. Choose a unique name to identify the new test image. Note: This example clones the image name ESL-XE-2.1.0 to ESL-test-image. Copy the image from the UNIX® prompt and wait for the copy to complete before cloning the image in Bright. This ensures the clone operation is complete before you continue. esms1# cp -pr /cm/images/ESL-XE-2.1.0 /cm/images/ESL-test-image esms1# cmsh [esms1]% softwareimage [esms1->softwareimage]% list Name (key) Path Kernel version --------------------- ------------------------------------- ----------------- default-image /cm/images/default-image 3.0.80-0.5-default ESL-XE-2.1.0 /cm/images/ESL-XE-2.1.0 3.0.80-0.5-default [esms1->softwareimage]% clone ESL-XE-2.1.0 ESL-test-image [esms1->softwareimage*[ESL-test-image*]]% commit b. Create a test category from your default slave node category (in this example, esLogin-XE) and assign the cloned image to that category. [esms1->softwareimage[ESL-test-image]]% category [esms1->category]% clone esLogin-XE esLogin-test [esms1->category*[esLogin-test*]]% set softwareimage ESL-test-image [esms1->category*[esLogin-test*]]% commit c. Temporarily assign a CDL node (in this example, eslogin1) to the esLogin-test category. [esms1->category[esLogin-test]]% device [esms1->device]% use eslogin1 [esms1->device[eslogin1]]% set category esLogin-test [esms1->device*[eslogin1*]]% commit [esms1->device[eslogin1]]% category [esms1->category]% list Name (key) Software image ------------------------ ------------------------ default default-image esLogin-XC default-image esLogin-XE ESL-XE-2.1.0 esLogin-test ESL-test-image [esms1->category]% usedby esLogin-test Category used by the following: Type Name Parameter Autochange ---------------- ------------------------ ------------------------ ------------ Device eslogin1 category no d. Open a new shell window and log in to the CIMS as root. e. Start cmsh, and launch a remote console (rconsole) on the CDL node. esms1# cmsh [esms1]% device use eslogin1 [esms1->device[eslogin1]]% rconsole 104 S–2327–B CIMS Configuration Tasks [3] f. In a separate CIMS window, login as root and reboot the slave node (eslogin1 in the example) using cmsh or use the Reboot button from the cmgui. esms1# cmsh [esms1]% device use eslogin1 [esms1->device[eslogin1]]% reboot 3. Verify the node boots without errors before you begin your testing. 4. Install and configure the new software image using the test node, and routinely test boot the system to verify proper operation. 5. After you have created the new software image, move the slave node out of the esLogin-test category, back into the default CDL category (in this example, esLogin-XE). esms1# cmsh [esms1]% device use eslogin1 [esms1->device[eslogin1]]% set category esLogin-XE [esms1->device*[eslogin1*]]% commit [esms1->device[eslogin1]]% list Type Hostname (key) MAC Category Ip Network -------------- ---------------- ------------------ ------------ ----------- --------- ... PhysicalNode eslogin1 00:00:00:00:00:00 esLogin-XE 10.141.0.2 esmaint-net 6. Clone the new test image (ESL-test-image) into the new working CDL software image (in this example, ESL-XE-2.1.0_CLE4.2). Copy the image from the UNIX® prompt and wait for the copy to complete before cloning the image in Bright. This ensures the clone operation is complete before you continue. [esms1->device[eslogin1]]% quit esms1# cp -pr /cm/images/ESL-test-image /cm/images/ESL-XE-2.1.0_CLE4.2 esms1# cmsh [esms1]% softwareimage clone ESL-test-image ESL-XE-2.1.0_CLE4.2 [esms1->softwareimage*[ESL-XE-2.1.0_CLE4.2*]]% commit [esms1->softwareimage[ESL-XE-2.1.0_CLE4.2]]% list Name (key) Path Kernel version ------------------------------ ----------------------------------------- ---------------------- ESL-XE-2.1.0 /cm/images/ESL-XE-2.1.0 2.6.32.59-0.7-default ESL-XE-2.1.0_CLE4.2 /cm/images/ESL-XE-2.1.0_CLE4.2 2.6.32.59-0.7-default ESL-test-image /cm/images/ESL-test-image 2.6.32.59-0.7-default default-image /cm/images/default-image 3.0.38-0.5-default 7. Use Bright to assign the new default CDL software (ESL-XE-2.1.0_CLE4.2) to the default CDL category (esLogin-XE). [esms1->softwareimage[ESL-XE-2.1.0_CLE4.2]]% category [esms1->category]% use esLogin-XE [esms1->category*[esLogin-XE*]]% set softwareimage ESL-XE-2.1.0_CLE4.2 [esms1->category*[esLogin-XE*]]% commit 8. Reboot all of slave nodes in the esLogin-XE category with the new image. [esms1->category[esLogin-XE]]% device reboot -c esLogin-XE S–2327–B 105 Data Management Platform (DMP) Administrator's Guide 3.13 Changing Node Category Sometimes it is necessary to change a nodes configuration for testing or in preparation for an image update. Procedure 19 describes how to change a node to a testing category which can be used to verify a new image update, etc. Procedure 19. Change node configuration 1. Log in to the CIMS as root. 2. Type cmsh on the command line to enter the command shell: esms1# cmsh [esms1]% 3. Type category at the cmsh prompt to enter category mode and display the category mode prompt: [esms1->category]% 4. Clone the category of the node you wish to test. For example, we have node eslogin01 in category eslogin. We wish to test a new image (eslogin-new-image) with this node. [esms1->category]% clone eslogin eslogin-test; commit 5. Now set the software image for eslogin-test to be the new image. [esms1->category]% set eslogin-test softwareimage eslogin-new-image; commit 6. Change to device mode and set the category for the node (eslogin01) to be eslogin-test. [esms1->category]% device [esms1->device]% set eslogin01 category eslogin-test; commit The new configuration is applied when the node reboots. 3.14 Creating a CDL Node Group Optional: Node groups can simplify and automate administration tasks by allowing management operations to be performed on groups of nodes. It is not necessary to configure node groups to manage the system. Nodes may belong to several groups at the same time. There are no parameters associated with a node group other than the member nodes. For sites with multiple CDL nodes, Cray recommends creating a node group for these nodes. Procedure 20. Creating a CDL node group 1. Log in to the CIMS as root and start cmsh. esms1# cmsh 106 S–2327–B CIMS Configuration Tasks [3] 2. Switch to nodegroup mode: [esms1]% nodegroup [esms1->nodegroup]% 3. Use the add command to add a node group. This example creates a new node group called Login. [esms1->nodegroup]% add Login [esms1->nodegroup*[Login*]]% 4. Use the append command to add nodes to the group. Note: Multiple nodes can be added as a list (N) or a range (node1..nodeN). [esms1->nodegroup*[Login*]]% append nodes eslogin1..eslogin5 5. Commit your changes. [esms1->nodegroup*[Login*]]% commit [esms1->nodegroup[Login]]% 6. Exit cmsh. [esms1->nodegroup[Login]]% quit esms1# 3.15 Configuring Virtual Network Computing (VNC®) on the CIMS Virtual Network Computing (VNC®) software enables you to view and interact with the CIMS from another computer. The Cray system provides a VNC server, Xvnc; you must download a VNC client to connect to it. Refer to TightVNC (http://www.tightvnc.com/) for more information. Note: The VNC software requires a TCP/IP connection between the server and the viewer. Firewalls or/and site security may restrict this connection. Cray configures a VNC account cray-vnc. Procedure 21. Starting the VNC server 1. Log in to the CIMS as root. 2. Use the chkconfig command to check the current status of the server: esms1# chkconfig vnc vnc off 3. Disable xinetd startup of Xvnc. If the chkconfig command you executed in step 2 reports that Xvnc was started by INET services (xinetd): esms1# chkconfig vnc vnc xinetd S–2327–B 107 Data Management Platform (DMP) Administrator's Guide Execute the following commands to disable xinetd startup of Xvnc (xinetd startup of Xvnc is the SLES 11 default, but it usually is disabled by chkconfig): esms1# chkconfig vnc off esm1# /etc/init.d/xinetd reload Reload INET services (xinetd). done If no other xinetd services have been enabled, the reload command will return failed instead of done. If the reload command returns failed, this is normal and you can ignore the failed notification. 4. Use the chkconfig command to start Xvnc at boot time: esms1# chkconfig vnc on 5. Start the Xvnc server immediately: esms1# /etc/init.d/vnc start If the password for cray-vnc has not already been established, the system prompts you for one. You must enter a password to access the server. Password: ******** Verify: Would you like to enter a view-only password (y/n)? n xauth: creating new authority file /home/cray-vnc/.Xauthority New 'X' desktop is esms1:1 Creating default startup script /home/cray-vnc/.vnc/xstartup Starting applications specified in /home/cray-vnc/.vnc/xstartup Log file is /home/cray-vnc/.vnc/esms1:1.log esms1# ps -eda | grep vnc 1839 pts/0 00:00:00 Xvnc Note: The startup script starts the Xvnc server for display :1. To access the Xvnc server, use a VNC client, such as vncviewer, tight_VNC, vnc4, or a web browser. Direct it to the CIMS that is running Xvnc. Many clients allow you to specify whether you want to connect in view-only or in an active mode. If you choose active participation, every mouse movement and keystroke made in your client is sent to the server. If more than one client is active at the same time, your typing and mouse movements are intermixed. Note: Commands entered through the VNC client affect the system as if they were entered from the CIMS. However, the main CIMS window and the VNC clients cannot detect each other. It is a good idea for the administrator who is sitting at the CIMS to access the system through a VNC client. 108 S–2327–B CIMS Configuration Tasks [3] Procedure 22. Connecting to VNC server through an ssh tunnel, using the vncviewer Important: This procedure is for use with the TightVNC client program. Verify that you have the vncviewer -via option available. If you do not, use Procedure 23 on page 109. • If you are connecting from a workstation or laptop running Linux™, enter the vncviewer command shown below. The first password you enter is for cray-vnc on the CIMS. The second password you enter is for the VNC server on the CIMS, which was set when the VNC server was started for the first time using /etc/init.d/vnc start on the CIMS. > vncviewer -via cray-vnc@esms1 localhost:1 Password: ******** VNC server supports protocol version 3.130 (viewer 3.3) Password: ******** VNC authentication succeeded Desktop name "cray-vnc's X desktop (esms:1)" Connected to VNC server, using protocol version 3.3 Procedure 23. Connecting to the VNC server through an ssh tunnel Note: This procedure assumes that the VNC server on the CIMS is running with the default port of 5901. 1. This ssh command starts an ssh session between the local Linux computer and the CIMS, and it also creates an SSH tunnel so that port 5902 on the local host is forwarded through the encrypted SSH tunnel to port 5901 on the CIMS. You will be prompted for the cray-vnc password on the CIMS. local_linux_prompt> ssh -L 5902:localhost:5901 esms1 -l cray-vnc Password: cray-vnc@esms1> 2. Now vncviewer can be started using the local side of the SSH tunnel, which is port 5902. You will be prompted for the password of the VNC server on the CIMS. This password was set when the VNC server was started for the first time using /etc/init.d/vnc start on the CIMS. remote% vncviewer localhost:2 Connected to RFB server, using protocol version 3.7 Performing standard VNC authentication Password: The VNC window from the CIMS appears. All traffic between the vncviewer on the local Linux computer and the VNC server on the CIMS is now encrypted through the SSH tunnel. S–2327–B 109 Data Management Platform (DMP) Administrator's Guide Procedure 24. Connecting an Apple® Mac® OS X system to the VNC server through an ssh tunnel Note: This procedure assumes that the VNC server on the CIMS is running with the default port of 5901. 1. The following ssh command starts an ssh session between the local Mac OS X® computer and the CIMS, and it also creates an SSH tunnel so that port 5902 on the localhost is forwarded through the encrypted SSH tunnel to port 5901 on the CIMS. You will be prompted for the cray-vnc password on the CIMS. local_mac_prompt> ssh -L 5902:localhost:5901 esms1 -l cray-vnc Password: cray-vnc@esms1> 2. The vncviewer can now be started using the local side of the SSH tunnel, which is port 5902. You will be prompted for the password of the VNC server on the CIMS. This password was set when the VNC server was started for the first time using /etc/init.d/vnc start on the CIMS. If you type this on the Mac OS X command line after having prepared the SSH tunnel, the vncviewer window displays. local_mac_prompt% open vnc://localhost:5902 The VNC window from the CIMS appears. All traffic between the vncviewer on the local Mac OS X computer and the VNC server on the CIMS is now encrypted through the SSH tunnel. Procedure 25. Connecting to the VNC server through an ssh tunnel with Windows® Note: If you are connecting from a computer running Windows®, then both a VNC client program, such as TightVNC and an SSH program, such as PuTTY, SecureCRT®, or OpenSSH are recommended. 1. The same method described in Procedure 23 can be used for computers running the Windows operating system. Although TightVNC encrypts VNC passwords sent over the network, the rest of the traffic is sent unencrypted. To avoid a security risk, install and configure an SSH program that creates an SSH tunnel between TightVNC on the local computer (localhost port 5902) and the remote VNC server (localhost port 5901). Note: Details about how to create the SSH tunnel vary amongst the different SSH programs for Windows computers. 2. After installing TightVNC, start the VNC viewer program by double-clicking on the TightVNC icon. Enter the host name and VNC screen number, localhost:number (such as, localhost:2 or localhost:5902), and then click on the Connect button. 110 S–2327–B CIMS Configuration Tasks [3] 3.16 Adding a Managed Switch or Device to the Bright Configuration You can include Ethernet, InfiniBand® (IB), Fibre Channel (FC), or serial-attached SCSI (SAS) switches, RAID controllers, or intelligent PDU to the Bright configuration. Refer to the device documentation supplied by the manufacturer for configuration and setup procedures. Bright uses SNMP community strings to communicate to devices. SNMP must be enabled for the device and the SNMP community strings should be configured correctly. By default, the SNMP community strings for switches and PDUs are typically set to public and private for respectively read and write access. Example 14 shows how to configure SNMP community strings for an Ethernet switch using cmsh. Example 14. Changing SNMP community strings for devices [esms1]% device use switch1-esmaint-net [esms1->device[switch1-esmaint-net]]% get readstring public [esms1->device[switch1-esmaint-net]]% get writestring private [esms1->device[switch1-esmaint-net]]% set readstring public2 [esms1->device*[switch1-esmaint-net*]]% set writestring private2 [esms1->device*[switch1-esmaint-net*]]% commit The following procedure describes how to setup a Mellanox IS50xx series IB switch and configure Bright to manage it. Refer to Figure 3 for the standard IP addressing scheme used on the esmaint-net network for switches or other devices. Note: Most device command-line interfaces (CLIs) have built-in help systems that can be displayed by entering ? on the command line. Some switches also support a context-sensitive help system that displays valid commands or command options when pressing the Tab key. Uplink ports (switch ports that are connected to other switches or to the esmaint-net) must be configured in Bright. CMDaemon must be told about any switch ports that are uplink ports, or the traffic passing through an uplink port will lead to mistakes in what CMDaemon knows about port and MAC correspondence. Procedure 26. Adding a Mellanox IS50XX series switch to the Bright configuration 1. Connect the serial management port on the switch to a laptop or PC running terminal emulator software (e.g., minicom, Putty, etc.). Note: Settings are typically 9600 Baud, 8N1, no flow control, and VT100 emulation. 2. Press return in emulator software console window to display the console prompt. It may be necessary to power cycle the switch to reset the console. 3. Do not use the setup wizard. Enter Ctrl-z to exit the wizard, if it starts. S–2327–B 111 Data Management Platform (DMP) Administrator's Guide 4. Enter the login and password (refer to the switch documentation for default login and password). Mellanox FabricIT Switch Management switch-5e0120 login: admin Password: admin Last login: Mon Aug 20 12:55:57 on ttyS0 Mellanox Switch 5. Type a question mark ? on the command line to display valid commands from the current mode. switch-5e0120 [standalone: master] > ? cli Configure CLI shell options enable Enter enable mode exit Log out of the CLI fabric Manage fabric diagnostics help View description of the interactive help system no Negate or clear certain configuration options ping Send ICMP echo requests to a specified host show Display system configuration or statistics slogin Log into another system securely using ssh telnet Log into another system using telnet terminal Set terminal parameters test Diagnostics traceroute Trace the route packets take to a destination ib-switch-1 [standalone: master] > 6. At the console prompt, run the following commands to enable switch configuration from a terminal: switch-5e0120 [standalone: master]# enable switch-5e0120 [standalone: master]# configure terminal 7. Set up simple network management protocol (SNMP). conswitch-5e0120 [standalone: master] (config)# snmp-server community public 8. Set the IP address and netmask of the Ethernet port used to connect to the esmaint-net network (in this example eth0). switch-5e0120 [standalone: master] (config)# interface eth0 ip address 10.141.200.1 255.255.255.0 9. Set a host name for the switch. switch-5e0120 [standalone: master] (config)# hostname ib-switch-1 ib-switch1 [standalone: master] (config) # 10. Write the configuration to memory and exit. ib-switch1 [standalone: master] (config) # write memory ib-switch1 [standalone: master] (config) # ib-switch-1 [standalone: master] (config) # exit ib-switch-1 [standalone: master] # exit Mellanox FabricIT Switch Management ib-switch-1 login: 112 S–2327–B CIMS Configuration Tasks [3] 11. Log in to the CIMS as root and start cmsh. esms1# cmsh [esms1]% 12. From device mode, add the IB switch to Bright. Your options for device type are: cloudnode, physicalnode, virtualsmpnode, headnode, ethernetswitch, ibswitch, myrinetswitch, powerdistributionunit, genericdevice, racksensor, chassis, gpuunit. Use the host name that you configured for the switch in step 9. [esms1]% device add ibswitch ib-switch-1 [esms1->device*[ib-switch-1*]]% 13. Set the management network to esmaint-net. [esms1->device*[ib-switch-1*]]% set network esmaint-net 14. Set the IP address for interface configured in step 8. [esms1->device*[ib-switch-1*]]% set ip 10.141.200.1 15. Configure the SNMP read string to public and write string to private. [esms1->device*[ib-switch-1*]]% set readstring public [esms1->device*[ib-switch-1*]]% set writestring private 16. (Optional) Use the cmsh set command to set other switch parameters such as rack ID, deviceheight (1U), deviceposition in rack, mac address, hardware tag, and administrator notes. All of these settings can be configured using the cmgui after the switch configured in Bright. 17. Commit the changes and list the devices managed by Bright. [esms1->device*[ib-switch-1*]]% commit [esms1->device[ib-switch-1]]% list esms1# [esms1->device[ib-switch-1]]% quit esms1#Type Hostname (key) MAC Category Ip Network ---------------------- ------------------- ------------------ ---------------- --------------- ------------ EthernetSwitch esmaint-net-switch 00:0F:8F:8E:9D:C0 10.141.50.1 esmaint-net EthernetSwitch ipmi-net-switch 00:0B:5F:CE:2F:40 10.148.50.1 ipmi-net EthernetSwitch wlm-net-switch 00:00:00:00:00:00 10.128.100.1 wlm-net HeadNode esms1 78:2B:CB:40:CE:CA 10.141.255.254 esmaint-net IBSwitch ib-switch-1 00:00:00:00:00:00 10.141.200.1 esmaint-net PhysicalNode eslogin-001 84:2B:2B:61:B0:04 esLogin-XC 10.141.0.37 esmaint-net PhysicalNode mds001 78:2B:CB:50:E9:A3 -mds 10.141.0.10 esmaint-net [esms1->device[ib-switch-1]]% 18. Exit cmsh. [esms1->device*[ib-switch-1*]]% quit esms1# S–2327–B 113 Data Management Platform (DMP) Administrator's Guide 3.17 DELL™ 5548 Switch Configuration The CIMS system should include configuration settings for Ethernet switches so that they can be monitored by Bright. Use this procedure to change switch settings for a Dell 5548 switch if your system is not pre-configured, or if you need to reconfigure another Ethernet switch. For the VLAN port assignments, see Figure 2. Procedure 27. Configuring the Dell 5548 1GbE switch Note: This procedure shows the instructions for a 48-port Ethernet switch. For a 24-port Ethernet switch, use the VLAN port rules and example commands to adapt the configuration for a smaller switch. 1. Connect the serial port of the switch to a suitable VT100 emulator (minicom, Putty, etc.). Note: Settings are 9600 Baud, 8N1, no flow control, VT100 emulation. 2. Power on the switch. 3. Do not use the setup wizard. Enter Ctrl-z to exit the wizard, if it starts. 4. At the console prompt, run the following commands: console> enable console# config 5. Set up SNMP. console (config)# snmp-server community public 6. Set up the VLANs. console (config)# vlan database console (config-vlan)# vlan 2 console (config-vlan)# vlan 3 console (config-vlan)# vlan 4 console (config-vlan)# exit console (config)# interface vlan 1 console (config-if)# name esmaint-net console (config-if)# interface vlan 2 console (config-if)# name ipmi-net console (config-if)# interface vlan 3 console (config-if)# name site-admin-net console (config-if)# interface vlan 4 console (config-if)# name site-user-net console (config-if)# exit 7. Configure management IP and the netmask. This is always on VLAN 1 (on esmaint-net). console (config)# interface vlan 1 console (config-if)# ip address 10.141.0.100 255.255.0.0 console (config-if)# exit 114 S–2327–B CIMS Configuration Tasks [3] 8. Set the default gateway of VLAN 1 to be the IP address of eth0 on the CIMS. console (config)# ip default-gateway 10.141.255.254 9. Configure the ports to the VLANs using the scheme shown in Figure 2. Note: In the commands below, replace NN with the appropriate VLAN port number. a. Configure the remaining VLAN 1 ports (on esmaint-net) by running the following two commands for each port on this VLAN. console (config)# interface gigabitethernet 0/NN console (config-if)# switchport access vlan 1 b. Configure the VLAN 2 ports (on ipmi-net) by running the following two commands for each port on this VLAN. console (config-if)# interface gigabitethernet 0/NN console (config-if)# switchport access vlan 2 c. Configure the VLAN 3 ports (on site-admin-net) by running the following two commands for each port on this VLAN. console (config-if)# interface gigabitethernet 0/NN console (config-if)# switchport access vlan 3 d. Configure the VLAN 4 ports (on site-user-net) by running the following two commands for each port on this VLAN. console (config-if)# interface gigabitethernet 0/NN console (config-if)# switchport access vlan 4 10. Disable spanning tree protocol (STP) to disable loop-free, redundant bridging paths between daisy-chained switches. console (config-if)# no spanning-tree 11. Set up the admin and root users: console (config-if)# exit console (config)# username admin privilege 15 password initial0 console (config)# username root privilege 15 password initial0 console (config)# exit 12. Save the configuration. console# exit console# write Overwrite file [startup-config]? [yes/press any key for no] yes console# exit S–2327–B 115 Data Management Platform (DMP) Administrator's Guide 3.18 Zoning the QLogic® FC Switch If your system includes QLogic® Fibre Channel (FC) switch, follow Procedure 28 on page 116 to zone the LUNs on your QLogic SANBox™ switch by using a utility called QuickTools. Note: If a LUN is to be shared between failover host pairs, each host must be given access to the LUN. The CIMS host port should be given access to all LUNs. QuickTools is an application that is embedded in the QLogic switch and is accessible from a workstation browser with a compatible Java™ plug-in. You must have a Java browser plugin, version 1.4.2 or later. These instructions assume that the disk device has four host ports connected to ports 0-3 for the QLogic SANbox switch. Zoning is implemented by creating a zone set, adding one or more zones to the zone set, and selecting the ports to use in the zone. This procedure presupposes that the SANBox is configured and on esmaint-net network. Procedure 28. Configuring zoning for a QLogic SANbox switch using QuickTools utility 1. Start a web browser. 2. Enter the IP address of your switch on the esmaint-net network. The IP address of each RAID controller is preconfigured by Cray and is listed on a sticker on the back of the RAID controller. 3. Enter the login name and password when the Add a New Fabric window pops up and prompts for them. The default administrative login name is admin, and the default password is password. 4. The QuickTools utility displays in your browser. Click Add Fabric. Note: If you receive a dialog box notification that the request failed to connect over a secured connection, click Yes and continue. 5. The switch is located and displayed in the window. Double-click the switch icon. Information about the switch displays in the right panel. 6. At the bottom of the panel, click the Configured Zonesets tab. 7. From the Tool Bar menu, select Zoning and then Edit Zoning. The Edit Zoning window displays. 8. Click the Zone Set button. The Create a Zone Set window displays. Create a new zone set. (In this example, assume that the zone set is named XT0.) 9. Right-click the XT0 zone and select Create a Zone. 116 S–2327–B CIMS Configuration Tasks [3] 10. Create a new zone name. 11. On the right panel, click the button in front of zonename to open a view of the domain members. 12. Define the ports in the zone to ensure that the discovery of LUNs is consistent among the CIMS, CLFS, and CDL nodes. Using the mouse, left-click on the desire port, and draft it to zonename. 13. Click Apply. The error-checking window displays. 14. When prompted, select Perform Error Check. 15. After confirming that no errors were found, click Save Zoning. 16. When prompted to activate a Zone Set, click Yes and then select the appropriate XT3 zone set. 17. At this point, Cray recommends that you create a backup of your switch configuration (Procedure 29 on page 117) before you close and exit the application. Procedure 29. Creating a backup of your QLogic switch configuration Create a backup of your QLogic switch configuration with the QuickTools utility. You must have a Java browser plugin, version 1.4.2 or later to use QuickTools. If you need to start your web browser and open the QuickTools utility, complete steps 1 through 4. If you currently have the QuickTools utility open, skip to step 5. 1. Start a web browser. 2. Enter the IP address of your switch on esmaint-net. The IP address of each RAID controller is preconfigured by Cray and is listed on a sticker on the back of the RAID controller. 3. Enter the login name and password when the Add a New Fabric window pops up and prompts for them. The RAID default administrative login name is admin, and the default password is password. 4. The QuickTools utility appears. Click Add Fabric. Note: If you receive a dialog box that states that the request failed to connect over a secured connection, click Yes and continue. 5. From within the QuickTools utility, complete the configuration backup. a. At the top bar, select Switch and then Archive. A Save window pops up with blanks for Save in: and File Name:. b. Enter the directory (for example, crayadm) and a file name (for example, sanbox_archive) for saving your QLogic switch configuration. S–2327–B 117 Data Management Platform (DMP) Administrator's Guide c. Click the Save button. 6. Close and exit the application. 3.19 Configuring the InfiniBand (ib-net) Network for Slave Nodes The InfiniBand® network (ib-net) is used only by slave nodes in the DMP system, but is configured on the CIMS. Before adding any slave nodes to the system, you must configure the InfiniBand network (ib-net) in Bright. Procedure 30. Configuring the InfiniBand (ib-net) network in Bright 1. Open a window on the CIMS, log in as root, and enter the cmsh command: esms1# cmsh [esms1]% 2. Switch to network mode: [esms1]% network [esms1->network]% 3. Check the currently configured networks with the list command. These networks were defined in the customized Cray XML configuration file. Note: The network globalnet is created by Bright but is not used in a DMP system. [esms1->network]% list Name (key) Type Netmask bits Base address Domain name IPv6 ---------------- ------------ ---------------- ---------------- -------------------- ---- esmaint-net Internal 16 10.141.0.0 esmaint-net.cluster no globalnet Global 16 0.0.0.0 cm.cluster no ipmi-net Internal 16 10.148.0.0 ipmi-net.cluster no site-admin-net External 20 aaa.bbb.ccc.ddd your.domain.com no 4. Use the clone subcommand to clone a similar network: [esms1->network]% clone ipmi-net ib-net Note: The prompt displayed in step 5.a is abbreviated as "%" in the following steps. 5. Set the network parameters. a. Set the base IP address to 10.149.0.0. % set baseaddress 10.149.0.0 b. Set the domain name to ib-net.cluster. % set domainname ib-net.cluster 118 S–2327–B CIMS Configuration Tasks [3] c. Set the netmask bits to 16. % set netmaskbits 16 d. Set the MTU: % set mtu 2044 e. Set the broadcast address: % set broadcastaddress 10.149.255.255 f. Use the show subcommand to view your changes. % show Parameter Value ------------------------------ ------------------------------------------------ Base address 10.149.0.0 Broadcast address 10.149.255.255 Domain Name ib-net.cluster Dynamic range end 0.0.0.0 Dynamic range start 0.0.0.0 Gateway 0.0.0.0 IPv6 no Lock down dhcpd no MTU 2044 Management allowed no Netmask bits 16 Node booting no Notes Revision Type Internal name ib-net g. Save your changes. % commit 6. Display the changed network list. [esms1->network]% list Name (key) Type Netmask bits Base address Domain name IPv6 ---------------- ------------ ---------------- ---------------- -------------------- ---- esmaint-net Internal 16 10.141.0.0 esmaint-net.cluster no globalnet Global 16 0.0.0.0 cm.cluster no ib-net Internal 16 10.149.0.0 ib-net.cluster no ipmi-net Internal 16 10.148.0.0 ipmi-net.cluster no site-admin-net External 20 aaa.bbb.ccc.ddd your.domain.com no 7. Exit cmsh. % quit S–2327–B 119 Data Management Platform (DMP) Administrator's Guide 3.20 iDRAC Remote Console In addition to the remote console capabilities of Bright, DELL™ servers provide a remote console and administrative interface through the IP address of the iDRAC port on the CIMS that is accessible from a web browser. The iDRAC interface enables CIMS control through the node's baseboard management controller (BMC). Note: Refer to Administrative Passwords on page 60 to change the iDRAC administrative password from the Bright cmsh shell. For detailed information, see the Dell iDRAC documentation: http://support.dell.com/support Procedure 31. Using the iDRAC7 web interface and remote console 1. From a web browser, open the IP address of the CIMS iDRAC port, https://idrac_port_IP. A login screen displays. Figure 39. iDRAC Login Page 2. Enter the account user name (root) and password (initial0). Then click on Submit. The System Summary window displays. CIMS power, control, and status information is available from this interface. 120 S–2327–B CIMS Configuration Tasks [3] Figure 40. iDRAC System Summary Page 3. To access the CIMS console, click on the Console tab. The Virtual Console window appears. 4. Click on Launch Virtual Console. Note: The iDRAC web interface may request to install a Virtual Console Java™ application. Confirm the installation of this application to use the Virtual Console feature. Figure 41. iDRAC Remote Console Window S–2327–B 121 Data Management Platform (DMP) Administrator's Guide Tip: Press the F9 key to access the console window menu. Tip: To logout of the virtual console, close the window or select File, then Exit from the console menu. You will still be logged into the iDRAC port from your web browser. 3.21 Configuring Administrator Email Alerts from the CIMS Procedure 32. Configuring administrator email alerts from the CIMS using cmgui 1. Open the cmgui and select the DMP system name in the RESOURCES tree. 2. Select the Settings tab. 3. Click on the + symbol (see Figure 42), and enter the administrator Email address in the Administrator e-mail field. Figure 42. Administrator's Email Address Procedure 33. Configuring administrator email alerts from the CIMS using cmsh 1. Log in to the CIMS as root and start cmsh. esms1# cmsh [esms1]% 122 S–2327–B CIMS Configuration Tasks [3] 2. Switch to partition mode. [esms1]%partition [esms1->partition]% 3. Use the base object in partition mode. [esms1->partition]% use base [esms1->partition[base]]% 4. Set the administrator's email address. [esms1->partition[base]]% set administratore-mail johndoe@server.com [esms1->partition*[base*]]% commit Tue Jun 4 10:09:27 2013 [notice] esms1: Service postfix was restarted 5. (Optional) To include additional administrator email addresses, enter: [esms1->partition[base]]% append administratore-mail janesmith@server.com [esms1->partition*[base*]]% get administratore-mail johndoe@server.com janesmith@server.com [esms1->partition*[base*]]% commit Tue Jun 4 10:11:12 2013 [notice] esms1: Service postfix was restarted 6. (Optional) To remove johndoe@server.com from the administrators email list enter: [esms1->partition[base]]% removefrom administratore-mail johndoe@server.com [esms1->partition*[base*]]% commit Tue Jun 4 10:11:12 2013 [notice] esms1: Service postfix was restarted 3.22 Configuring SSH Keys for eswrap on CDL and Internal Login Nodes This procedures describes how to configure SSH Keys on CDL and internal login nodes by using ssh-agent or passphrase-less RSA®/DSA keys. By default, SSH prompts for a password on each command wrapped by eswrap. Note: Consult the site security policies before configuring transparent SSH access. This procedure creates DSA keys. Procedure 34. Configuring SSH Keys for eswrap on CDL and internal login nodes 1. Log in to the CIMS as root. 2. SSH to a CDL node as root and create a key with ssh-keygen. esms1# ssh eslogin1 Last login: Tue May 7 10:52:08 2013 from esms1.cm.cluster eslogin1# S–2327–B 123 Data Management Platform (DMP) Administrator's Guide 3. Select the type of key, either RSA or DSA and the number of bits (DSA keys must be 1024 bits). The choice depends on the site security policy. Save the key in id_dsa.pub. eslogin1# ssh-keygen -t dsa -b 1024 Generating public/private dsa key pair. Enter file in which to save the key (/root/.ssh/id_dsa): /root/.ssh/id_dsa.pub Enter passphrase (empty for no passphrase): Enter same passphrase again: Your identification has been saved in /root/.ssh/id_dsa.pub Your public key has been saved in /root/.ssh/id_dsa.pub The key fingerprint is: 56:cd:94:e4:3f:ef:a2:9a:a2:bc:17:c4:63:8b:a6:1a root@eslogin1 The key's randomart image is: +--[ DSA 1024]----+ | .o. | | =. | | . .+ | | =. . | | +So o | | o.o o | | E o . .| | ... o . .. | | .. ++ .o... ..| +-----------------+ 4. Copy the contents of the public key file (id_dsa.pub or id_dsa.pub). 5. On the Cray internal login node (aka login gateway), edit the file $HOME/.ssh/authorized_keys and append the public key from the CDL file id_dsa.pub. 6. Repeat step 2 through step 4 for all CDL nodes that can access that Cray system. Note: On a system that does not share user home directories between CDL nodes and the Cray system, you can remove the SSH key prompts for an unknown host by setting up a known_hosts file ($HOME/.ssh/known_hosts) and/or authorized_keys file ($HOME/.ssh/authorized_keys) on the Cray system. 7. SSH to an internal Cray login node (craylogin). eslogin1# ssh craylogin The authenticity of host 'craylogin (10.128.1.132)' can't be established. DSA key fingerprint is a8:0d:b0:5c:f8:d2:ec:4f:00:b8:69:87:7d:28:ac:05. Are you sure you want to continue connecting (yes/no)? yes Warning: Permanently added 'craylogin,10.128.1.132' (DSA) to the list of known hosts. Password: Creating directory '/home/users/username'. Welcome to XE system craylogin 8. Create a .ssh directory for login username and cd into that directory. craylogin users/username> mkdir -p ~/.ssh craylogin users/username> cd .ssh 124 S–2327–B CIMS Configuration Tasks [3] 9. Use secure FTP to transfer the id_dsa.pub file you create in step 3 from the eslogin1 node to the Cray internal login node. craylogin users/username/.ssh> sftp username@eslogin1:.ssh/id_dsa.pub . Connecting to eslogin1... The authenticity of host 'eslogin1 (aaa.bbb.ccc.ddd)' can't be established. DSA key fingerprint is b8:87:2c:43:31:2d:9f:64:2b:30:e3:08:45:cb:78:65. Are you sure you want to continue connecting (yes/no)? yes Warning: Permanently added 'eslogin1,aa.bbb.ccc.ddd' (DSA) to the list of known hosts. Password: Fetching /home/users/username/.ssh/id_dsa.pub to ./id_dsa.pub /home/users/username/.ssh/id_dsa.pub 100% 737 0.7KB/s 00:00 craylogin users/username/.ssh> ls id_rsa.pub known_hosts 10. Copy the id_dsa.pub key to an authorized_keys file. craylogin users/username/.ssh> cat id_dsa.pub >> authorized_keys ccraylogin users/username/.ssh> ls authorized_keys id_dsa id_dsa.pub known_hosts 11. Logout of the Cray internal login node. craylogin users/username/.ssh> logout Connection to craylogin closed. S–2327–B 125 Data Management Platform (DMP) Administrator's Guide 12. Verify you can run wrapped commands without entering a password. eslogin1# eswrap --help eslogin1# echo $ESWRAP_LOGIN eslogin1# which xtnodestat /opt/cray/eslogin/eswrap/1.0.15/bin/xtnodestat eslogin1# xtnodestat Current Allocation Status at Mon Sep 10 13:13:00 2012 C0-0 C1-0 C2-0 C3-0 n3 ---------------- AAAAAAAAAAAAAAaa ---------------- --------------AA n2 ---------------- AAAAAAAAAAAAAAaa ---------------- --------------AA n1 ---------------- AAAAAAAAAAAAAAaa ---------------- --------------AA c2n0 ---------------- AAAAAAAAAAAAAAaa ---------------- --------------AA n3 - -------------- AAAAAAAAAAAAAAAA a aaaaaaaaaaaaaa ---------------- n2 -S-------------- AAAAAAAAAAAAAAAA aSaaaaaaaaaaaaaa ---------------- n1 -S-------------- AAAAAAAAAAAAAAAA aSaaaaaaaaaaaaaa ---------------- c1n0 - -------------- AAAAAAAAAAAAAAAA a aaaaaaaaaaaaaa ---------------- n3 --------------- --------------AA aaaaaaaaaaaaaaa ---------------- n2 S--------------- --------------AA Saaaaaaaaaaaaaaa ---------------- n1 S--------------- --------------AA Saaaaaaaaaaaaaaa ---------------- c0n0 --------------- --------------AA aaaaaaaaaaaaaaa ---------------- s0123456789abcdef 0123456789abcdef 0123456789abcdef 0123456789abcdef Legend: nonexistent node S service node ; free interactive compute node - free batch compute node A allocated (idle) compute or ccm node ? suspect compute node W waiting or non-running job X down compute node Y down or admindown service node Z admindown compute node Available compute nodes: 0 interactive, 488 batch Job ID User Size Age State command line --- ------ -------- ----- --------- -------- ---------------------------------- a 302511 addy 128 0h17m run gamess.ga.x eslogin1# 3.23 Save the System Configuration After you have installed the CIMS software and configured all of the nodes and devices in Bright, save the site configuration to an XML file so you can recover the system configuration from a backup. To save your site customizations, use the cmd -x siteconfigfile.xmlcommand, to generate a human-readable configuration file. ! Caution: The cmd -i siteconfigfile.xml command can restore your system configuration only when all the system images available in /cm/images are present and all other site customizations have been configured in Bright. 126 S–2327–B CIMS Configuration Tasks [3] Procedure 35. Save the system configuration settings to an XML file 1. Log in to the CIMS as root. remote% ssh root@esms1 2. Stop the cmd daemon. esms1#/etc/init.d/cmd stop Waiting for CMDaemon (26823) to terminate... done 3. Dump the system configuration to an XML file. # cmd -x SiteConfig_Dump.xml Wed Feb 13 13:37:50 2013 Info: CMDaemon version 1.4 (r15096) Wed Feb 13 13:37:50 2013 Info: Reading configuration from /cm/local/apps/cmd/etc/cmd.conf Wed Feb 13 13:37:50 2013 Info: CMDaemon auditing is disabled Wed Feb 13 13:37:50 2013 Info: Initialize cmdaemon database Wed Feb 13 13:37:50 2013 Info: Initialize monitoring database Wed Feb 13 13:37:50 2013 Info: Database: Mirroring not required to remote master. no partition Wed Feb 13 13:37:50 2013 Info: Database: no index on timestamp present in MonData table Wed Feb 13 13:37:50 2013 Info: Database: using mysql's bulkinsert with interval of 3600s Wed Feb 13 13:37:51 2013 Info: Succesfully stored configuration in SiteConfig_Dump.xml 4. Start the cmd daemon. # /etc/init.d/cmd start Waiting for CMDaemon to start... done Procedure 36. Load system configuration settings from an XML file ! Caution: Use the cmd -i siteconfigfile.xml command with caution. You can restore your system configuration using the cmd -i command only if you have all the system images available in /cm/images and all other site customizations properly configured in Bright. If the cmd -i command fails, it could potentially render the system inoperable. 1. Log in to the CIMS as root. remote% ssh root@esms1 2. Stop the cmd daemon. esms1# /etc/init.d/cmd stop Waiting for CMDaemon (26823) to terminate... done S–2327–B 127 Data Management Platform (DMP) Administrator's Guide 3. Dump the system configuration to an XML file. esms1# cmd -i SiteConfig.xml Wed Feb 13 13:44:48 2013 Info: CMDaemon version 1.4 (r15096) Wed Feb 13 13:44:48 2013 Info: Reading configuration from /cm/local/apps/cmd/etc/cmd.conf Wed Feb 13 13:44:48 2013 Info: CMDaemon auditing is disabled Wed Feb 13 13:44:49 2013 Info: Billing Service enabled Wed Feb 13 13:44:49 2013 Info: Initialize cmdaemon database Wed Feb 13 13:44:49 2013 Info: Drop cmdaemon database Wed Feb 13 13:44:49 2013 Info: Create cmdaemon database Wed Feb 13 13:44:52 2013 Info: Recreate monitoring database . . . 4. Start the cmd daemon. esms1# /etc/init.d/cmd start Waiting for CMDaemon to start... done Note: All slave nodes appear DOWN in Bright until you restart the cmd service on each slave node. 3.24 Resizing Partitions on the CIMS If the CIMS was not configured as an HA system and is being integrated into an HA system, then you must resize the /cm partition to accommodate the required Distributed Replicated Block Device (DRBD) partitions. If DRBD partitions are already defined, do not perform this procedure. Procedure 37. Resizing partitions on a CIMS 1. Log in to the CIMS as root. remote% ssh root@esms1 2. Determine which devices holds the /cm partition. esms1# df -h Filesystem Size Used Avail Use% Mounted on /dev/sda5 177G 3.1G 165G 2% / devtmpfs 32G 240K 32G 1% /dev tmpfs 32G 0 32G 0% /dev/shm /dev/sda1 473M 22M 427M 5% /boot /dev/sdb3 3.3T 30G 3.1T 1% /cm /dev/sda3 61G 180M 57G 1% /tmp /dev/sdb1 962G 373M 913G 1% /var /dev/sdb2 9.4G 155M 8.8G 2% /var/lib/mysql/cmdaemon_mon 3. Backup the existing /cm partition to a tarball in /cm-backup.tar.gz. esms1# tar zcf cm-backup.tar.gz /cm tar: Removing leading /` ' from member names tar: Removing leading /` ' from hard link targets esms1# ls -l cm-backup.tar.gz -rw-r--r-- 1 root root 16384685013 Sep 19 14:25 cm-backup.tar.gz 128 S–2327–B CIMS Configuration Tasks [3] 4. Use the parted utility to verify where /cm is mounted on /dev/sdb (/dev/sdb was determined to hold the /cm partition from step 2. esms1# parted /dev/sdb GNU Parted 2.3 Using /dev/sdb Welcome to GNU Parted! Type 'help' to view a list of commands. (parted) p Model: DELL PERC H710P (scsi) Disk /dev/sdb: 4723GB Sector size (logical/physical): 512B/512B Partition Table: gpt Number Start End Size File system Name Flags 1 17.4kB 1049GB 1049GB ext3 /var 2 1049GB 1059GB 10.2GB ext3 /var/lib/mysql/cmdaemon_mon 3 1059GB 4723GB 3665GB ext3 /cm (parted) quit esms1# 5. Shut down all slave nodes to free /cm/shared which is NFS® mounted on all slave nodes. This example has only one slave node named eslogin1. esms1# cmsh [esms1]% device [esms1->device]% status eslogin1 ............. [ UP ] esms1 .................... [ UP ] sw-1ge ................... [ DOWN ] [esms1->device]% shutdown eslogin1 eslogin1: Shutdown in progress ... [esms1->device]% Wed Sep 19 14:28:20 2012 [notice] esms1: eslogin1 [ DOWN ] [esms1->device]% power status ipmi0 .................... [ OFF ] eslogin1 ipmi0 .................... [ ON ] esms1 No power control ......... [ UNKNOWN ] sw-1ge [esms1->device]% quit 6. Shut down the CMDaemon (cmd). esms1# /etc/init.d/cmd stop Waiting for CMDaemon (4838) to terminate... done 7. Use lsof to determine if /cm is being used. esms1# lsof /cm COMMAND PID USER FD TYPE DEVICE SIZE/OFF NODE NAME slapd 3475 ldap txt REG 8,19 2949776 159072337 /cm/local/apps/openldap/sbin/slapd slapd 3475 ldap mem REG 8,19 441348 159072328 /cm/local/apps/openldap/lib/libldap_r-2.4.so.2.8.3 slapd 3475 ldap mem REG 8,19 81962 159072318 /cm/local/apps/openldap/lib/liblber-2.4.so.2.8.3 slapd 3475 ldap mem REG 8,19 10036285 159065221 /cm/local/apps/openldap/db4/lib/libdb-4.6.so conmand 5035 root txt REG 8,19 387897 159121453 /cm/local/apps/conman/sbin/conmand conmand 5035 root mem REG 8,19 781050 159072826 /cm/local/apps/freeipmi/1.1.3/lib/libipmiconsole.so.2.2.1 conmand 5035 root mem REG 8,19 6524146 159072821 /cm/local/apps/freeipmi/1.1.3/lib/libfreeipmi.so.12.0.2 conmand 5035 root 5rR REG 8,19 414 159121420 /cm/local/apps/conman/etc/conman.conf S–2327–B 129 Data Management Platform (DMP) Administrator's Guide 8. step 7 shows that ldap and conman are using the /cm partition. Stop the ldap and conman services. Because /cm contains /cm/shared, which is NFS exported, stop the nfsserver service, as well. esms1# /etc/init.d/ldap status Checking for service ldap: running esms1# /etc/init.d/ldap stop Shutting down ldap-server done esms1# /etc/init.d/conman stop Stopping ConMan: conmand done esms1# /etc/init.d/nfsserver stop Shutting down kernel based NFS server: nfsd statd mountd done 9. Unmount /cm and run parted to resize the /cm partition by removing it and creating it with a smaller size. a. Unmount /cm. esms1# umount /cm b. Start parted and list the existing partitions esms1# parted /dev/sdb GNU Parted 2.3 Using /dev/sdb Welcome to GNU Parted! Type 'help' to view a list of commands. (parted) p Model: DELL PERC H710P (scsi) Disk /dev/sdb: 4723GB Sector size (logical/physical): 512B/512B Partition Table: gpt Number Start End Size File system Name Flags 1 17.4kB 1049GB 1049GB ext3 /var 2 1049GB 1059GB 10.2GB ext3 /var/lib/mysql/cmdaemon_mon 3 1059GB 4723GB 3665GB ext3 /cm (parted) c. Remove the /cm partition. (parted) rm 3 (parted) p Model: DELL PERC H710P (scsi) Disk /dev/sdb: 4723GB Sector size (logical/physical): 512B/512B Partition Table: gpt Number Start End Size File system Name Flags 1 17.4kB 1049GB 1049GB ext3 /var 2 1049GB 1059GB 10.2GB ext3 /var/lib/mysql/cmdaemon_mon (parted) 130 S–2327–B CIMS Configuration Tasks [3] d. Make a new smaller sized /cm partition and quit. (parted) mkpart /cm 1059GB 3930GB (parted) p Model: DELL PERC H710P (scsi) Disk /dev/sdb: 4723GB Sector size (logical/physical): 512B/512B Partition Table: gpt Number Start End Size File system Name Flags 1 17.4kB 1049GB 1049GB ext3 /var 2 1049GB 1059GB 10.2GB ext3 /var/lib/mysql/cmdaemon_mon 3 1059GB 3930GB 2871GB ext3 /cm (parted) quit Information: You may need to run /etc/fstab. 10. Run makefs.ext3 on the device that contained /cm (/dev/sdb3 in this example). esms1# mkfs.ext3 /dev/sdb3 mke2fs 1.41.9 (22-Aug-2009) Filesystem label= OS type: Linux Block size=4096 (log=2) Fragment size=4096 (log=2) 175243264 inodes, 700972288 blocks 35048614 blocks (5.00%) reserved for the super user First data block=0 Maximum filesystem blocks=4294967296 21392 block groups 32768 blocks per group, 32768 fragments per group 8192 inodes per group Superblock backups stored on blocks: 32768, 98304, 163840, 229376, 294912, 819200, 884736, 1605632, 2654208, 4096000, 7962624, 11239424, 20480000, 23887872, 71663616, 78675968, 102400000, 214990848, 512000000, 550731776, 644972544 Writing inode tables: 10/21392 done Creating journal (32768 blocks): done Writing superblocks and filesystem accounting information: done 11. Mount the /cm partition. esms1# mount /cm esms1# df -h Filesystem Size Used Avail Use% Mounted on /dev/sda5 177G 19G 150G 11% / devtmpfs 32G 244K 32G 1% /dev tmpfs 32G 0 32G 0% /dev/shm /dev/sda1 473M 22M 427M 5% /boot /dev/sda3 61G 180M 57G 1% /tmp /dev/sdb1 962G 373M 913G 1% /var /dev/sdb2 9.4G 155M 8.8G 2% /var/lib/mysql/cmdaemon_mon /dev/sdb3 2.6T 202M 2.5T 1% /cm S–2327–B 131 Data Management Platform (DMP) Administrator's Guide 12. Restore /cm from the backup tarball. esms1# cd /cm esms1:/cm # cp /root/cm-backup.tar.gz . esms1:/cm # tar zxf cm-backup.tar.gz esms1:/ # ls /cm CLUSTERMANAGERID conf local node-installer shared README images lost+found nodeinstaller 13. Start nfsserver and cmd. esms1:/ # /etc/init.d/nfsserver start Starting kernel based NFS server: mountd statd nfsdrpc.nfsd: address family inet6 not supported by protocol TCP sm-notify done esms1:/ # /etc/init.d/cmd start Waiting for CMDaemon to start... done 14. Boot the slave nodes. esms1:/ # cmsh [esms1]% device [esms1->device]% list Type Hostname (key) MAC Category Ip Network PowerDistributio ---------------------- ---------------- ------------------ ---------------- --------------- -------------- EthernetSwitch sw-1ge 00:00:00:00:00:00 10.141.253.1 esmaint-net MasterNode esms1 D4:AE:52:B5:E2:64 10.141.255.254 esmaint-net PhysicalNode eslogin1 D4:AE:52:B5:A4:68 eslogin 10.141.0.1 esmaint-net [esms1->device]% power status ipmi0 .................... [ OFF ] eslogin1 ipmi0 .................... [ ON ] esms1 No power control ......... [ UNKNOWN ] sw-1ge [esms1->device]% power -n eslogin1 on ipmi0 .................... [ ON ] eslogin1 [esms1->device]% Wed Sep 19 15:16:08 2012 [notice] esms1: eslogin1 [ INSTALLING ] (nod e installer started) [esms1->device]% Wed Sep 19 15:16:38 2012 [notice] esms1: eslogin1 [ INSTALLER_CALLINGINIT ] (s witching to local root) [esms1->device]% Wed Sep 19 15:17:35 2012 [notice] esms1: eslogin1 [ UP ] [esms1->device]% status eslogin1 ..................[ UP ] esms1 .................... [ UP ] sw-1ge ................... [ DOWN ] health check failed Wed Sep 19 15:18:01 2012 [notice] eslogin1: Check 'DeviceIsUp' is in state PASS on eslogin1 [esms1->device]% pexec -n eslogin1 "df -h" [eslogin1] : Filesystem Size Used Avail Use% Mounted on /dev/sda5 767G 3.0G 725G 1% / devtmpfs 32G 192K 32G 1% /dev tmpfs 32G 0 32G 0% /dev/shm /dev/sda3 31G 659M 28G 3% /tmp /dev/sda2 61G 601M 57G 2% /var master:/cm/shared 2.6T 31G 2.5T 2% /cm/shared master:/home 177G 19G 150G 11% /home [esms1->device]% quit 15. Create new drbd partitions in the free space of /dev/sdb 132 S–2327–B CIMS Configuration Tasks [3] a. Start parted and list partitions on /dev/sdb. esms1:/ # parted /dev/sdb GNU Parted 2.3 Using /dev/sdb Welcome to GNU Parted! Type 'help' to view a list of commands. (parted) p Model: DELL PERC H710P (scsi) Disk /dev/sdb: 4723GB Sector size (logical/physical): 512B/512B Partition Table: gpt Number Start End Size File system Name Flags 1 17.4kB 1049GB 1049GB ext3 /var 2 1049GB 1059GB 10.2GB ext3 /var/lib/mysql/cmdaemon_mon 3 1059GB 3930GB 2871GB ext3 /cm (parted) b. Make a 20GB, /drbd1 partition with an ext2 file system. (parted) mkpart Partition name? []? /drbd1 File system type? [ext2]? Start? 3930GB End? 3950GB (parted) p Model: DELL PERC H710P (scsi) Disk /dev/sdb: 4723GB Sector size (logical/physical): 512B/512B Partition Table: gpt Number Start End Size File system Name Flags 1 17.4kB 1049GB 1049GB ext3 /var 2 1049GB 1059GB 10.2GB ext3 /var/lib/mysql/cmdaemon_mon 3 1059GB 3930GB 2871GB ext3 /cm 4 3930GB 3950GB 20.0GB /drbd1 c. Make a 1GB, /drbd2 partition. (parted) mkpart Partition name? []? /drbd2 File system type? [ext2]? Start? 3950GB End? 3951GB d. Make a 1GB, /drbd3 partition with an ext2 file system. (parted) mkpart Partition name? []? /drbd3 File system type? [ext2]? Start? 3951GB End? 3952GB S–2327–B 133 Data Management Platform (DMP) Administrator's Guide e. Make a 16GB, /drbd4 partition. (parted) mkpart Partition name? []? /drbd4 File system type? [ext2]? Start? 3952GB End? 3968GB f. Make a /drbd5 partition with the remaining space (in this example 755GB). (parted) mkpart Partition name? []? /drbd4 File system type? [ext2]? Start? 3968GB End? 4723GB (parted) quit Information: You may need to update /etc/fstab. 16. Show free disk blocks and files. esms1:/ # df -h Filesystem Size Used Avail Use% Mounted on /dev/sda5 177G 19G 150G 11% / devtmpfs 32G 264K 32G 1% /dev tmpfs 32G 0 32G 0% /dev/shm /dev/sda1 473M 22M 427M 5% /boot /dev/sda3 61G 180M 57G 1% /tmp /dev/sdb1 962G 349M 913G 1% /var /dev/sdb2 9.4G 156M 8.8G 2% /var/lib/mysql/cmdaemon_mon /dev/sdb3 2.6T 31G 2.5T 2% /cm 17. Make an ext3 file system on /dev/sdb4. esms1:/ # mkfs.ext3 /dev/sdb4 18. Make as ext3 file system on /dev/sdb5. esms1:/ # mkfs.ext3 /dev/sdb5 19. Make as ext3 file system on /dev/sdb6. esms1:/ # mkfs.ext3 /dev/sdb6 20. Make as ext3 file system on /dev/sdb7. esms1:/ # mkfs.ext3 /dev/sdb7 21. Make as ext3 file system on /dev/sdb8. esms1:/ # mkfs.ext3 /dev/sdb8 134 S–2327–B CIMS Configuration Tasks [3] 22. Verify the partitions are correct. esms1:/ # parted /dev/sdb GNU Parted 2.3 Using /dev/sdb Welcome to GNU Parted! Type 'help' to view a list of commands. (parted) p Model: DELL PERC H710P (scsi) Disk /dev/sdb: 4723GB Sector size (logical/physical): 512B/512B Partition Table: gpt Number Start End Size File system Name Flags 1 17.4kB 1049GB 1049GB ext3 /var 2 1049GB 1059GB 10.2GB ext3 /var/lib/mysql/cmdaemon_mon 3 1059GB 3930GB 2871GB ext3 /cm 4 3930GB 3950GB 20.0GB /drbd1 5 3950GB 3951GB 999MB /drbd2 6 3951GB 3952GB 1000MB /drbd3 7 3952GB 3968GB 16.0GB /drbd4 8 3968GB 4723GB 755GB /drbd5 (parted) quit 23. Create mount points and mount the new drbd partitions. esms1# mkdir -p /drbd1 /drbd2 /drbd3 /drbd4 /drbd5 esms1# mount /dev/sdb4 /drbd1 esms1# mount /dev/sdb5 /drbd2 esms1# mount /dev/sdb6 /drbd3 esms1# mount /dev/sdb7 /drbd4 esms1# mount /dev/sdb8 /drbd5 esms1:/ # df -h Filesystem Size Used Avail Use% Mounted on /dev/sda5 177G 19G 150G 11% / devtmpfs 32G 264K 32G 1% /dev tmpfs 32G 0 32G 0% /dev/shm /dev/sda1 473M 22M 427M 5% /boot /dev/sda3 61G 180M 57G 1% /tmp /dev/sdb1 962G 349M 913G 1% /var /dev/sdb2 9.4G 156M 8.8G 2% /var/lib/mysql/cmdaemon_mon /dev/sdb3 2.6T 31G 2.5T 2% /cm /dev/sdb4 19G 173M 18G 1% /drbd1 /dev/sdb5 938M 18M 874M 2% /drbd2 /dev/sdb6 939M 18M 875M 2% /drbd3 /dev/sdb7 15G 166M 14G 2% /drbd4 /dev/sdb8 693G 198M 658G 1% /drbd5 esms1:/ # exit S–2327–B 135 Data Management Platform (DMP) Administrator's Guide 3.25 Configuring DHCP to Allow Requests from Unknown Nodes By default, the CIMS node is configured to ignore PXE boot requests from nodes with unknown MAC addresses. If your CIMS administration network (esmaint-net) contains only DMP nodes, you can configure the Dynamic Host Configuration Protocol (DHCP) server to allow the CIMS to answer PXE boot requests from nodes with unknown MAC addresses. Note: Consult your site's security policy before you enable this feature. Procedure 38. Configuring DHCP to allow requests from unknown nodes 1. Edit /cm/local/apps/cmd/etc/cmd.conf. 2. Set LockDownDhcpd = false 3. Save your changes and exit. 3.26 Changing the CIMS Firewall Configuration The CIMS runs the Shorewall Firewall package. Configuration files are located in /etc/shorewall (such as the firewall rules in /etc/shorewall/rules). By default, the CIMS allows ICMP traffic from the external interface (site-admin-net) and SSH traffic over ports 22 and 8081. Note: Port 8081 (SSL) must be open to use the Bright Cluster Management GUI (cmgui) from an external server. Depending on site requirements, the default firewall settings may need to be modified. If changes are made to the Shorewall configuration, you must restart the shorewall service using the following command. esms1# /etc/init.d/shorewall restart 136 S–2327–B CDL Administration Tasks [4] 4.1 Creating a CDL Node in Bright Cluster Manager® The easiest way to add a new slave node is to clone an existing node that is configured and fully functional in Bright Cluster Manager® (Bright). When you do not have a functioning CDL node, you must clone the default node (node001) created during the CIMS installation process. Note: In this guide, some examples are left-justified to fit command lines or display output on a single line. Left-justification has no special significance. Procedure 39. Creating a CDL node in Bright Note: The following procedures use the Bright management shell (cmsh). They may also be performed using the Bright GUI (cmgui). The cmsh command prompt displays an asterisk (*) when you have uncommitted changes. Be sure to commit your changes using the commit command before exiting cmsh, or your changes will be lost. When using the cmgui, be sure to click the Save button as needed to save and commit your changes. 1. Log in to the CIMS as root and start cmsh. esms1# cmsh [esms1]% 2. Enter device mode. [esms1]% device [esms1->device]% 3. List the available devices. [esms1->device]% list Type Hostname (key) MAC Category Ip Network ---------------------- ---------------- ------------------ ---------- --------------- -------------- EthernetSwitch switch01 00:00:00:00:00:00 10.141.253.1 esmaint-net MasterNode esms1 78:2B:CB:40:CE:CA 10.141.255.254 esmaint-net PhysicalNode node001 00:00:00:00:00:00 default 10.141.0.1 esmaint-net S–2327–B 137 Data Management Platform (DMP) Administrator's Guide 4. Clone an existing node such as the default node. This example creates a new CDL node named eslogin1. [esms1->device]% clone node001 eslogin1 Base name mismatch, IP settings will not be modified! Note: When the CIMS is installed, Bright creates a default node, node001, which is placed in the default category and uses the default slave image (/cm/images/default-image). This image is assigned to the newly cloned node. Figure 43. Default Node in Bright Note: When repeating this procedure for additional CDL nodes, clone the first fully configured and functional CDL node (eslogin1) instead of the default node (node001). 5. Change the interface settings for the new (cloned) node. a. Switch to interfaces mode and list interfaces on eslogin1. [esms1->device*[eslogin1*]]% interfaces [esms1->device*[eslogin1*]->interfaces]% list Type Network device name IP Network ------------ -------------------- ---------------- ---------------- bmc ipmi0 10.148.0.1 ipmi-net physical BOOTIF [prov] 10.141.0.1 esmaint-net 138 S–2327–B CDL Administration Tasks [4] b. Set the BOOTIF and ipmi0 interface addresses for the new node. These addresses must be different from those used by the default or original node. [esms1->device*[eslogin1*]->interfaces]% set bootif ip 10.141.0.2 [esms1->device*[eslogin1*]->interfaces*]% set ipmi0 ip 10.148.0.2 [esms1->device*[eslogin1*]->interfaces*]% list Type Network device name IP Network ------------ -------------------- ---------------- ---------------- bmc ipmi0 10.148.0.2 ipmi-net physical BOOTIF [prov] 10.141.0.2 esmaint-net c. Set up the ib-net network and interface. [esms1->device*[eslogin1*]->interfaces*]% add physical ib0 [esms1->device*[eslogin1*]->interfaces*[ib0*]]% set network ib-net [esms1->device*[eslogin1*]->interfaces*[ib0*]]% set ip 10.149.0.2 [esms1->device*[eslogin1*]->interfaces[ib0]]% show Parameter Value ------------------------------ ------------------------------------------------ Additional Hostnames Card Type DHCP no IP 10.149.0.2 MAC 00:00:00:00:00:00 Network ib-net Network device name ib0 Revision Speed Type physical d. Commit your changes. [esms1->device*[eslogin1*]->interfaces*[ib0*]]% commit e. Exit ib0. [esms1->device[eslogin1]->interfaces[ib0]]% exit f. Check the results. [esms1->device[eslogin1]->interfaces]% list Type Network device name IP Network ------------ -------------------- ---------------- ---------------- bmc ipmi0 10.148.0.2 ipmi-net physical BOOTIF [prov] 10.141.0.2 esmaint-net physical ib0 10.149.0.2 ib-net g. Exit interface mode and return to device mode. [esms1->device[eslogin1]->interfaces]% exit h. Display the status of the new node. [esms1->device[eslogin1]]% status eslogin1 ............... [ DOWN ] (Unassigned) Note: The node is unassigned because the MAC address has not been set in Bright. S–2327–B 139 Data Management Platform (DMP) Administrator's Guide 6. Set the MAC address for the CDL node (eth0 on esmaint-net). [esms1->device[eslogin1]]% set mac MACaddress 7. Set the management network to esmaint-net. [esms1->device[eslogin1*]]% set managementnetwork esmaint-net 8. Commit your changes and exit cmsh. [esms1->device*[eslogin1*]]% commit [esms1->device[eslogin1]]% quit esms1# 4.2 Configuring Network Parameters for the Site User Network Before using a CDL node, you must configure the site network that is used by slave CDL nodes and set up the network parameters for each node. Procedure 40. Setting network parameters for site-user-net Note: This procedure uses the network name site-user-net. Substitute the name of your external user (site) network for user access. 1. Log in to the CIMS as root and run the cmsh command. esms1# cmsh [esms1]% 2. Switch to network mode: [esms1]% network [esms1->network]% 3. Display the existing networks. [esms1->network]% list Name (key) Type Netmask bits Base address Domain name IPv6 ---------------- ------------ ---------------- ---------------- -------------------- ---- esmaint-net Internal 16 10.141.0.0 esmaint-net.cluster no globalnet Global 16 0.0.0.0 cm.cluster no ib-net Internal 16 10.149.0.0 ib-net.cluster no ipmi-net Internal 16 10.148.0.0 ipmi.net.cluster no site-admin-net External 24 aaa.bbb.ccc.ddd your.domain.com no 4. Determine whether a site-user-net exists on the CIMS. a. If a site-user-net already exists on the CIMS, proceed to step 7. b. Create site-user-net by cloning the site-admin-net network. [esms1->network]% clone site-admin-net site-user-net 5. The cloned network inherits the same settings as the original network (site-admin-net). You must change several settings for site-user-net: base address, broadcast address, domain name, gateway, and (if necessary) netmask bits. 140 S–2327–B CDL Administration Tasks [4] a. Display the existing settings. [esms1->network*[site-user-net*]% show Parameter Value ------------------------------ ---------------------- Base address aaa.bbb.ccc.ddd Broadcast address aaa.bbb.255.255 Domain Name your.domain.com Dynamic range end 0.0.0.0 Dynamic range start 0.0.0.0 Gateway aaa.bbb.ccc.ddd IPv6 no Lock down dhcpd no MTU 1500 Management allowed no Netmask bits 24 Node booting no Notes Revision Type External name site-user-net b. Change the base address. [esms1->network*[site-user-net*]% set baseaddress site-user-netBaseAddress c. Change the broadcast address. [esms1->network*[site-user-net*]% set broadcastaddress site-user-netBroadcastAddress d. Change the domain name. [esms1->network*[site-user-net*]% set domainname site-user-netDomainName e. Change the gateway. [esms1->network*[site-user-net*]% set gateway site-user-netGateway f. If necessary, change the netmask bits. [esms1->network*[site-user-net*]% set netmaskbits NN 6. Commit your changes. [esms1->network*[site-user-net*]% commit 7. Switch to device mode. [esms1->network[site-user-net]% device 8. Add an interface to the site-user-net network. This example shows the host name eslogin1, the Ethernet port eth2, and the example IP address aaa.bbb.ccc.ddd. Substitute your CDL node's host name and IP address when configuring the eth2 interface. S–2327–B 141 Data Management Platform (DMP) Administrator's Guide Important: If necessary, specify eth0 instead of eth2. [esms1->device]% addinterface -n eslogin1 physical eth2 site-user-net aaa.bbb.ccc.ddd Note: You must repeat this step for each CDL node that is added to the site-user-net network. 9. Commit your changes. [esms1->device*[eslogin1*]]% commit 10. Show the interfaces on eslogin1. [esms1->device% use eslogin1 [esms1->device[eslogin1]]% interfaces; list Type Network device name IP Network ------------ -------------------- ---------------- ---------------- bmc ipmi0 10.148.0.37 ipmi-net physical BOOTIF [prov] 10.141.0.37 esmaint-net physical eth2 aaa.bbb.ccc.ddd site-user-net physical ib0 10.149.0.37 ib-net 11. Display the existing networks. [esms1->device[eslogin1]->interfaces]% network list Name (key) Type Netmask bits Base address Domain name IPv6 ---------------- ------------ ---------------- ---------------- -------------------- ---- esmaint-net Internal 16 10.141.0.0 esmaint-net.cluster no globalnet Global 16 0.0.0.0 cm.cluster no ib-net Internal 16 10.149.0.0 ib-net.cluster no ipmi-net Internal 16 10.148.0.0 ipmi-net.cluster no site-admin-net External 24 aaa.bbb.ccc.ddd your.domain.com no site-user-net External 24 aaa.bbb.ccc.ddd your.domain.com no [esms1]% 12. Exit cmsh. [esms1->device[eslogin1]->interfaces]% quit esms1# 4.3 Creating the Workload Manager Network (wlm-net) The workload manager network (wlm-net) connects the CDL nodes to a switch or gateway and then to the internal login nodes on the Cray system. Procedure 41. Creating the workload manager network (wlm-net) 1. Log in to the CIMS as root and run the cmsh command. esms1# cmsh [esms1]% 2. Switch to network mode: [esms1]% network [esms1->network]% 142 S–2327–B CDL Administration Tasks [4] 3. Display the existing networks. [esms1->network]% list Name (key) Type Netmask bits Base address Domain name IPv6 ---------------- ------------ ---------------- ---------------- -------------------- ---- esmaint-net Internal 16 10.141.0.0 esmaint-net.cluster no globalnet Global 16 0.0.0.0 cm.cluster no ib-net Internal 16 10.149.0.0 ib-net.cluster no ipmi-net Internal 16 10.148.0.0 ipmi-net.cluster no site-admin-net External 24 aaa.bbb.ccc.ddd your.domain.com no site-user-net External 24 aaa.bbb.ccc.ddd your.domain.com no [esms1->network]% 4. Create wlm-net by cloning the site-admin-net network. [esms1->network]% clone site-admin-net wlm-net [esms1->network*[wlm-net*]% 5. The cloned network inherits the same settings as the original network (site-admin-net). You must change the several settings for the new wlm-net: base address, broadcast address, domain name, gateway, and (if necessary) netmask bits. a. Display the existing settings. [esms1->network*[wlm-net*]% show Parameter Value ------------------------------ ---------------------- Base address aaa.bbb.ccc.ddd Broadcast address aaa.bbb.255.255 Domain Name your.domain.com Dynamic range end 0.0.0.0 Dynamic range start 0.0.0.0 Gateway aaa.bbb.ccc.ddd IPv6 no Lock down dhcpd no MTU 1500 Management allowed no Netmask bits 24 Node booting no Notes Revision Type External name wlm-net b. Change the base address. [esms1->network*[wlm-net*]% set baseaddress wlm-netBaseAddress c. Change the broadcast address. [esms1->network*[wlm-net*]% set broadcastaddress wlm-netBroadcastAddress d. Change the domain name. [esms1->network*[wlm-net*]% set domainname wlm-netDomainName e. Change the gateway. [esms1->network*[wlm-net*]% set gateway wlm-netGateway S–2327–B 143 Data Management Platform (DMP) Administrator's Guide f. If necessary, change the netmask bits. [esms1->network*[wlm-net*]% set netmaskbits NN 6. Commit your changes. [esms1->network*[wlm-net*]% commit [esms1->network[wlm-net]% 7. Verify the wlm-net network settings. [esms1->network[wlm-net]]% show Parameter Value ------------------------------ ------------------------------------ Base address 10.150.0.0 Broadcast address 10.150.255.255 Domain Name your.domain.com Dynamic range end 0.0.0.0 Dynamic range start 0.0.0.0 Gateway aaa.bbb.ccc.ddd IPv6 no Lock down dhcpd yes MTU 1500 Management allowed no Netmask bits 24 Node booting no Notes Revision Type External name wlm-net [esms1->network[wlm-net]]% 8. Switch to device mode. [esms1->network[wlm-net]% device [esms1->device]% 9. Add an interface to the wlm-net network. This example shows the host name eslogin1, the Ethernet port eth1, and the example IP address 10.150.0.1. Substitute your CDL node's host name and IP address when configuring the eth1 interface. Important: If necessary, specify eth3 instead of eth1. [esms1->device]% addinterface -n eslogin1 physical eth1 wlm-net 10.150.0.1 Note: You must repeat this step for each CDL node that is added to the wlm-net network. 10. Commit your changes. [esms1->device*[eslogin1*]% commit 144 S–2327–B CDL Administration Tasks [4] 11. Show the interfaces on eslogin1. [esms1->device[eslogin1]% interfaces; list Type Network device name IP Network ------------ -------------------- ---------------- ---------------- bmc ipmi0 10.148.0.37 ipmi-net physical BOOTIF [prov] 10.141.0.37 esmaint-net physical eth1 10.150.0.1 wlm-net physical eth2 aaa.bbb.ccc.ddd site-admin-net physical ib0 10.149.0.37 ib-net [esms1->device*[eslogin1]->interfaces]% 12. Display the existing networks. [esms1->device[eslogin1]->interfaces]% network list Name (key) Type Netmask bits Base address Domain name IPv6 ---------------- ------------ ---------------- ---------------- -------------------- ---- esmaint-net Internal 16 10.141.0.0 esmaint-net.cluster no globalnet Global 16 0.0.0.0 cm.cluster no ib-net Internal 16 10.149.0.0 ib-net.cluster no ipmi-net Internal 16 10.148.0.0 ipmi-net.cluster no site-admin-net External 24 aaa.bbb.0.0 your.domain.com no site-user-net External 24 aaa.bbb.0.0 your.domain.com no wlm-net External 24 10.150.0.1 your.domain.com no 13. Exit cmsh. [esms1->device[eslogin1]->interfaces]% quit esms1# 4.4 Configuring Bright Categories for the CDL Nodes A Bright category controls which software image is used for nodes in that category. During the installation, default categories for each type of login node are created, (esLogin-XE and esLogin-XC). Note: The default category was assigned to the first CDL node when it was cloned. The following procedure describes how to customize the esLogin-XC category (substitute your site CDL category). The default category for CDL nodes must have a software image and default gateway configured. Note: Optionally, you may want to modify the default disk setup XML file or the finalize script for your site configuration. Procedure 42. Configure category settings for the CDL image 1. Log into the CIMS as root. 2. Run the cmsh command. esms1# cmsh [esms1]% S–2327–B 145 Data Management Platform (DMP) Administrator's Guide 3. Switch to category mode and list categories and associated software images. [esms1]% category [esms1->category]% list Name (key) Software image ------------------------ ------------------------ default default-image esLogin-XC default-image esLogin-XE default-image 4. Use the esLogin-XC category (substitute your site CDL category). [esms1->category]% use esLogin-XC [esms1->category[esLogin-XC]]% 5. Set the default gateway for the esLogin-XC category. For gatewayIP, use the IP address of your site's gateway (usually on site-user-net). [esms1->category*[esLogin-XC*]]% set defaultgateway gatewayIP 6. Assign a CDL software image (imagename) to the esLogin-XC category. [esms1->category*[esLogin-XC*]]% set softwareimage imagename [esms1->category*[esLogin-XC*]]% commit Important: Files created or modified by a finalize script must be listed in the excludelistupdate exclude list for the category. Software updates will over write customized files if the files are not specified in an exclude list for the category. Customized files must also be specified in the excludelistgrab, and excludelistgrabnew exclude lists to prevent customized files from being copied to the CIMS. 7. (Optional) Change the Bright finalize script for this category. Bright runs the finalize script during node provisioning (node installation) just before turning control over to the software image initialization process. Note: The settings in: /opt/cray/esms/cray-es-finalize-scripts-XX/default/eslogin_finalize.sh were added to the esLogin-XC (or esLogin-XE) category by ESLinstall. Changes to eslogin_finalize.sh do not occur until they are saved in cmgui, or committed using cmsh, and the node is rebooted. a. Copy the eslogin_finalize.sh file. [esms1->category[esLogin-XC]]% quit esms1# cd /opt/cray/esms/cray-es-finalize-scripts-XX esms1# mkdir -p etc esms1# cp -p default/eslogin_finalize.sh etc/site.eslogin_finalize.sh 146 S–2327–B CDL Administration Tasks [4] b. (Optional) Edit the site.eslogin_finalize.sh file to make any adjustments. Note: A finalize script (run before init) is used to set a file configuration or to initialize special hardware, sometimes after a hardware check. It is run in order to make software or hardware work before, or during the later init stage of boot. Use a finalize script to execute commands before init, and the commands cannot be stored persistently anywhere else, or it is needed because a choice between (otherwise non-persistent) configuration files must be made based on the hardware before init starts. esms1# vi etc/site.eslogin_finalize.sh esms1# cmsh [esms1->category]% category use esLogin-XC [esms1->category[esLogin-XC]]% set finalizescript /opt/cray/esms/cray-es-finalize-scripts-XX /etc/site.eslogin_finalize.sh [esms1->category*[esLogin-XC*]]% commit c. Confirm the finalize script. [esms1->category[esLogin-XC]]% get finalizescript 8. (Optional) Change the disk partitions sizes for the CDL node. Note: The following partition sizes were added to the esLogin-XC (or esLogin-XE) category from /opt/cray/esms/cray-es-diskpartitions-XX /default/eslogin-diskfull.xml. Changes to eslogin-diskfull.xml do not occur until they are saved in cmgui, or committed using cmsh, and the node is rebooted. swap - 64 GB /tmp - 32 GB /var - 64 GB / - Remainder of disk a. Quit cmsh and copy XML configuration file. [esms1->category[esLogin-XC]]% quit esms1# cd /opt/cray/esms/cray-es-diskpartitions-XX esms1# mkdir -p etc esms1# cp -p default/eslogin-diskfull.xml etc/site.eslogin-diskfull.xml Important: If the default disk setup XML files are updated in a ESM release and the site disk setup XML files have been customized, system administrators must compare the newly released disk setup XML files with the current production disk setup XML files, and merge the changes manually. After the changes have been merged, you must load the updated disk setup file into the Bright database for the node category and reboot the nodes that use that category. S–2327–B 147 Data Management Platform (DMP) Administrator's Guide b. Edit the site.eslogin-diskfull.xml file to change partition sizes or configuration. esms1# vi etc/site.eslogin-diskfull.xml esms1# cmsh [esms1->category]% category use esLogin-XC [esms1->category*[esLogin-XE*]]% set disksetup /opt/cray/esms/cray-es-diskpartitions-XX/etc/ site.eslogin-diskfull.xml esms1->category[esLogin-XC]]% commit c. Confirm the disk setup is loaded. [esms1->category[esLogin-XC]]% get disksetup 4.5 Configuring Kdump on CDL Nodes (SLES) Kdump is configured on a DMP system by modifying the configuration files on CDL systems. Dump files from slave nodes are stored either on the CIMS using NFS®, or on the slave node local disk. To save dump files to a local disk on a slave node, create a persistent /var/crash partition. Procedure 43. Configuring kdump on CDL nodes (SLES) 1. Log in to the CIMS as root. 2. Choose a slave node that you can use to test the kdump procedure (in this example eslogin1) and clone that slave node's software image. This example clones ESL-XE-2.1.0-201309252116 to ESL-XE-2.1.0-kdump. esms1# cp -pr /cm/images/ESL-XE-2.1.0-201309252116 /cm/images/ESL-XE-2.1.0-kdump esms1# cmsh esms1% softwareimage [esms1->softwareimage]% clone ESL-XE-2.1.0-201309252116 ESL-XE-2.1.0-kdump [esms1->softwareimage*[ESL-XE-2.1.0-kdump*]]% 3. Commit your changes. [esms1->softwareimage*[ESL-XE-2.1.0-kdump*]]% commit 4. Create a test category to configure kdump Switch to category mode to create a test category. [esms1->->softwareimage[ESL-XE-2.1.0-kdump]]% category [esms1->category]% 5. Clone an existing CDL category to create a test category. Note: Be sure to clone the functioning esLogin-XE or esLogin-XC category to an esLogin-XE-test or esLogin-XC-test category. These categories have different configurations and software images and are not interchangeable. This procedure creates an esLogin-XE-test category. [esms1->category]% clone esLogin-XE esLogin-XE-test [esms1->category*[esLogin-XE-test*]% 148 S–2327–B CDL Administration Tasks [4] 6. Assign the kdump CDL image (ESL-XE-2.1.0-kdump) to the test category. [esms1->category*[esLogin-XE-test]*]% set softwareimage ESL-XE-2.1.0-kdump 7. Add /var/crash to the exclude lists for the esLogin-XC-testcategory. The vi editor launches which enables you to edit each of the exclude list files. Add - /var/crash/* to the following exclude lists: [esms1->category*[esLogin-XE-test]*]% set excludelistsyncinstall [esms1->category*[esLogin-XE-test]*]% set excludelistupdate [esms1->category*[esLogin-XE-test]*]% set excludelistgrab [esms1->category*[esLogin-XE-test]*]% set excludelistgrabnew 8. Save each file and commit your changes. [esms1->category*[esLogin-XE-test*]% commit [esms1->category[esLogin-XE-test]% 9. Assign the test category and test image to the test node (eslogin1) and commit your changes. [esms1->category[esLogin-XE-test]]% device use eslogin1 [esms1->device[eslogin1]]% set category esLogin-XE-test [esms1->device*[eslogin1*]]% commit [esms1->device[eslogin1]]% 10. If you are saving kdump crash files to the slave node local disk, add the following lines to the slave node's finalize script. This command opens the vi editor. [esms1->device[eslogin1]]% set finalizescript DEV=$( awk -- '{ if ($2 == "/localdisk/var/crash") { print $1; exit 0 } }' < /proc/mounts ) [ -n "$DEV" ] && e2label $DEV crash 11. Commit your changes. [esms1->device*]]% commit 12. Set the storage location for crash dumps. • If crash dumps will be saved to the CIMS proceed to step 13. • If crash dumps will be saved to the slave node's local disk, proceed to step 14. 13. Save crash files to /var/crash on the primary CIMS S–2327–B 149 Data Management Platform (DMP) Administrator's Guide a. Use fsexports to determine whether the CIMS is exporting /var/crash. [esms1->device]]% use esms1 [esms1->device[esms1]]% fsexports [esms1>device[esms1]->fsexports]% list Name (key) Path -------------------------------------------- ------------------------------- /cm/shared@esmaint-net /cm/shared /home@esmaint-net /home /var/spool/burn@esmaint-net /var/spool/burn /cm/node-installer/certificates@esmaint-net /cm/node-installer/certificates /cm/node-installer@esmaint-net /cm/node-installer /var/crash@esmaint-net /var/crash b. Export /var/crash from the CIMS and configure it so that slave nodes can access it. [esms1->device[esms1]->fsexports]% add /var/crash [esms1->device*[esms1*]->fsexports*[/var/crash*]]% set name /var/crash@esmaint.net [esms1->device*[esms1*]->fsexports*[/var/crash*]]% set extraoptions no_subtree_check [esms1->device*[esms1*]->fsexports*[/var/crash*]]% set hosts esmaint-net [esms1->device[esms1]*]->fsexports*[/var/crash*]]% set write yes [esms1->device*[esms1*]->fsexports*[/var/crash*]]% commit c. Exit cmsh. [esms1->device[esms1]->fsexports[/var/crash]]% quit esms1# d. Update the exports. esms1# exportfs -a 14. Use the chroot shell, edit the /boot/pxelinux.cfg/default file for kdump test image created in step 2 (ESL-XE-2.1.0-kdump). a. Quit cmsh. [esms1->device[esms1]]% quit b. Use the chroot shell to edit the kdump test image. esms1# chroot /cm/images/ESL-XE-2.1.0-kdump esms:/>vi /boot/pxelinux.cfg/default c. Edit /boot/pxelinux.cfg/default file. d. Scroll down and locate the following line: # End of documentation, configuration follows: 150 S–2327–B CDL Administration Tasks [4] e. Enter the following lines in the default configuration file: LABEL kdump KERNEL vmlinuz IPAPPEND 3 APPEND initrd=initrd crashkernel=512M CMDS console=tty0 console=ttyS1,115200n8 CMDE MENU LABEL ^KDUMP - Normal boot mode with kdump MENU DEFAULT f. Examine the other LABEL entries in the default configuration file and remove the line: MENU DEFAULT. g. Exit and save the file. h. Verify that the /var/crash partition exists in the ESL-XE-2.1.0-kdump image. 15. Edit the /etc/sysconfig/kdump file and modify the following lines: esms1:/> vi /etc/sysconfig/kdump a. Scroll down and locate the KDUMP_SAVEDIR entry: If you want to save crash dump files using NFS to /var/crash on the CIMS, modify the KDUMP_SAVEDIR line as follows: KDUMP_SAVEDIR="nfs://master/var/crash/" If you want to save crash dump files to the local disk, modify KDUMP_SAVEDIR as follows: KDUMP_SAVEDIR="file:///var/crash" Note: Create a persistent partition (/var/crash) in the disk setup XML file for the kdump test category (esLogin-XE-test). Creating a separate partition for crash dumps on the slave node software image prevents /var from filling up and causing problems for the operating system. b. Locate KDUMP_DUMPLEVEL and change it to: KDUMP_DUMPLEVEL=27 c. Locate KDUMP_CONTINUE_ON_ERROR and change it to: KDUMP_CONTINUE_ON_ERROR="true" d. Locate KDUMP_NETCONFIG and change it to you esmaint-net interface, eth0 or eth2): KDUMP_NETCONFIG="eth0:dhcp" e. Exit and save the file. 16. Enable the kdump service. esms1:/> chkconfig --set boot.kdump on S–2327–B 151 Data Management Platform (DMP) Administrator's Guide 17. Verify that /lib/mkinitrd/scripts/setup-storage.sh and setup-kdumpfs.sh exist. a. If these files do not exist, copy them from the CIMS to /lib/mkinitrd/scripts/ in the /cm/images/ESL-XE-2.1.0-kdump image. You must exit the chroot shell to do this step. esms1:/> exit esms1# cp -p /lib/mkinitrd/scripts/setup-storage.sh /cm/images/ESL-XE-2.1.0-kdump/lib/mkinitrd/scripts esms1# cp -p /lib/mkinitrd/scripts/setup-kdumpfs.sh /cm/images/ESL-XE-2.1.0-kdump/lib/mkinitrd/scripts b. Return the chroot shell in the ESL image. esms1# chroot /cm/images/ESL-XE-2.1.0-kdump c. Run mkinitrd_setup to update symbolic links in /lib/mkinitrd/setup and /lib/mkinitrd/boot. esms1:/> mkinitrd_setup Scanning scripts ... Resolve dependencies ... Install symlinks in /lib/mkinitrd/setup ... Install symlinks in /lib/mkinitrd/boot ... esms1:/lib/mkinitrd/scripts> 18. Exit the chroot shell. esms1:/> exit esms1# 19. Reboot the test node and run kdump a. Start a console window on the test slave node (eslogin1). esms1# cmsh [esms1]% device; use eslogin1 [esms1->device[eslogin1]]% rconsole b. Reboot the test node (eslogin1). eslogin1: reboot eslogin1: Reboot in progress ... c. When the node reboots, initiate kdump. esms1# ssh eslogin1 eslogin1# echo c > /proc/sysrq-trigger Note: If dumping over NFS, the dump file should be created in /var/crash on the CIMS node. If dumping to the slave node's local disk, the dump file should be created in /var/crash on the slave node's local disk. 20. Make the kdump image the default image for all CDL nodes. 152 S–2327–B CDL Administration Tasks [4] a. Start cmsh and assign the test node (eslogin1) to the default CDL category esLogin-XE. esms1# cmsh [esms1]% device; use eslogin1 [esms1->device[eslogin1]]% set category esLogin-XE [esms1->device*[eslogin1*]]% commit b. Switch to category mode and configure the default esLogin-XE category to use the kdump software image. [esms1->device[eslogin1]]% category [esms1->category]% use esLogin-XE [esms1->category[esLogin-XE]% set softwareimage ESL-XE-2.1.0-kdump 21. Reboot all of the nodes in the esLogin-XE category, so that they use the kdump software image. [esms1->category[esLogin-XE]% device [esms1->device]% reboot -c esLogin-XE eslogin001: Reboot in progress ... 22. Exit cmsh. [esms1->device]% quit 4.6 Creating a kdump Crash Partition on a Slave Node Local Disk If the slave node does not already have a /var/crash partition, then you must create a persistent partition (/var/crash) on a slave node's disk setup to store kdump crash files. Procedure 44. Create a kdump /var/crash partition on a slave node 1. Log in to the CIMS as root and run the cmsh command. esms1# cmsh [esms1]% 2. Clone the working category used or the slave node to create a test category. This procedure creates a category named esLogin-XE-test. esms1# category [esms1->category]% clone esLogin-XE esLogin-XE-test [esms1->category*[esLogin-XE-test*]]% 3. Edit the disk setup XML file for the test category. [esms1->category*[esLogin-XE-test*]]% set disksetup S–2327–B 153 Data Management Platform (DMP) Administrator's Guide The vi editor starts which enables you to edit the setup XML file. Scroll down in the disk setup XML file and create a new /var/crash disk partition. ... 10G linux ext3 /var/crash defaults,noatime,nodiratime ... 4. Commit your changes. [esms1->category*[esLogin-XE-test*]]% commit [esms1->category[esLogin-XE-test]]% 5. Reboot the slave node (in this example eslogin1) to re-partition the disk. Note: When rebooting a system, it is helpful to start a remote console shell. In a new window log in to the CIMS as root, start cmsh, and launch a remote console on eslogin1 while you reboot. esms1# cmsh [esms1]% device; use eslogin1 [esms1->device[eslogin1]]% rconsole In a another window, reboot the slave node. The node installer will recognize the disk setup has changed, and re-partition the disks. [esms1->category[esLogin-XE-test]]% device; use eslogin1 [esms1->device[eslogin1]]% reboot 6. Enter quit to exit cmsh. 7. When the node reboots, SSH to the node and verify that the /var/crash partition was created. esms1# ssh eslogin1 Last login: Thu May 23 08:09:22 2013 eslogin1# df Filesystem 1K-blocks Used Available Use% Mounted on /dev/sda1 19685912 9735588 8950324 53% / udev 65879960 200 65879760 1% /dev tmpfs 65879960 0 65879960 0% /dev/shm /dev/sda7 189409992 191948 179596496 1% /local /dev/sda3 1969424 35784 1833596 2% /tmp /dev/sda2 19685656 654584 18031088 4% /var /dev/sda6 17718140 176196 16641900 2% /var/crash master:/cm/shared 125991776 60142976 59448768 51% /cm/shared master:/home 217873344 56468864 150337120 28% /home 154 S–2327–B CDL Administration Tasks [4] 4.7 Seting CDL Device Parameters When you add a new CDL node to Bright, it must be assigned to a Bright category (esLogin-XE or esLogin-XC, for example) that configures node-specific device information. Note: The physical node and network interface information should already exists in Bright. Procedure 45. Setting CDL slave node parameters in Bright 1. Log in to the CIMS as root and run the cmsh command. esms1# cmsh [esms1]% 2. Switch to device mode: [esms1]% device [esms1->device]% 3. Add the new node to the esLogin-XC category. Note: This procedure uses the example host name eslogin1 and the category name esLogin-XC. Substitute your actual CDL host name and the category name used in the previous procedure. [esms1->device]% use eslogin1 [esms1->device[eslogin1]]% set category esLogin-XE [esms1->device[eslogin1*]]% commit [esms1->device[eslogin1]]% 4. Set the rack information (device height, position in the rack, and rack index). This example assumes that the device height is 2U, its position in the rack is 0, and the rack index is 0. Substitute the actual values for your CDL node. [esms1->device[eslogin1]]% set deviceheight 2 [esms1->device[eslogin1*]]% set deviceposition 0 [esms1->device[eslogin1*]]% set rack 1 5. Commit your device changes. [esms1->device[eslogin1*]]% commit [esms1->device[eslogin1]]% 6. Exit cmsh. [esms1->device[eslogin1]]% quit esms1# S–2327–B 155 Data Management Platform (DMP) Administrator's Guide 4.8 Configuring eswrap The eswrap utility is a wrapper that lets users access a subset of Cray Linux Environment (CLE) and Programming Environment (PE) commands from a CDL node. eswrap uses Secure Shell (ssh) to launch the wrapped command on the Cray system, then displays the output on the CDL node so that it appears to the user that the wrapped command is actually running on the CDL node. Running eswrap creates a symbolic link for each wrapped command in the directory /opt/cray/eslogin/eswrap/default/bin. Each symbolic link points to the eswrap command, so that running a wrapped command (such as apstat) actually runs eswrap with the wrapped command as an argument. eswrap uses ssh to run the command on the specified node of the Cray system (by default, the internal login node with the host name login, unless the $ESWRAP_LOGIN environment variable specifies a different node). The initialization file eswrap.ini controls the list of wrapped commands. An administrator can edit this file to include or exclude SLURM commands, Application Level Placement Scheduler (ALPS) commands, STAT (Stack Trace Analysis Tool) commands, Cluster Compatibility Mode (CCM) commands, aprun, qsub, and selected Cray CLE commands. Note: See the eswrap(8) man page for more information on the eswrap configuration variables, including which commands are enabled by default. Use the following commands to display the image-specific version of this CDL man page: esms1# cd /cm/images/imagename/opt/cray/eslogin/eswrap/default/man/man8 esms1# man ./eswrap.8 Procedure 46. Configuring eswrap Note: You must execute the following steps as root whenever the eswrap.ini is modified. 1. Use chroot to change to the root directory of the CDL image. For imagename, substitute the directory name of the CDL image. esms1# chroot /cm/images/imagename esms1:/> 2. Edit the eswrap.ini configuration file to enable (wrap) the necessary commands for your environment. Refer to Support for Native SLURM on page 157 for information about support for native SLURM. esms1:/> vi /opt/cray/eslogin/eswrap/eswrap.ini 3. Run the eswrap command. esms1:/> /opt/cray/eslogin/eswrap/default/bin/eswrap --install 156 S–2327–B CDL Administration Tasks [4] 4. Exit the chroot environment. esms1:/> exit esms1# Note: Use this procedure to rerun eswrap --install to update the links for the wrapped commands after changes are made to the eswrap.ini file. 4.8.1 Support for Native SLURM The ESL software now supports the Simple Linux™ Utility for Resource Management (SLURM) by wrapping related commands with the native SLURM architect. If any of these commands are installed on the CDL node, they are moved out of the way before they are wrapped to prevent them from appearing in any paths. Important: to support native SLURM, the latest version of eswrap no longer performs the eswrap --install automatically. This is because the path environment is not valid during the RPM install but is outside the RPM scripts. You must run /opt/cray/eslogin/eswrap/default/bin/eswrap --install from a chroot shell in the ESL image after verifying that /opt/cray/eslogin/eswrap/eswrap.ini is correct in the image. A new /opt/cray/eslogin/eswrap/default/etc/eswrap.ini file includes a slurm option. You must merge this file with /opt/cray/eslogin/eswrap/default/eswrap.ini. If slurm=true is set in the eswrap.ini file, the following commands are wrapped when eswrap --install runs. If slurm=false is set in the eswrap.ini file, the following commands are unwrapped and any commands that were moved out of the way are restored when eswrap --install runs: salloc scontrol smap sshare sattach diag sprio sstat sbatch sinfo squeue strigger sacct sbcast sjobexitmod sreport sview sacctmgr scancel sjstat srun 4.9 Configuring a CDL Node to Mount an External Lustre® File System This section describes how to configure a CDL node to mount a Lustre file system. S–2327–B 157 Data Management Platform (DMP) Administrator's Guide Procedure 47. Configuring a CDL node to mount an external Lustre file system Note: This procedure modifies the software image imagename, to mount a Lustre file system which is mounted as /scratch on the Lustre MDS server at IP address 10.149.0.1 to the local mount point on the CDL of /lus/scratch. 1. Make a mount point in the CDL software image. esms1# chroot /cm/images/imagename esms1:/> mkdir -p /lus/scratch esms1:/> exit 2. Use the cmsh imageupdate command to update what is on the running node from imagename. The command performs a dry run, then use -w to perform the actual update. esms1# cmsh [esms1]% device use eslogin1 [esms1->device[eslogin1]]% imageupdate Performing dry run (use synclog command to review result, then pass -w to perform real update)... [esms1->device[eslogin1]]% imageupdate -w Mon Mar 4 14:08:26 2013 [notice] esms1: Provisioning started: sending esms1:/cm/images/imagename to eslogin1:/, mode UPDATE, dry run = no 3. Start a remote console on eslogin1. [esms1->device[eslogin1]]% rconsole 4. Start a new cmsh session and reboot eslogin1 to add the new mount point. esms1# cmsh [esms1]% device use eslogin1 [esms1->device[eslogin1]]% reboot 5. After eslogin1 reboots, login as root and enter the following commands to mount the file system: esms1# ssh root@eslogin1 eslogin1# mount -t lustre -o rw,flock,lazystatfs 10.149.0.1@o2ib:/scratch /lus/scratch 6. Verify file system mounted. eslogin1# mount | grep lustre 10.149.0.1@o2ib:/scratch on /lus/scratch type lustre (rw,flock,lazystatfs) eslogin1# df /lus/scratch Filesystem 1K-blocks Used Available Use% Mounted on 10.149.0.1@o2ib:/scratch 15611116960 922104 14829205576 1% /lus/scratch 158 S–2327–B CDL Administration Tasks [4] 4.10 Mounting a Lustre® File System on a CDL Node Procedure 48. Mounting a Lustre file system on a CDL node 1. Use the chroot shell to make a mount point in the CDL software image (in this example ESL-XC-1.3.0-201309252103). esms1# chroot /cm/images/ESL-XC-1.3.0-201309252103 esms1:/> mkdir -p /lus/scratch esms1:/> exit 2. Update the CDL nodes with the modified ESL-XC-1.3.0-201309252103 image. a. Start the cmsh. esms1# cmsh b. Go into device mode. [esms1]% device c. Update the image for the CDL node (in this example, eslogin1). [esms1->device]% imageupdate -n eslogin1 Performing dry run (use synclog command to review result, then pass -w to perform real update)... [esms1->device]% ... [esms1->device]% imageupdate -w -n eslogin1 d. Exit cmsh. [esms1->device]% quit 3. Use SSH to log in to the CDL node (eslogin1). esms1# ssh eslogin1 Last login: Fri Apr 15 11:28:06 2013 from esms1.cm.cluster eslogin1# 4. Mount the file system on the CDL node (eslogin1). Note: Add the following mount command and file system information to /etc/fstab . 10.149.0.2 is the IP address of the esfs-mds001 node on ib-net. eslogin1# mount -t lustre -o rw,flock,lazystatfs 10.149.0.2@o2ib:/scratch /lus/scratch 5. Verify the scratch file system is mounted. eslogin1# mount | grep lustre 10.149.0.1@o2ib:/scratch on /lus/scratch type lustre (rw,flock,lazystatfs) eslogin1# df /lus/scratch Filesystem 1K-blocks Used Available Use% Mounted on 10.149.0.1@o2ib:/scratch 15611116960 922104 14829205576 1% /lus/scratch S–2327–B 159 Data Management Platform (DMP) Administrator's Guide 4.11 Setting Up Exclude Lists Exclude lists may be configured for the esLogin-XC, esLogin-XE, esFS-MDS, or esFS-OSS categories. Files must be synchronized by pushing them from software image on the CIMS node to the slave node, or pulling them from the slave node to the CIMS node. There are three exclude lists for pushing from the CIMS node to the slave node which are excludelistfullinstall, excludelistsyncinstall, and excludelistupdate. These lists contains files that will not be pushed to the slave node during software image installation. Two exclude lists control how files are pulled from the slave node to the CIMS software image which are exludelistgrab, excludelistgrabnew. 160 S–2327–B CDL Administration Tasks [4] Important: Files created or modified by a finalize script must be listed in the excludelistupdate exclude list for the category. Software updates will over write customized files if the files are not specified in an exclude list for the category. Customized files must also be specified in the excludelistgrab, and excludelistgrabnew exclude lists to prevent customized files from being copied to the CIMS. excludelistfullinstall — Push During a full software image installation at boot time, all files from the software image in the CIMS are pushed to the slave node unless they are included in the excludelistfullinstall exclude list. excludelistsyncinstall — Push During a sync software image installation at boot time, all files from the software image on the CIMS are pushed to the slave node unless they are entered in the excludelistsyncinstall exclude list. excludelistupdate — Push If the node is already booted, running the cmsh imageupdate command will push all files from the slave node software image on the CIMS to the software image on the running node, except those entered on the excludelistupdate exclude list. excludelistgrab — Pull Using the cmsh grabimage command, or cmgui Grab to Image button synchronizes files from the slave node to its existing software image, unless the files or directories are entered in the excludelistgrab exclude list. excludelistgrabnew — Pull Running the cmsh grabimage -n newimage command synchronizes files from the slave node to a new software image, unless the files or directories are entered in the excludelistgrabnew exclude list. Important: Be sure to configure the excludelistgrab and excludelistgrabnew exclude lists to exclude all network file systems such as NFS®, Lustre®, or GPFS™ file systems. S–2327–B 161 Data Management Platform (DMP) Administrator's Guide Figure 44 shows the category exclude lists under the cmgui Node Category->Settings tab. Figure 44. Setting up Exclude Lists in cmgui 4.11.1 Check Exclude Lists Each of the exclude lists has specific comments about the pre-configured exclusions. To check what is currently in one of the pre-configured exclude lists, run the following cmsh command or use the cmgui to select a node category from the resource tree, then select the Settings tab. esms1# cmsh -c "category use esLogin-XC; get excludelistfullinstall" # For details on the exclude patterns defined here please refer to # the FILTER RULES section of the rsync man page. # # Files that match these patterns will not be installed onto the node. - lost+found/ - /proc/* - /sys/* 4.11.2 Changing Exclude Lists To change an exclude list, run this command. It will open an editor so you can make changes to the list. esms1# cmsh -c "category use esLogin-XC; set excludelistfullinstall; commit" 162 S–2327–B CDL Administration Tasks [4] 4.11.3 Exclude User Home Directories If user home directories in /home/users are mounted from an NFS server, then add /home/users/* to excludelistfullinstall, excludelistsyncinstall, and excludelistupdate, to prevent those directories from being removed when synchronizing from a software image to an CDL node. If user home directories in /home/users are mounted from an NFS server and also mount a Lustre file system in /lus/scratch, then add /home/users/* and /lus/scratch to the excludelistgrab and excludelistgrabnew exclude lists. This prevents a grabimage command from copying all of the files from a remote file server to the software image. /home/users/* /lus/scratch S–2327–B 163 Data Management Platform (DMP) Administrator's Guide 164 S–2327–B CLFS Administration Tasks [5] 5.1 Recommended MDS/MGT Volume Size Cray recommends an MGT volume size of 100MB. 5.2 Configuring CLFS Failover (esfsmon) The Lustre® file system uses multiple Cray CLFS server nodes to supply data storage services to provide a unified view of a cluster file system. As such, there are multiple points of failure. However, there are also multiple routes for data flow and multiple services that can respond to I/O service requests in the event of failure. As file system infrastructures become very large and complex, failures are inevitable. Failures are detected by hardware, software, or firmware within components or system that comprise the Lustre file system. Generally, a good failover strategy is based on the expectation that failures are spread over time fairly evenly and that there is a reasonable amount of time for human intervention to correct failures (e.g., hard drive failures and replacements). Cray's failover strategy is that the normal mode of operations is to have the whole system functional and running without on-going errors. When a failure occurs, the automated failover mechanisms move operations in the failed path to other functional resources and provide notification to the system administrator. After the system is running in a failed over state, subsequent failures are detected and additional notifications are sent, however, additional failover operations are not executed without human intervention. When multiple failures occur in short periods of time, the primary cause may be difficult to identify without significant investigation. Continued automatic failover actions may lead to catastrophic file system failures. 5.2.1 Storage Configuration Overview Multiple Lustre file systems are supported on external storage nodes. The resiliency features in CLFS node failover ensure that under normal conditions, failures will not result in the loss of access to the file system or the loss of data. Those features, described below in detail, include: • OSS path redundancy to the storage arrays via failover driver software • Rebuilding RAID after a disk failure S–2327–B 165 Data Management Platform (DMP) Administrator's Guide • Failing over OSTs when an OSS suffers various types of failures • MDT failover when an MDS suffers a failure • Loss of an LNET router network connection 5.2.2 Failover Conditions The following conditions trigger a failover action. After a failure monitor-only mode (RUNSAFE) is activated. • Power failure of a node • Failure to TCP ping the node • Failure of a node to LNET ping at least one other node • Failure of a node to have the expected complement of mounts available 5.2.3 Failover Functional Tests Functional tests are performed on all nodes by category. See esfsmon healthcheck on page 168 for more information about node categories. • All nodes in a particular category are tested in parallel – Failed nodes are not tested – Some tests do not apply to the standby MDS node unless it is acting as the primary • Test descriptions – DRAC function Verify connectivity with the DRAC by checking power status of each node. – IP connectivity with CIMS Ping each node from the CIMS over the management network. – LNET ping MDS nodes will attempt to LNET ping two OSS nodes. OSS nodes will attempt to ping two MDS nodes, if available, or the MDS and an OSS node. Failure of both ping attempts is a failure. Status of the IB port is also checked. – Lustre Mounts Check that the appropriate lustre mounts are mounted. 5.2.4 HA/Failover of I/O Paths for Servers Connected to Storage Arrays The driver for the RAID arrays supports path failover when redundant paths between an OSS and its attached array fails. When the path failure has been corrected, failback is automatic. 166 S–2327–B CLFS Administration Tasks [5] 5.2.5 Disk Failure and Rebuild: RAID Hardware Capability Disk failure detection and automatic rebuild of the replaced drive are features of the RAID array in each LUN. The only manual portion of the repair process is the physical replacement of the failed disk drive. 5.2.6 Failover Features and Bright Monitoring Bright software provides a monitoring framework that enables administrators to: • Inspect monitoring data to the required level for existing resources • Configure gathering of monitoring data for new resources • See current and past problems or abnormal behavior • Notice trends that help the administrator predict likely future problems • Manage current and likely future problems by triggering alerts, taking necessary actions to improve the situation, or investigate further CLFS failover under Bright Cluster Manager® (Bright) consists of five parts: a monitor script, an action script, a failback command, a configuration file, and the lustre_control utility. • esfsmon_healthcheck is the monitoring script which resides in /cm/local/apps/cmd/scripts/healthchecks on the CIMS node. The esfsmon_healthcheck monitoring script is configured as the esfsmon monitoring healthcheck in Bright. A separate esfsmon monitor process runs for each Lustre file system. See Configuring the esfsmon healthcheck Monitor on page 169 for more information about esfsmon_healthcheck monitoring. The monitoring process runs on the CIMS as a Bright healthcheck every 120 seconds by default and can be reconfigured. See Tuning the esfsmon:filesystem Check Interval on page 177 for more information about changing the healthcheck monitoring interval. The esfsmon_healthcheck monitoring script executes commands on the CIMS and all monitored nodes via the Bright daemon (CMDaemon) to assess the ability of each node to serve the Lustre file system. Status messages are sent to both the Bright software and /var/log/messages on the CIMS. • esfsmon_action is the action script which resides in /cm/local/apps/cmd/scripts/action on the CIMS and is configured as the esfsmon_action monitoring action in Bright.esfsmon_action runs on the CIMS as a Bright healthcheck action when the esfsmon healthcheck reports a failure, and takes appropriate failover action if the esfsmon healthcheck is not in RUNSAFE mode. esfsmon_action executes lustre_control failover on the CIMS to affect MDS or OSS failover. Status messages are sent to both the Bright software and /var/log/messages. S–2327–B 167 Data Management Platform (DMP) Administrator's Guide • esfsmon_failback is the failback command which runs on the CIMS to bring previously failed CLFS server nodes back to service after any faults have been corrected. • esfsmon.conf is the configuration file that contains environmental variables and file system definitions used by esfsmon. • lustre_control runs on the CIMS to perform the failover and failback of the Lustre assets. 5.2.6.1 esfsmon healthcheck The testing sequence of the esfsmon healthcheck monitor is as follows: • Validate configuration – Need at least one MDS and two OSS nodes for failover to be active. – No monitoring will be performed until the esfsmon healthcheck discovers at least one active MDS and two active OSS nodes. – Check to see if monitoring should be suspended. Existence of /var/esfsmon/esfsmon_suspend_filesystem file indicates a suspended mode. This is entered automatically during the failover process to prevent false positive error indications. • Check for degraded nodes – Degraded nodes indicate that a failover action has taken place but has not been resolved so monitor-only mode is used. If there are any nodes in the esfs-mds-failed-filesystem or esfs-oss-failed-filesystem categories, degraded nodes exist. – Continuing to failover nodes may cascade into a catastrophic configuration where the CLFS becomes unusable. Failed nodes should be addressed and returned to service as soon as possible. • Functional testing by node category. Nodes must be in the Bright categories shown below: – esfs-mds-filesystem All MDS nodes in filesystem – esfs-mds-fo-filesystem All Standby MDS nodes in filesystem – esfs-oss-even-filesystem All even numbered OSS nodes in filesystem – esfs-oss-odd-filesystem All odd numbered OSS nodes in filesystem – esfs-mds-failed-filesystem All MDS nodes that have failed over in filesystem – esfs-oss-failed-filesystem All OSS nodes that have failed over in filesystem 168 S–2327–B CLFS Administration Tasks [5] 5.2.7 Configuring the esfsmon healthcheck Monitor The esfsmon healthcheck monitor is installed as a master node healthcheck in Bright by Cray. It is available to monitor any Lustre file system whose MDS and OSS servers are managed by a CIMS server running Bright. The following procedures describe how to configure esfsmon to monitor a Lustre file system and how to view the esfsmon configuration. 5.2.7.1 Installing esfsmon_healthcheck and esfsmon_action Scripts Before esfsmon can be configured and used, the esfsmon_healthcheck and esfsmon_action scripts must be installed into Bright Cluster Manager. The following procedure installs the esfsmon_healthcheck and esfsmon_action scripts in Bright Cluster Manager. This must be done only once, regardless of the number of Lustre file systems that will be monitored. Procedure 49. Installing esfsmon_healthcheck and esfsmon_action 1. Log in to the CIMS as root and startcmsh. esms1# cmsh [esms1]% 2. Switch to monitoring mode and add esfsmon. [esms1]% monitoring healthchecks [esms1->monitoring->healthchecks]% add esfsmon [esms1->monitoring->healthchecks*[esfsmon*]]% show Parameter Value ------------------------------ --------------- Class of healthcheck misc Command Description Disabled no Extended environment no Name esfsmon Notes Only when idle no Parameter permissions optional Revision Sampling method samplingonnode State flapping count 7 Timeout 5 Valid for node,headnode S–2327–B 169 Data Management Platform (DMP) Administrator's Guide 3. Set the command. [esms1->monitoring->healthchecks]% set command /cm/local/apps/cmd/scripts/healthchecks/esfsmon_healthcheck [esms1->monitoring->healthchecks*[esfsmon*]]% set description "esFS Lustre Filesystem Monitor" [esms1->monitoring->healthchecks*[esfsmon*]]% set parameterpermissions required [esms1->monitoring->healthchecks*[esfsmon*]]% set timeout 120 [esms1->monitoring->healthchecks*[esfsmon*]]% show Parameter Value ------------------------------ ------------------------------------------------------------ Class of healthcheck misc Command /cm/local/apps/cmd/scripts/healthchecks/esfsmon_healthcheck Description esFS Lustre Filesystem Monitor Disabled no Extended environment no Name esfsmon Notes Only when idle no Parameter permissions required Revision Sampling method samplingonnode State flapping count 7 Timeout 120 Valid for node,headnode 4. Commit your changes. [esms1->monitoring->healthchecks*[esfsmon*]]% commit [esms1->monitoring->healthchecks[esfsmon]]% 5. Configure sfsmon_action. [esms1->monitoring->healthchecks[esfsmon]]% monitoring actions [esms1->monitoring->actions]% add esfsmon_action [esms1->monitoring->actions*[esfsmon_action*]]% show Parameter Value ------------------------------ ------------------- Command Description Name esfsmon_action Revision Run on headnode Timeout 5 isCustom yes 6. Set the command and parameters. [esms1->monitoring->actions*[esfsmon_action*]]% set command /cm/local/apps/cmd/scripts/actions/esfsmon_action [esms1->monitoring->actions*[esfsmon_action*]]% set timeout 660 [esms1->monitoring->actions*[esfsmon_action*]]% set description "Action for esfsmon failures" [esms1->monitoring->actions*[esfsmon_action*]]% show Parameter Value ------------------------------ -------------------------------------------------- Command /cm/local/apps/cmd/scripts/actions/esfsmon_action Description Action for esfsmon failures Name esfsmon_action Revision Run on headnode Timeout 660 isCustom yes 170 S–2327–B CLFS Administration Tasks [5] 7. Commit your changes. [esms1->monitoring->actions*[esfsmon_action*]]% commit [esms1->monitoring->actions[esfsmon_action]]% 5.2.8 Configure esfsmon.conf The esfmon.conf file configures esfsmon to monitor a file system (in this example, the file system is scratch). S–2327–B 171 Data Management Platform (DMP) Administrator's Guide Edit the esfmon.conf and make the following changes: ESFSMON_STATE_DIR=/var/esfsmon Path to esfsmon operational state (DO NOT CHANGE!) ESFSMON_DATA_DIR=/tmp/esfsmon Path to esfsmon operational data (DO NOT CHANGE!) ESFSMON_LUSTRE_CONTROL=/opt/cray/esms/cray-lustre-control-XX/bin/lustre_control Full path to lustre_control ESFSMON_SUSPEND_BASE=$ESFSMON_STATE_DIR/esfsmon_suspend_ Flag indicating that esfsmon is in a suspended state. Existence of this file indicates suspended state. Each esfsmon instance will append the Lustre file system name to the file. ESFSMON_RUNSAFE_BASE=$ESFSMON_STATE_DIR/esfsmon_runsafe_ Flag indicating that esfsmon is in a monitor-only (RUNSAFE) mode. Existence of this file indicates monitor-only (RUNSAFE) mode. Each esfsmon instance will append the Lustre file system name to the file. ESFSMON_MDS_FO_DISABLED_BASE=$ESFSMON_STATE_DIR/esfsmon_mds_fo_disabled_ Flag indicating that MDS fail over (FO) is disabled. Existence of this file indicates MDS FO is disabled. Each esfsmon instance will append the Lustre file system name to the file. ESFSMON_FO_DATA_BASE=$ESFSMON_STATE_DIR/esfsmon_fo_ This file contains the host name of the failed node. Each esfsmon instance will append the Lustre file system name to the file. ESFSMON_LOG_FREQ_BASE=$ESFSMON_STATE_DIR/esfsmon_log_freq_ The following control logging frequency to prevent filling the syslog if problems are not corrected quickly. The ESFSMON_LOG_FREQ contains parameters for how often to log. Valid frequencies are: 1. 0m = always log 2. 1 = log first time only - this is the default 3. xm = log every 'x' minutes 4. xh = log every 'x' hours 5. xd = log every 'x' days 172 S–2327–B CLFS Administration Tasks [5] ESFSMON_DEBUG_BASE=$ESFSMON_STATE_DIR/esfsmon_debug_ Flag indicating whether to run debug mode or not. Each esfsmon instance will append the Lustre file system name to the file. ESFSMON_MDS_NODE_scratch=lustre01-mds001 MDS node for each file system. Variable name is ESFSMON_MDS_NODE_filesystem=failover-mds-hostname. ESFSMON_MDS_CAT_scratch=esfs-mds-scratch MDS node category for each file system. ESFSMON_MDS_FO_NODE_scratch=lustre01-mds002 MDS failover node for each file system. Variable name is ESFSMON_MDS_FO_NODE_filesystem=failover-mds-hostname. ESFSMON_MDS_FO_CAT_scratch=esfs-mds-fo-scratch MDS failover node category for each file system. ESFSMON_MDS_FAILED_CAT_scratch=esfs-mds-failed-scratch MDS failed node category for each file system. ESFSMON_OSS_EVEN_CAT_scratch=esfs-oss-even-scratch OSS even numbered node category for each file system. ESFSMON_OSS_ODD_CAT_scratch=esfs-oss-odd-scratch OSS odd numbered node category for each file system. ESFSMON_OSS_FAILED_CAT_scratch=esfs-oss-failed-scratch OSS failed node category for each file system. ESFSMON_BASENAME_scratch=lustre01- CLFS node base name for each file system. Variable name is ESFSMON_BASENAME_filesystem=base-hostname. For example, lustre1-mds001 and lustre1-oss001 have a basename of lustre1-. CMSH="/cm/local/apps/cmd/bin/cmsh -c" Command path and argument to run cmsh commands. S–2327–B 173 Data Management Platform (DMP) Administrator's Guide 5.2.9 Activating esfsmon Before activating esfsmon in Bright, it is important that the file system-specific support files are in place. These include the esfsmon suspend and RUNSAFE control files in /var/esfsmon for the file system being monitored. Without these in place, an inadvertent failover may be attempted when esfsmon is activated for the file system. The following files need to be in place on the CIMS, before configuring esfsmon for a file system. • /var/esfsmon/esfsmon_suspend_filesystem — This file suspends monitoring of filesystem • /var/esfsmon/esfsmon_runsafe_filesystem — This file puts efsmon into RUNSAFE (monitor only) mode of filesystem 5.2.10 Configuring esfsmon healthcheck in Bright Procedure 50. Configuring esfsmon healthcheck in Bright 1. Log in to the CIMS as root and run the cmsh command. esms1# cmsh [esms1]% 2. Enter healthcheck configuration mode for the CIMS. [esms1]% monitoring setup healthconf headnode [esms1->monitoring->setup[HeadNode]->healthconf]% 3. From this prompt you can list what healthchecks are currently configured with the list command. [esms1->monitoring->setup[HeadNode]->healthconf]% list HealthCheck HealthCheck Param Check Interval ------------------------ ------------------ ------------------------ DeviceIsUp 120 ManagedServicesOk 120 chrootprocess 900 cmsh 1800 diskspace 2% 10% 20% 1800 exports 1800 failedprejob 900 failover 1800 interfaces 1800 ldap 1800 mounts 1800 mysql 1800 ntp 300 oomkiller 1800 schedulers 1800 smart 1800 [esms1->monitoring->setup[MasterNode]->healthconf]% 174 S–2327–B CLFS Administration Tasks [5] 4. Use the following commands to add a monitor for a file system named scratch. The first command adds another instance of esfsmon without any parameters. This is a base which we will configure for scratch. Note that none of these configuration activities take effect until we commit them to the Bright database with the cmsh commit command. [esms1->monitoring->setup[HeadNode]->healthconf]% add esfsmon [esms1->monitoring->setup*[HeadNode*]->healthconf*[esfsmon*]]% show Parameter Value ------------------------------ ------------------------------------------------ Check Interval 120 Disabled no Fail Actions Fail severity 10 GapThreshold 2 HealthCheck esfsmon HealthCheckParam LogLength 3000 Only when idle no Pass Actions Stateflapping Actions Store yes ThresholdDuration 1 Unknown Actions Unknown severity 10 [hopms->monitoring->setup*[HeadNode*]->healthconf*[esfsmon*]]% 5. Set the fail action and the healthcheck parameters to scratch. [esms1->monitoring->setup*[HeadNode*]->healthconf*[esfsmon*]]% set checkinterval 60 [esms1->monitoring->setup*[HeadNode*]->healthconf*[esfsmon*]]% set failactions esfsmon_action [esms1->monitoring->setup*[HeadNode*]->healthconf*[esfsmon*]]% set healthcheckparam scratch [esms1->monitoring->setup*[HeadNode*]->healthconf*[esfsmon*]]% show Parameter Value ------------------------------ ------------------------------------------------ Check Interval 120 Disabled no Fail Actions enter: esfsmon_action() Fail severity 10 GapThreshold 2 HealthCheck esfsmon HealthCheckParam scratch LogLength 3000 Only when idle no Pass Actions Stateflapping Actions Store yes ThresholdDuration 1 Unknown Actions Unknown severity 10 [hopms->monitoring->setup*[HeadNode*]->healthconf*[esfsmon*]]% commit [hopms->monitoring->setup[HeadNode]->healthconf[esfsmon]]% quit esms# 6. The esfsmon monitor for the scratch file system is setup but is suspended due to the existence of the/var/esfsmon/esfsmon_suspend_scratch file. Activate monitoring the scratch file system by removing S–2327–B 175 Data Management Platform (DMP) Administrator's Guide /var/esfsmon/esfsmon_suspend_scratch. To enable failover action to take place, remove the /var/esfsmon/esfsmon_runsafe_scratch file. 5.2.11 Controlling the Bright Cluster Manager Failover Monitor (esfsmon) The Bright failover monitor and failover action are controlled by the existence or non-existence of the following files in the /var/esfsmon directory on the CIMS: • esfsmon_suspend_filesystem – If present, monitoring of file system is suspended. – Set by the esfsmon_action script during failover to prevent false positive failure indications. – Set by the esfsmon_failback script during failback to prevent false positive failure indications. – May be set by the administrator to manually suspend monitoring. If manually set, it must be manually removed. • esfsmon_runsafe_filesystem – If present, monitoring of filesystem is in RUNSAFE mode. – Errors will be reported but no failover action will be performed. – Set automatically by the monitor when failures compromise the ability to perform a failover. – May be set by the administrator to manually enter RUNSAFE mode. – If set, it must be manually removed to de-activate RUNSAFE mode and enable failover action. • esfsmon_mds_fo_disabled_filesystem – If present, failover of MDS nodes is disabled. – Set automatically by the monitor if a standby MDS node is not present or if the primary MDS node has failed over to the standby MDS node. – May be set by the administrator to manually disable MDS failover. – If set, it must be manually removed to enable MDS failover. 176 S–2327–B CLFS Administration Tasks [5] 5.2.12 Avoiding Inadvertent Failover Operations In order to ensure that inadvertent failover actions are avoided, the administrator should suspend esfsmon by touching /var/esfsmon/esfsmon_suspend_filesystem during operations where any of the following may occur. • Powering off or powering on a node • Loss of InfiniBand® connectivity to a node • Loss of Lustre mounts or change in the number of Lustre mounts from the normal value • Performing a manual failover Remove the /var/esfsmon/esfsmon_suspend_filesystem file to resume failover monitoring when the above conditions no longer apply. 5.2.13 Tuning the esfsmon:filesystem Check Interval The esfsmon monitor is configured in Bright as the esfsmon health check on the CIMS. It takes a file system name as a parameter. Each esfsmon monitor will be named esfsmon:filesystem within Bright and can be tuned separately. The following commands will list the current check intervals (in seconds) of the master node health checks. The suggested check interval for esfsmon is 120 seconds and should not be changed without consulting Cray. Procedure 51. Tuning the esfsmon:filesystem check interval 1. Log in to the CIMS as root and run the cmsh command. esms1# cmsh [esms1]% 2. Show health check intervals for the CIMS. [esms1]% monitoring setup healthconf headnode [esms1->monitoring->setup[HeadNode]->healthconf]% list HealthCheck HealthCheck Param Check Interval ------------------------ ------------------ ------------------------ DeviceIsUp 120 ManagedServicesOk 120 cmsh 1800 esfsmon scratch1 120 esfsmon scratch2 120 exports 1800 failedprejob 900 failover 1800 ldap 1800 mounts 1800 mysql 1800 [hopms->monitoring->setup[MasterNode]->healthconf]% quit esms:# S–2327–B 177 Data Management Platform (DMP) Administrator's Guide 3. To change the interval of the esfsmon:filesystem, perform the following commands: (using the scratch1 file system and setting the interval to 160 seconds in this example). esms# cmsh [esms1]% monitoring setup healthconf headnode [esms1->monitoring->setup[HeadNode]->healthconf]% set esfsmon:scratch1 checkinterval 160 [esms1->monitoring->setup*[HeadNode*]->healthconf*]% commit [esms1->monitoring->setup[HeadNode]->healthconf]% quit esms# 5.2.14 Returning a Node to the CLFS Service Once a node has been failed over it will no longer be monitored. The failover action moves the failed node to an internal failed node category and sets flags on the takeover node to indicate it is now serving extra file system targets. To return a node to service the administrator must execute the esfsmon_failback command with the failed node host name as the sole argument. The esfsmon_failback command performs the following operations. • Verifies that the node being restored is currently in the failed node category • Moves the node to the proper active node category • Calls lustre_control to failback the Lustre targets The file system will be in RUNSAFE (monitor-only) mode. To enable automated failover, you must remove the /var/esfsmon/esfsmon_runsafe_filesystem file for the file system. 5.2.15 OSS Failover A custom monitor script runs periodically on the CIMS as a Bright healthcheck to verify the operational health of all CLFS nodes. Each OSS has three primary functions, the first two of which are absolutely required for proper CLFS operation: • Communicating with other nodes in the LNET InfiniBand fabric • Mounting OSTs and providing file system services for them • Communicating with a management server via TCP/IP. Loss of IP communication would not necessarily require a failover action, depending on site policies. When the monitor is operating, any hardware or software event that prevents the OSS from performing its primary functions will trigger an automated failover process that attempts to move the OSTs to a designated failover OSS that can continue to provide file system services for the OSTs. 178 S–2327–B CLFS Administration Tasks [5] 5.2.15.1 Failover Actions The underlying mechanism supporting failover of OSTs from one OSS to another is the format on each OST that specifies another OSS as a failed node. Custom health checks in Bright verify the basic functions listed above and if a failure is detected or if an OSS fails to respond to the Bright queries, the CIMS monitor script will shutdown power for the failed OSS, which un-mounts its OSTs, and then call lustre_control to perform a failover of the OSTs. Note that this typically doubles the OST load on the failover OSS. 5.2.15.2 Corrective Actions The actions required after an OSS failover are to determine the cause of the failure and correct it. Log entries by Bright and the monitor script should point toward the primary cause. 5.2.15.3 Recovery Actions Once the OSS has been repaired or patched and is ready for service, its OSTs should be un-mounted from the failover OSS and re-mounted on their home OSS. This can be done with the esfsmon_failback command and typically can be done with the file system up and active. It is highly recommended that the esfsmon_failback command be used to recover the previously failed OSS back into the system. The esfsmon_failback script takes the OSS host name as an argument and performs the necessary OST unmounts and re-mount operations via the lustre_control utility. It also performs all the administrative housekeeping to put the OSS back into the correct node category so it will be monitored again. Once the OSS is returned to service, the administrator needs to remove the /var/esfsmon/esfsmon_runsafe_filesystem file to return the monitor to an active failover mode. 5.2.16 MDT/MGS Failover At a high level, the primary functions of the metadata server (MDS) and metadata targets, which are storage devices (MDT), are the same as those of the OSS: Communicating on two networks (LNET and IP) and mounting and providing file system services for disks formatted for Lustre. The only difference is that the disks are the MDT and the MGS volumes (MGT). Cray recommends that MGT volumes are at least 100MB. When the esfsmon monitor is operating, any hardware or software event that prevents the MDS from performing its primary function, trigger an automated failover process that attempts to move the MDTs to a designated failover MDS that can continue to provide file system services for the MDTs. S–2327–B 179 Data Management Platform (DMP) Administrator's Guide 5.2.16.1 Failover Actions The designation in format of a failnode is as described above for OSS except in this case the MDT is formatted, typically to reference a standby MDS or another active MDS. Failure monitoring with Bright queries and the CIMS monitor script is the same as described above for OSS. An MDS failure also triggers the CIMS to initiate a power reset followed by a call to the standby or designated MDS to mount the MDT(s) of the failed MDS. 5.2.16.2 Corrective Actions The action required after an MDS failover are to determine the cause of the failure and correct it. Log entries by Bright and the monitor script should point toward the primary cause. 5.2.16.3 Recovery Actions Once the MDS has been repaired or patched and is ready for service, its MDT(s) should be un-mounted from the failover MDS and re-mounted on their home MDSs. This can be done with the esfsmon_failback command and can often be done while the file system is mounted and in use by clients. It is highly recommended that the esfsmon_failback command be used to recover the previously failed MDS back into the system. The esfsmon_failback script takes the MDS host name as an argument and performs the necessary unmount and re-mount operations via the lustre_control utility. It also performs all the administrative housekeeping to put the MDS back into the correct node category so it will be monitored again. After the MDS is returned to service, the administrator must remove the /var/esfsmon/esfsmon_runsafe_filesystem file to return the monitor to an active failover mode. 5.2.16.4 MGS Failover Most configurations feature a merged MDS and MGS server, with separate disks for MGT and MDT. All of the MDS discussion applies equally to MGS failover with the added requirement for different formatting. Specifically the MGT LUN itself does not specify a failnode parameter. Since it is the repository for configuration data of the file system, it does not need to report its new location and is functional by simply being re-mounted on a failover server. However, the corollary requirement to this one is that every other CLFS node (OST and MDT) in the configuration must be aware of this new location. 180 S–2327–B CLFS Administration Tasks [5] This is done by adding a second (i.e. alternate) mgsnode parameter to the format of every OST and MDT which points to the MGS/MDS failover node. Finally, MGS failover can take significantly longer than MDS failure alone or OSS failures because every client in the configuration must become aware of the new MGS node address. For large configurations this may exceed existing timeout values. 5.3 Configuring Kdump on CentOS™ (Optional) Kdump is configured on a DMP system by modifying the configuration files on CDL systems. Dump files from slave nodes are stored either on the CIMS using NFS™, or on the slave node local disk. To save dump files to a local disk on a slave node, create a persistent /var/crash partition. Procedure 52. Configuring kdump on CentOS 1. Log in to the CIMS as root. 2. Chose a slave node that you can use to test the kdump procedure (in this example esfs-oss1) and clone that slave node's software image. This example clones ESF-XX-2.1.0-201309252230 to ESF-XX-2.1.0-kdump. esms1# cd /cm/images esms1# cp -pr ESF-XX-2.1.0-201309252230 ESF-XX-2.1.0-kdump esms1# cmsh [esms1%] softwareimage [esms1->softwareimage]% clone ESF-XX-2.1.0-201309252230 ESF-XX-2.1.0-kdump 3. Commit your changes. [esms1->->softwareimage*[ESF-XX-2.1.0-kdump*]]% commit [esms1->->softwareimage[ESF-XX-2.1.0-kdump]]% 4. Create a test category to configure kdump Switch to category mode to create a test category. [esms1->->softwareimage[ESF-XX-2.1.0-kdump]]% category [esms1->category]% 5. Clone the production esFS-OSS or esFS-MDS category to create a test category. Note: Be sure to clone the production esFS-OSS category to an OSS-test category, and the production esFS-MDS category to an MDS-test category. These categories have different configurations and are not interchangeable. The ESF-XX-2.1.0-kdump software image is interchangeable for both node types. This procedure creates an OSS-test category. [esms1->category]% clone esFS-OSS OSS-test [esms1->category*[OSS-test*]% S–2327–B 181 Data Management Platform (DMP) Administrator's Guide 6. Assign the kdump esFS-OSS software image (ESF-XX-2.1.0-kdump) to the OSS-test category. [esms1->category*[OSS-test*]% set softwareimage ESF-XX-2.1.0-kdump 7. Add /var/crash to the exclude lists for the ESF-XX-2.1.0-kdump image. The vi editor launches which enables you to edit each of the exclude list files. Add - /var/crash/* to the list of excluded files: [esms1->category*[OSS-test*]% set excludelistsyncinstall [esms1->category*[OSS-test*]% set excludelistupdate [esms1->category*[OSS-test*]% set excludelistgrab [esms1->category*[OSS-test*]% set excludelistgrabnew 8. Save each file and commit your changes. [esms1->category*[OSS-test*]% commit [esms1->category[OSS-test]% 9. Assign the OSS-test test category to the test node (esfs-oss1) and commit your changes. [esms1->category[OSS-test]]% device use esfs-oss1 [esms1->device[esfs-oss1]]% set category OSS-test [esms1->device*[esfs-oss1*]]% commit [esms1->device[esfs-oss1]]% 10. If you are saving kdump crash files to the slave node local disk, add the following lines to the slave node's finalize script. This command opens the vi editor. [esms1->device[esfs-oss1]]% set finalizescript DEV=$( awk -- '{ if ($2 == "/localdisk/var/crash") { print $1; exit 0 } }' < /proc/mounts ) [ -n "$DEV" ] && e2label $DEV crash 11. Commit your changes. [esms1->device*]]% commit 12. Set the storage location for crash dumps If crash dumps will be saved to the CIMS proceed to step 13. If crash dumps will be saved to the slave node's local disk, proceed to step 14. 13. Save crash files to /var/crash on the primary CIMS 182 S–2327–B CLFS Administration Tasks [5] a. Use fsexports to determine whether the CIMS is exporting /var/crash. [esms1->device]]% use esms1 [esms1->device[esms1]]% fsexports [esms1->device[esms1]->fsexports]% list Name (key) Path -------------------------------------------- ------------------------------- /cm/shared@esmaint-net /cm/shared /home@esmaint-net /home /var/spool/burn@esmaint-net /var/spool/burn /cm/node-installer/certificates@esmaint-net /cm/node-installer/certificates /cm/node-installer@esmaint-net /cm/node-installer /var/crash@esmaint-net /var/crash b. Export /var/crash from the CIMS and configure it so that slave nodes can access it. [esms1->device[esms1]->fsexports]% add /var/crash [esms1->device*[esms1*]->fsexports*[/var/crash*]]% set name /var/crash@esmaint-net [esms1->device*[esms1*]->fsexports*[/var/crash*]]% set extraoptions no_subtree_check [esms1->device*[esms1*]->fsexports*[/var/crash*]]% set hosts esmaint-net [esms1->device[esms1]*]->fsexports*[/var/crash*]]% set write yes [esms1->device*[esms1*]->fsexports*[/var/crash*]]% commit c. Exit cmsh. [esms1->device[esms1]->fsexports[/var/crash]]% quit esms1# d. Update the exports. esms1# exportfs -a 14. Use the chroot shell, edit the /boot/pxelinux.cfg/default file for kdump test image created in step 2 (ESF-XX-2.1.0-kdump). a. Quit cmsh. [esms1->device[esms1]]% quit b. Use the chroot shell to edit the kdump test image. esms1# chroot /cm/images/ESF-XX-2.1.0-kdump esms:/>vi /boot/pxelinux.cfg/default c. Edit /boot/pxelinux.cfg/default file. d. Scroll down and locate the following line: # End of documentation, configuration follows: S–2327–B 183 Data Management Platform (DMP) Administrator's Guide e. Enter the following lines in the default configuration file: LABEL kdump KERNEL vmlinuz IPAPPEND 3 APPEND initrd=initrd crashkernel=512M CMDS console=tty0 console=ttyS1,115200n8 CMDE MENU LABEL ^KDUMP - Normal boot mode with kdump MENU DEFAULT f. Examine the other LABEL entries in the default configuration file and remove the line: MENU DEFAULT. g. Exit and save the file. 15. Verify that /var/crash exists in the ESF-XX-2.1.0-kdump image and is a directory. If necessary: esms1:/> mkdir /var/crash esms1:/> ls -l /var/crash 16. Edit the /etc/kdump.conf file and add or modify the following lines: esms1:/> vi /etc/kdump.conf a. The following lines should be uncommented and all other lines should be commented. path / core_collector makedumpfile -c --message-level 1 -d 27 link_delay 60 default reboot If you want to save crash dump files to the local add this line to the kdump.conf file. If the file system type is ext4, then replace ext3 with ext4. ext3 LABEL=crash Note: Create a persistent partition (/var/crash) in the disk setup XML file for the kdump test category (ESF-XX-2.1.0-kdump). Creating a separate partition for crash dumps on the slave node software image prevents /var from filling up and causing problems for the operating system. b. Exit and save the file. 17. Enable the kdump service. esms1:/> chkconfig kdump on 18. Exit the chroot shell. esms1:/> exit esms1# 19. Reboot the test node and run kdump 184 S–2327–B CLFS Administration Tasks [5] a. Start a console window on the test slave node (esfs-oss1). esms1# cmsh [esms1]% device; use esfs-oss1 [esms1->device[esfs-oss1]]% rconsole b. Reboot the test node (esfs-oss1). esfs-oss1: reboot esfs-oss1: Reboot in progress ... c. When the node reboots, initiate kdump. esms1# ssh esfs-oss1 esfs-oss1#echo c > /proc/sysrq-trigger Note: If dumping over NFS to the CIMS, the dump file is created in /var/crash on the CIMS node. If dumping to the slave node's local disk, the dump file is created in /var/crash on the slave node's local disk. 20. Make the kdump image the default image for all CDL nodes. a. Start cmsh and assign the test node (esfs-oss1) to the default esFS-OSS. esms1# cmsh [esms1]% device use esfs-oss1 [esms1->device[esfs-oss1]]% set category esFS-OSS [esms1->device*[esfs-oss1*]]% commit b. Switch to category mode and configure the default esFS-OSS category to use the kdump software image. [esms1->device[esfs-oss1]]% category [esms1->category]% use esFS-OSS [esms1->category[esFS-OSS]% set softwareimage ESF-XX-2.1.0-kdump 21. Reboot all of the nodes in the esFS-OSS category, so that they use the kdump software image. [esms1->category[esFS-OSS]% device [esms1->device]% reboot -c esFS-OSS esfs-oss1: Reboot in progress ... 22. Exit cmsh. [esms1->device]% quit 23. If necessary, repeat Procedure 52 on page 181 to configure a kdump image for MDS nodes. S–2327–B 185 Data Management Platform (DMP) Administrator's Guide 5.4 Configuring a NetApp™ Storage System Note: The instructions in this section apply for both SAS (Serial Attached SCSI) and Fibre Channel (FC) RAIDs and supersede the documentation supplied by the RAID manufacturer. Use the SANtricity™ storage management software from NetApp, Inc. to manage external NetApp RAID storage devices. SANtricity is provided as a separate package and is installed from a CD on the CIMS. The RAID controllers are set to IP addresses on the esmaint-net network in the 10.141.100.xxx range. See Figure 2 and CIMS Network Configuration on page 25. 5.4.1 Installing SANtricity Storage Manager Software for NetApp Devices The SANtricity software is generally preinstalled and the SANtricity media is shipped with the system. However, if the CIMS does not have the software installed, you can install it. The SANtricity SMClient executable is found in /opt/SMgr/client. Procedure 53. Installing SANtricity storage management software 1. Log in to the CIMS as root. 2. If you are installing from the SANtricity CD, insert it into the CIMS CD drive and mount it. esms1# mount /dev/cdrom /media/cdrom mount: block device /dev/sr0 is write-protected, mounting read-only Or, if you are installing from the SMIA-LINUXX64-10.80.A0.47.bin file, copy SMIA-LINUXX64-10.80.A0.47.bin to /root/release. esms1# cp ./SMIA-LINUX-10.70.A0.25.bin /root/release/ 3. Set the DISPLAY environment variable. esms1# export DISPLAY=:0.0 4. Verify that the X Window System is functioning by launching xterm or executing the xlogo utility. esms1# xterm Exit the xterm window. 5. Run the executable file. If you are installing from the CD: esms1# /bin/bash /media/cdrom/Linux*x86_64/install/SMIA-LINUXX64-10.80.A0.47.bin 186 S–2327–B CLFS Administration Tasks [5] Or, if you are installing from a directory: esms1# /root/release/SMIA-LINUXX64-10.80.A0.47.bin /root/release/SMIA-LINUXX64-10.80.A0.47.bin Preparing to install... Extracting the JRE from the installer archive... Unpacking the JRE... Extracting the installation resources from the installer archive... Configuring the installer for this system's environment... Launching installer... 6. Click Next. The License Agreement window displays. 7. Accept the license agreement and click Next. The Select Installation Type window displays. 8. Click Typical (Full Installation), then click Next. The Multipathing Driver Warning window displays. 9. Click OK. The Pre-Installation Summary window displays. 10. Click Install. The Installing SANtricity window displays and shows the installation progress. When the installation completes, an Install Complete window appears. 11. Click Done. The SANtricity client is installed in /usr/bin/SMclient and is currently running. 12. Close the file browser and eject the CD. esms1# eject 5.4.2 Configuring LUNs for NetApp Devices Create a Volume Group and the LUNs that are members of it. Procedure 54. Creating a volume group for NetApp devices You must be logged on to the CIMS as root. 1. Start the SANtricity software. esms1# /usr/bin/SMclient The SANtricity Storage Manager window displays. 2. If the Select Addition Method window displays, choose one of the following options; otherwise, skip to step 3: • Automatic — Select this option if you did not assign IP addresses to the storage array controllers using a serial connection. The SANtricity software automatically detects the available controllers, in-band, using the Fibre Channel or InfiniBand link. S–2327–B 187 Data Management Platform (DMP) Administrator's Guide • Manual — Select this option if esmaint-net IP addresses are assigned to the RAID controllers. Refer to IP address scheme discussed in CIMS Network Configuration on page 25. Figure 2 shows how RAID controllers are connected to the es-maint network. Note: The rest of this procedure assumes that you selected the Manual option. 3. Double-click the name for the Storage Array that you want to configure. The Array Management window displays. 4. Click the Logical/Physical tab. 5. Right-click Unconfigured Capacity and select Create Volume. The Create Volume wizard displays. 6. Click Next on the Introduction (Create Volume) window. 7. Select the Manual option on the Specify Volume Group (Create Volume) window. 8. Select the tray, and desired slots, and click Add. 9. Verify that the RAID level is correct (for example RAID 5). 10. Click Calculate Capacity. 11. Click Next on the Specify Volume Group (Create Volume) window. When you create the first Volume Group, you are prompted to create the first volume. Procedure 55. Creating and configuring volumes for NetApp devices 1. Enter a new volume capacity. 2. Specify units as GB or MB. 3. Enter a name. 4. Select the Customize Settings option. 5. Click Next in the Specify Capacity/Name (Create Volume) window. 6. Verify the settings on the Customize Advanced Volume Parameters (Create Volume) window. These settings are used for the all of the LUNs. • For Volume I/O characteristics type, verify that File System is selected. • For Preferred Controller Ownership, verify that Slot A is selected. This places the LUN on the A Controller. 7. Click Next in the Customize Advanced Volume Parameters (Create Volume) window. 188 S–2327–B CLFS Administration Tasks [5] 8. In the Specify Volume to LUN Mapping window, select the Default mapping option. 9. If not configuring multipath, select Host type, and select Linux™ from the drop-down menu. If you intend to configure multipath, set Host type to Linux (DM-MP). 10. Click Finish in the Specify Volume to LUN Mapping window. 11. When prompted to create more LUNs in the Creation Successful (Create Volume) window, select Yes unless this is the last volume you are creating. If this is the last volume, select No and skip to step 15. 12. In the Allocate Capacity (Create Volume) window, verify that Free Capacity is selected on Volume Group 1 (RAID 5). 13. Click Next in the Allocate Capacity (Create Volume) window. 14. Repeat step 1 through step 13 to create all of the volumes. 15. Click OK in the Completed (Create Volume) window. 16. Create a hot spare. The hot spare provides a ready backup if any of the drives in the Volume Group fail. a. Right-click on a drive in the right portion of the window and select Hot Spare Coverage. b. Select the Manually Assign Individual Drives option. c. Click OK. d. Click Close. 17. Exit the tool. 5.4.3 Host Mappings If you are configuring multipath, and are experiencing errors from NetApp™ storage devices such as Volume not on preferred path due to AVT/RDAC failover, be sure to set the host mappings parameter to Linux (DM-MP) or Linux (on older firmware). Use the SANtricity™ Storage Manager Software command Host Mappings->Default Group->Change Default Host Operating System to configure the host mappings setting. S–2327–B 189 Data Management Platform (DMP) Administrator's Guide 5.4.4 Configuring Remote Logging of NetApp™ Storage System Messages NetApp storage systems use Simple Network Management Protocol (SNMP) to provide boot RAID messages on the esmaint-net network. Configure the community settings for the RAID device using the serial console connection (a custom cable is shipped with NetApp devices for this purpose). Make the console connection in a similar manner to how you configure a switch. (See Adding a Managed Switch or Device to the Bright Configuration on page 111.) Refer to your NetApp storage system documentation for more information. 5.4.5 Add a NetApp RAID Storage System to Bright RAID devices are added as type genericdevice in cmsh or as Other devices using the cmgui. The procedure below shows how to add a RAID device to Bright using cmsh. Procedure 56. Adding a NetApp RAID storage system to Bright 1. Install the SANtricity (SMclient) software on the CIMS. Refer to Installing SANtricity Storage Manager Software for NetApp Devices on page 186. 2. Set the RAID controller IP addresses to valid esmaint-net addresses (in the 10.141.100.xxx range). Refer to IP address scheme discussed in CIMS Network Configuration on page 25. Figure 2 shows how RAID controllers are connected to the esmaint-net network. 3. Log in to the CIMS as root and run cmsh. esms1# cmsh [esms1]% 4. Switch to device mode. [esms1]% device 5. Add the NetApp RAID controller(s) under Other devices in the cmgui resource tree, or as a genericdevice in cmsh. This procedure adds a NetApp 3992 controller A for example. [esms1->network[storage-net]]% device [esms1->device]% add genericdevice netapp3992-cntrlA 6. Set the device information for model, rack number, device height in rack units, and device U position in the rack. [esms1->device*[netapp3992-cntrlA*]]% set model LSI3992 [esms1->device*[netapp3992-cntrlA*]]% set rack rack_num [esms1->device*[netapp3992-cntrlA*]]% set deviceheight rack_units [esms1->device*[netapp3992-cntrlA*]]% set deviceposition rack_position 190 S–2327–B CLFS Administration Tasks [5] 7. Set the RAID controller Ethernet port IP address on the esmaint-net, for example 10.141.100.10. [esms1->device*[netapp3992-cntrlA*]]% set ip controller_ip_address 8. Set MAC address for the RAID controller Ethernet port. [esms1->device*[netapp3992-cntrlA*]]% set mac MAC_address 9. Set the network to esmaint-net. 10. Set power control to custom (power is controlled from the SANtricity client software). [esms1->device*[netapp3992-cntrlA*]]% set powercontrol custom 11. (Optional) Add notes for this controller. The following command opens the vi editor where notes can be added for this device. For example: oss001 connection to RAID controller A, Even LUNs. [esm1->device*[netapp3992-cntrlA*]]% set notes 12. Show the controller settings. [esm1->device*[netapp3992-cntrlA*]]% show Parameter Value ------------------------------ -------------------------------------- Activation Tue, 14 May 2013 08:20:30 CDT Additional Hostnames Container index 0 Custom ping script Custom ping script argument Custom power script Custom power script argument Device height 4 Device position 30 Ethernet switch Hostname netapp3992-cntrlA Ip 10.141.100.10 Mac 00:0B:5F:CE:2F:40 Model Network esmaint-net Notes Partition base Power control custom PowerDistributionUnits Rack 1 Revision Tag 00000000a000 Type GenericDevice Userdefined1 Userdefined2 S–2327–B 191 Data Management Platform (DMP) Administrator's Guide 5.5 Rediscovering the New LUNs This procedure causes the CLFS to rediscover the new LUNs created in Procedure 55 on page 188. Procedure 57. Rebooting the CLFS and verifying LUNs are recognized 1. Log in to the CIMS as the root. 2. Use SSH to log in to the CLFS nodes and enter the following command to ensure that the LUNs are recognized: esms1# ssh esfs-oss001 3. Reboot the node or probe the SCSI bus to verify the LUNs are available to the MDS or OSS node. Note: Two paths to LUN 5 (/dev/sdg and /dev/sdm) indicate multipath is enabled. [root@esfs-oss001 ~]# lsscsi [0:2:0:0] disk DELL PERC H710P 3.13 /dev/sda [5:0:0:0] cd/dvd PLDS DVD+-RW DS-8A8SH KD51 /dev/sr0 [7:0:0:0] disk LSI INF-01-00 0780 /dev/sdb [7:0:0:1] disk LSI INF-01-00 0780 /dev/sdc [7:0:0:2] disk LSI INF-01-00 0780 /dev/sdd [7:0:0:3] disk LSI INF-01-00 0780 /dev/sde [7:0:0:4] disk LSI INF-01-00 0780 /dev/sdf [7:0:0:5] disk LSI INF-01-00 0780 /dev/sdg [8:0:0:0] disk LSI INF-01-00 0780 /dev/sdh [8:0:0:1] disk LSI INF-01-00 0780 /dev/sdi [8:0:0:2] disk LSI INF-01-00 0780 /dev/sdj [8:0:0:3] disk LSI INF-01-00 0780 /dev/sdk [8:0:0:4] disk LSI INF-01-00 0780 /dev/sdl [8:0:0:5] disk LSI INF-01-00 0780 /dev/sdm 4. List the disk devices by using the fdisk command to verify that the LUNs (volumes) are configured. [root@esfs-oss001 ~]# fdisk -l 5.6 Partitioning the LUNs After you finish creating, formatting, and zoning the LUNs on the RAID, you must partition them. Refer to the NetApp documentation for the SANtricity Storage Manager software for more information. 192 S–2327–B CLFS Administration Tasks [5] 5.7 Adding Nodes to CLFS Categories Each CLFS slave node must be added to Bright and assigned to the esFS-MDS or esFS-OSS category to configure node-specific device information. The physical node and network interface information should already exist in the Bright database. Note: This procedure adds MDS nodes to the esFS-MDS category. Repeat this procedure to add OSS nodes to the esFS-OSS category. Procedure 58. Adding nodes to a CLFS category 1. Log in to the CIMS as root and run cmsh. esms1# cmsh [esms1]% 2. Switch to device mode: [esms1]% device 3. Add the new node to the esFS-MDS category. Note: This procedure uses the example host name esfs-mds001 and the category name esFS-MDS. Substitute your actual CLFS node host name and the category name used in the previous procedure. [esms1->device]% use esfs-mds001 [esms1->device[esfs-mds001]]% set category esFS-MDS [esms1->device[esfs-mds001*]]% 4. Set the rack information (device height, position in the rack, and rack number). This example assumes that the device height is 2U, its position in the rack is 12, and the rack number is 1. Substitute the actual values from your CLFS node position. [esms1->device[esfs-mds001*]]% set deviceheight 2 [esms1->device[esfs-mds001*]]% set deviceposition 12 [esms1->device[esfs-mds001*]]% set rack 1 5. Commit your device changes. [esms1->device[esfs-mds001*]]% commit [esms1->device[esfs-mds001]]% 6. Repeat step 3 through step 5 to add each MDS node to the esFS-MDS category and each OSS node to the esFS-OSS category. S–2327–B 193 Data Management Platform (DMP) Administrator's Guide 7. Check the MDS and OSS categories to verify they are configured properly. [esms1->device[esfs-mds001]]% category [esms1->category]]% list Name (key) Software image ------------------------ ------------------------ default default-image default-diskless default-image esFS-MDS ESF-XX-2.1.0-201309252230 esFS-OSS ESF-XX-2.1.0-201309252230 esLogin-XC ESL-XC-1.0.2-2013022113+ esLogin-XE ESL-XE-1.1.1_CLE4.1 [esms1->category]% usedby esFS-MDS Category used by the following: Type Name Parameter Autochange ---------------- ------------------------ ------------------------ ------------ Device esfs-mds001 category no Device esfs-mds002 category no [esms1->category]% usedby esFS-OSS Category used by the following: Type Name Parameter Autochange ---------------- ------------------------ ------------------------ ------------ Device esfs-oss001 category no Device esfs-oss002 category no Device esfs-oss003 category no Device esfs-oss004 category no 8. Exit cmsh. [esms1->category]% quit esms1# 5.8 Configuring Bright Categories for CLFS Nodes Note: Use Procedure 59 on page 194 to customize the Bright esFS-MDS and esFS-OSS categories for MDS and OSS nodes. Then use Procedure 58 on page 193 to add CLFS nodes to the Bright category. The default category is assigned to the first CLFS node when it was cloned. The esFS-MDS and esFS-OSS categories must be assigned the software image created during installation. You must also define the default gateway, default disk partitions, and finalize script for your site configuration. Procedure 59. Configuring Bright CLFS categories Note: This procedure customizes the esFS-MDScategory for your site configuration. Repeat this procedure to customize the esFS-OSS category. 194 S–2327–B CLFS Administration Tasks [5] 1. Log in to the CIMS as root and run cmsh. esms1# cmsh [esms1]% 2. Switch to category mode and list categories and associated software images. [esms1]% category [esms1->category]% list Name (key) Software image ------------------------ ------------------------ default default-image default-diskless default-image esFS-MDS default-image esFS-OSS default-image esLogin-XC ESL-XC-1.0.2-2013022113+ esLogin-XE ESL-XE-1.1.1_CLE4.1 3. Assign the CLFS image (ESF-XX-2.1.0-201309252230) created by ESFinstall, to the esFS-MDS category. [esms1->category]% use esFS-MDS [esms1->category[esFS-MDS]]% set softwareimage ESF-XX-2.1.0-201309252230 [esms1->category*[esFS-MDS*]]% 4. Continue at step 5 if configuring the esFS-OSS category. For the esFS-MDS category, set the default gateway so that MDS nodes can contact an external LDAP server on site-user-net. For gateway, use the IP address of your site's gateway (on site-user-net). [esms1->category*[esFS-MDS*]]% set defaultgateway gateway 5. Commit the changes. [esms1->category*[esFS-MDS*]]% commit [[esms1->category[esFS-MDS]]% S–2327–B 195 Data Management Platform (DMP) Administrator's Guide 6. Configure the esFS-MDS or esFS-OSS category with the proper disk partition sizes and configuration. Note: A custom disk partition layout can be applied using an XML schema to a category of nodes such as esFS-MDS or esFS-OSS. A disk partition layout that is applied to an individual node within a category overrides the category setting. Add XML entries to the site.esfs-diskfull.xml file if there are several LUNs on a SAS RAID that are used for MDTs. The XML file currently defines /dev/sda through /dev/sdz. Changes made to the esfs-diskfull.xml file do not occur until they are saved in cmgui, or committed in cmsh, and the CLFS node is rebooted. Bright detects that the partitioning has changed and invokes a FULL install. (Bright also ignores the NOSYNC install mode and will repartition the drives on other CLFS nodes in the category as well.) Older model CLFS nodes can use esfs-small-diskfull.xml for disk partitioning to accommodate smaller hard drives. Important: If the default disk setup XML files are updated in a ESM release and the site disk setup XML files have been customized, system administrators must compare the newly released disk setup XML files with the current production disk setup XML files, and merge the changes manually. After the changes have been merged, you must load the updated disk setup file into the Bright database for the node category and reboot all the nodes that use that category. a. Quit cmsh and copy the default XML configuration to an etc directory to prevent it from being overwritten during software updates. [esms1->category[esFS-MDS]]% quit esms1# cd /opt/cray/esms/cray-es-diskpartitions-XX esms1# mkdir -p etc esms1# cp -p default/esfs-diskfull.xml etc/site.esfs-diskfull.xml b. Edit the site.esfs-diskfull.xml file to change partition sizes or configuration. esms1# vi etc/site.esfs-diskfull.xml c. Save the changes and return to cmsh. 7. Set the disk setup parameter for the category and commit the changes. esms1# cmsh [esms1]% category use esFS-MDS [esms1->category[esFS-MDS]]% set disksetup /opt/cray/esms/cray-es-diskpartitions-XX /etc/site.esfs-diskfull.xml [esms1->category*[esFS-MDS*]]% commit 8. Confirm your changes to the site.esfs-diskfull.xml script. [esms1->category[esFS-MDS]]% get disksetup 196 S–2327–B CLFS Administration Tasks [5] 9. Repeat this procedure to configure the esFS-OSS category. Note: You must also configure the CLFS node finalize scripts. 10. Refer to Adding Nodes to CLFS Categories on page 193 to add CLFS nodes to the esFS-MDS and esFS-OSS categories. 5.9 Configuring Lustre® File Systems on CLFS Nodes This section describes how to configure Lustre file systems on CLFS nodes from the CIMS. Procedure 60. Configuring Lustre file systems on CLFS nodes from the CIMS 1. Log in to the CIMS as root. 2. Copy /opt/cray/esms/cray-lustre-control-XX/default/etc/example.fs_defs to /opt/cray/esms/cray-lustre-control-XX/etc/scratch.fs_defs to prevent it from being overwritten during software updates. esms1# mkdir -p /opt/cray/esms/cray-lustre-control-XX/etc esms1# cd /opt/cray/esms/cray-lustre-control-XX/etc esms1# cp -p /opt/cray/esms/cray-lustre-control-XX/default/etc/example.fs_defs scratch.fs_defs 3. Display the Lustre network identifier (NID) map information to include in the scratch.fs_defs file. esms1# lustre_control dump_nid_map -w esfs-mds00[1-2],esfs-oss00[1-4] Performing 'dump_nid_map' from esms1 at Thu Jan 31 14:47:49 CST 2013 Hostname to LNET nid mapping for the nodes: esfs-mds001,esfs-mds002,esfs-oss001,esfs-oss002,esfs-oss003,esfs-oss004 nid_map: nodes=esfs-mds00[1-2] nids=10.149.0.[1-2]@o2ib nid_map: nodes=esfs-oss00[1-4] nids=10.149.0.[3-6]@o2ib 4. To configure persistent device names for the RAID disk devices you configured in step 6, Cray recommends that you use the /dev/disk/by-id/ persistent S–2327–B 197 Data Management Platform (DMP) Administrator's Guide device names. To determine the persistent drive names for the devices you configured in step 6, use SSH to log in to the MDS or OSS node and list the contents of the /dev/disk/by-id directory. esms1# ssh esfs-mds001 Last login: Fri Apr 19 11:08:03 2013 from esms1.cm.cluster [root@esfs-mds001 by-id]# ls -l /dev/disk/by-id total 0 lrwxrwxrwx 1 root root 9 Apr 18 21:14 scsi-3600508e000000000faef8330f46c9c09 -> ../../sda lrwxrwxrwx 1 root root 10 Apr 18 21:14 scsi-3600508e000000000faef8330f46c9c09-part1 -> ../../sda1 lrwxrwxrwx 1 root root 10 Apr 18 21:14 scsi-3600508e000000000faef8330f46c9c09-part2 -> ../../sda2 lrwxrwxrwx 1 root root 9 Apr 18 16:43 scsi-3600a0b800026cfe400002dfd4becf544 -> ../../sdb lrwxrwxrwx 1 root root 9 Apr 18 21:14 wwn-0x600508e000000000faef8330f46c9c09 -> ../../sda lrwxrwxrwx 1 root root 10 Apr 18 21:14 wwn-0x600508e000000000faef8330f46c9c09-part1 -> ../../sda1 lrwxrwxrwx 1 root root 10 Apr 18 21:14 wwn-0x600508e000000000faef8330f46c9c09-part2 -> ../../sda2 lrwxrwxrwx 1 root root 9 Apr 18 16:43 wwn-0x600a0b800026cfe400002dfd4becf544 -> ../../sdb Note: The persistent device name for sdb is scsi-3600a0b800026cfe400002dfd4becf544. Enter the value in the device configuration section of scratch.fs_defs. Important: For a multipath configuration, device names should link to a logical alias under /dev/mapper/alias . This can be configured in the node finalize script either for the node, or the node category (such as eslogin-XC). Multipath configuration device names can also be configured as a dm-uuid path name such as /dev/disk/by-id/dm-uuid-mpath-wwid. Either option in step 4 is a link that points to the /dev/dm-x device. 5. Edit the scratch.fs_defs file to set values. esms1# vi /opt/cray/esms/cray-lustre-control-XX/etc/scratch.fs_defs 6. Modify the fs_name, nid_map, mds, mgt, ost, and stripe_count settings in the scratch.fs_defs file. a. Enter the fs_name: (such as scratch). # file system name - must be 8 characters or less fs_name: scratch b. Enter the nid_map information obtained in step 3. nid_map: nodes=esfs-mds00[1-2] nids=10.149.0.[2-3]@o2ib nid_map: nodes=esfs-oss00[1-4] nids=10.149.0.[4-7]@o2ib c. Enter the mdt and mgt definitions and disk device name(s) for your system 198 S–2327–B CLFS Administration Tasks [5] obtained from step 4. Include the failover node definition (fo_node) if configuring a failover system. If there is no failover node definition, remove this line. Note: In this example, the metadata target (MDT) and management target (MGT) functions share the same CLFS node (esfs-mds001). ## MDT ## MetaData Target mdt: node=esfs-mds001 dev=/dev/disk/by-id/scsi-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx fo_node=esfs-mds002 ## MGT ## Management Target mgt: node=esfs-mds001 dev=/dev/disk/by-id/scsi-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx fo_node=esfs-mds002 7. Enter the ost definitions and disk device name(s) for your system obtained from step 4. Include the failover node definition (fo_node) if configuring a failover system. If there is no failover node definition, remove this line. ## OST ## Object Storage Target(s) ost: node=esfs-oss001 dev=/dev/disk/by-id/scsi-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx fo_node=esfs-oss002 ost: node=esfs-oss003 dev=/dev/disk/by-id/scsi-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx fo_node=esfs-oss004 8. Exit and save the scratch.fs_defs file and proceed to Configuring Lustre® File Systems on MDS and OSS Nodes on page 199. 5.10 Configuring Lustre® File Systems on MDS and OSS Nodes Procedure 61. Configuring Lustre file systems on MDS and OSS nodes 1. From the CIMS as root, install the Lustre file system definitions. esms1# cd /opt/cray/esms/cray-lustre-control-XX/etc esms1# lustre_control install scratch.fs_defs Performing 'install' from esms1 at Mon Apr 22 14:38:42 CDT 2013 Parsing file system definitions file: scratch.fs_defs Parsed file system definitions file: scratch.fs_defs The 'scratch' file system definitions were successfully installed! S–2327–B 199 Data Management Platform (DMP) Administrator's Guide 2. Format the file system on scratch. esms1# lustre_control reformat -f scratch Performing 'reformat' from esms1 at Mon Apr 22 14:42:52 CDT 2013 About to reformat all targets for the following file system(s): scratch Continue? (y|n|q) y 3. Start the Lustre file system scratch. esms1# lustre_control start -p -f scratch 4. Check the status of the scratch file system. esms1# lustre_control status -f scratch Performing 'status' from esms1 at Mon Apr 22 14:44:14 CDT 2013 File system: scratch Device Host Mount OST Active Recovery Status MGS esfs-mds001 Mounted N/A Unknown MGS* esfs-mds002 Unmounted N/A N/A scratch-MDT0000 esfs-mds001 Mounted N/A INACTIVE scratch-MDT0000* esfs-mds002 Unmounted N/A N/A scratch-OST0000 esfs-oss1 Mounted Active INACTIVE scratch-OST0000* esfs-oss2 Unmounted Active N/A . . . 5.11 Creating a Generic CLFS Node in Bright Cluster Manager® The easiest way to add a new slave node is to clone an existing node that is configured and fully functional in Bright Cluster Manager® (Bright). When you do not have a functioning CLFS node, you must clone the default node (node001) created during the CIMS installation process. Note: In this guide, some examples are left-justified to fit command lines or display output on a single line. Left-justification has no special significance. Procedure 62. Creating a CLFS node in Bright Note: The following procedures use the Bright management shell (cmsh). They may also be performed using the Bright GUI (cmgui). The cmsh command prompt displays an asterisk (*) when you have uncommitted changes. Be sure to commit your changes using the commit command before exiting cmsh, or your changes will be lost. Alternatively, the cmgui provides a Save button to save and commit your changes. 200 S–2327–B CLFS Administration Tasks [5] 1. Log in to the CIMS as root and run cmsh. esms1# cmsh [esms1]% 2. Enter device mode. [esms1]% device [esms1->device]% 3. Perform this procedure for each CLFS node. List the available devices. [esms1->device]% list Type Hostname (key) MAC Category Ip Network ---------------------- ---------------- ------------------ ---------- --------------- -------------- EthernetSwitch switch01 00:00:00:00:00:00 10.141.253.1 esmaint-net MasterNode esms1 78:2B:CB:40:CE:CA 10.141.255.254 esmaint-net PhysicalNode node001 00:00:00:00:00:00 default 10.141.0.1 esmaint-net 4. If this is the first CLFS node created for the system, you must clone the default node which is node001 in step 3. This example creates a new CLFS node named esfs-mds001. Note: When the CIMS software is installed, Bright creates a default node, node001, that uses the default slave image in /cm/images/default-image. This image is assigned to the newly cloned node. [esms1->device]% device list |grep node001 PhysicalNode node001 00:00:00:00:00:00 default 10.141.0.1 esmaint-net [esms1->device]% clone node001 esfs-mds001 Base name mismatch, IP settings will not be modified! If you have already created a CLFS node, then clone the configured CLFS to create another MDS or OSS node. Important: Always make sure that newly cloned nodes boot from default-image without errors before you begin other configuration tasks. When repeating this procedure to create additional CLFS nodes, set the newly cloned node category to default and boot the node using the default-image to configure the node for Bright. [esms1->device]% clone esfsnode esfs-mds001 Base name mismatch, IP settings will not be modified! [esms1->device*[esfs-mds001*]]% commit [esms1->device[esfs-mds001]]% exit [esms1->device*]% set esfs-mds001 category default [esms1->device*]% commit 5. Change the interface settings for the new (cloned) node. S–2327–B 201 Data Management Platform (DMP) Administrator's Guide a. Switch to interfaces mode and list interfaces on esfs-mds001. [esms1->device]% use esfs-mds001 [esms1->device]% interfaces [esms1->device[esfs-mds001]->interfaces]% list Type Network device name IP Network ------------ -------------------- ---------------- ---------------- bmc ipmi0 10.148.0.1 ipmi-net physical BOOTIF [prov] 10.141.0.1 esmaint-net b. Set the BOOTIF and ipmi0 interface addresses for the new node. These addresses must be different from those used by the default or original node. [esms1->device*[esfs-mds001*]->interfaces]% set bootif ip 10.141.0.2 [esms1->device*[esfs-mds001*]->interfaces*]% set ipmi0 ip 10.148.0.2 [esms1->device*[esfs-mds001*]->interfaces*]% list Type Network device name IP Network ------------ -------------------- ---------------- ---------------- bmc ipmi0 10.148.0.2 ipmi-net physical BOOTIF [prov] 10.141.0.2 esmaint-net c. Configure the ib-net network and interface. Note: Cray recommends that ib0 on OSS nodes be connected to the IB switch, ib1 is should not be used, and ib2 and ib3 connect to the storage array controllers. [esms1->device*[esfs-mds001*]->interfaces*]% add physical ib0 [esms1->device*[esfs-mds001*]->interfaces*[ib0*]]% set network ib-net [esms1->device*[esfs-mds001*]->interfaces*[ib0*]]% set ip 10.149.0.2 [esms1->device*[esfs-mds001*]->interfaces[ib0]]% show Parameter Value ------------------------------ ------------------------------------------------ Additional Hostnames Card Type DHCP no IP 10.149.0.2 MAC 00:00:00:00:00:00 Network ib-net Network device name ib0 Revision Speed Type physical d. Commit your changes. [esms1->device*[esfs-mds001*]->interfaces*[ib0*]]% commit e. Exit ib0. [esms1->device[esfs-mds001]->interfaces[ib0]]% exit 202 S–2327–B CLFS Administration Tasks [5] f. Check the results. [esms1->device[esfs-mds001]->interfaces]% list Type Network device name IP Network ------------ -------------------- ---------------- ---------------- bmc ipmi0 10.148.0.2 ipmi-net physical BOOTIF [prov] 10.141.0.2 esmaint-net physical ib0 10.149.0.2 ib-net g. Exit interface mode and return to device mode. [esms1->device[esfs-mds001]->interfaces]% exit h. Display the status of the new node. [esms1->device[esfs-mds001]]% status esfs-mds001 ............... [ DOWN ] (Unassigned) Note: The node is unassigned because the MAC address has not been set in Bright. 6. Set the MAC address for the CLFS node (eth0 on esmaint-net). [esms1->device[esfs-mds001]]% set mac MACaddress 7. Set the management network to esmaint-net. [esms1->device[esfs-mds001*]]% set managementnetwork esmaint-net 8. Commit your changes and quit cmsh. [esms1->device*[esfs-mds001*]]% commit [esms1->device[esfs-mds001]]% quit esms1# 9. If you are configuring an MDS node, refer to Configuring the Site User Network (for MDS Nodes Only) on page 208, to configure network parameters for the site-user-network. If you are configuring an OSS node, test-boot the node. 5.12 Configuring Node Finalize Script for MDS Nodes Prepare the node finalize script for the MDS nodes and load it for the esFS-MDS category. A finalize script (run before init) is used to set a file configuration or to initialize special hardware, sometimes after a hardware check. It is run in order to make software or hardware work before, or during the later init stage of boot. Use a finalize script to execute commands before init, when the commands cannot be stored persistently anywhere else, or when it is needed because a choice between (otherwise non-persistent) configuration files must be made based on the hardware before init starts. S–2327–B 203 Data Management Platform (DMP) Administrator's Guide Important: Files created or modified by a finalize script must be listed in the excludelistupdate exclude list for the category. Software updates will over write customized files if the files are not specified in an exclude list for the category. Customized files must also be specified in the excludelistgrab, and excludelistgrabnew exclude lists to prevent customized files from being copied to the CIMS. Procedure 63. Configuring the node finalize script for MDS nodes Note: The named, nslcd, nscd, and ldap services are turned off in the default finalize script for MDS nodes. To enable these services, uncomment the corresponding lines in the finalize script. 1. Log in to the CIMS as root. 2. Make a copy of the default MDS finalize script to an etc directory to prevent it from being overwritten during software updates. esms1# cd /opt/cray/esms/cray-es-finalize-scripts-XX esms1# mkdir -p etc esms1# cp -p default/mds_finalize.sh etc/site.mds_finalize.sh Note: Cray recommends that the subnet manager for the IB fabric should be run on ib0 of the MDS node(s) for system configurations that have many CLFS file system nodes connected to an IB switch as described in step 3. 3. (Optional) Add/uncomment the following command line if you want to start the IB subnet manager on ib0 for the esFS-MDS category. esms1# vi etc/site.mds_finalize.sh echo "/usr/sbin/opensm --daemon -g /` usr/sbin/ibstat mlx4_0 1 | grep GUID | awk '{print $3}' "` >> /localdisk/etc/rc.d/rc.local exit 0 4. To configure LDAP, refer to Configure LDAP on MDS Nodes on page 206 after completing this procedure. 5. Run the cmsh and set the finalize script for the esFS-MDS category. esms1# cmsh [esms1]% category [esms1->category]% use esFS-MDS [esms1->category[esFS-MDS]]% set finalizescript /opt/cray/esms/cray-es-finalize-scripts-XX /etc/site.mds_finalize.sh [esms1->category*[esFS-MDS*]]% commit [esms1->category[esFS-MDS]]% 6. Verify the new lines have been added to the site.mds_finalize.sh finalize script. [esms1->category[esFS-MDS]]% get finalizescript [esms1->category[esFS-MDS]]% quit 204 S–2327–B CLFS Administration Tasks [5] 5.13 Configuring Node Finalize Script for OSS Nodes Prepare the node finalize script for the OSS nodes and load it for the esFS-OSS category. Important: Files created or modified by a finalize script must be listed in the excludelistupdate exclude list for the category. Software updates will over write customized files if the files are not specified in an exclude list for the category. Customized files must also be specified in the excludelistgrab, and excludelistgrabnew exclude lists to prevent customized files from being copied to the CIMS. Procedure 64. Configuring the node finalize script for OSS nodes 1. Log in to the CIMS as root. 2. Make a copy of the default OSS finalize script to an etc directory to prevent it from being overwritten during software updates. esms1# cd /opt/cray/esms/cray-es-finalize-scripts-XX esms1# mkdir -p etc esms1# cp -p default/oss_finalize.sh etc/site.oss_finalize.sh 3. (Optional) If the OSS node connects to the storage array using InfiniBand® (IB), then start a subnet manager and configure SCSI RDMA protocol (SRP) on the interface that is connected to the storage array controllers. a. Edit the site.oss_finalize.sh file add/uncomment the following lines before the exit 0 line. esms1# vi etc/site.oss_finalize.sh b. Add/uncomment the following lines to disable IB cards which are not present. sed -i -e "s/MTHCA_LOAD=yes/MTHCA_LOAD=no/" /localdisk/etc/infiniband/openib.conf sed -i -e "s/QIB_LOAD=yes/QIB_LOAD=no/" /localdisk/etc/infiniband/openib.conf sed -i -e "s/MLX4_EN_LOAD=yes/MLX4_EN_LOAD=no/" /localdisk/etc/infiniband/openib.conf sed -i -e "s/CXGB3_LOAD=yes/CXGB3_LOAD=no/" /localdisk/etc/infiniband/openib.conf sed -i -e "s/NES_LOAD=yes/NES_LOAD=no/" /localdisk/etc/infiniband/openib.conf c. Add/uncomment these lines to configure SRP. # configure SRP for IB connected RAID sed -i -e "s/SDP_LOAD=yes/SDP_LOAD=no/" /localdisk/etc/infiniband/openib.conf sed -i -e "s/SRP_LOAD=no/SRP_LOAD=yes/" /localdisk/etc/infiniband/openib.conf sed -i -e "s/SRP_DAEMON_ENABLE=no/SRP_DAEMON_ENABLE=yes/" /localdisk/etc/infiniband/openib.conf Note: Cray recommends that ib0 on OSS nodes be connected to the IB switch, ib1 is should not be used, and ib2 and ib3 connect to the storage array controllers. d. Add/uncomment the following command line to the end of the S–2327–B 205 Data Management Platform (DMP) Administrator's Guide site.oss_finalize.sh script (before exit 0 line), to start the IB subnet manager on ib2 (the IB interface connected to the storage array). Change the 2 to designate the IB interface port connected to the storage array. echo "/usr/sbin/opensm --daemon -g /` usr/sbin/ibstat mlx4_1 2 | grep GUID | awk '{print $3}' "` >> /localdisk/etc/rc.d/rc.local exit 0 (Optional) If both ports of the IB card (mlx4_1 in this example) are connected to storage, then add (or uncomment) a separate line for each port: echo "/usr/sbin/opensm --daemon -g /` usr/sbin/ibstat mlx4_1 1 | grep GUID | awk '{print $3}' "` >> /localdisk/etc/rc.d/rc.local echo "/usr/sbin/opensm --daemon -g /` usr/sbin/ibstat mlx4_1 2 | grep GUID | awk '{print $3}' "` >> /localdisk/etc/rc.d/rc.local exit 0 4. Run the cmsh and set the finalize script for the esFS-OSS category. esms1# cmsh [esms1]% category [esms1->category]% use esFS-OSS [esms1->category[esFS-OSS]]% set finalizescript /opt/cray/esms/cray-es-finalize-scripts-XX/etc/site.oss_finalize.sh [esms1->category*[esFS-OSS*]]% commit [esms1->category[esFS-OSS]]% 5. Verify the site.oss_finalize.sh finalize script is correct (or, if you added lines in step 3, verify they are present). [esms1->category[esFS-OSS]]% get finalizescript [esms1->category[esFS-OSS]]% quit 6. Verify the lustre service is running for the ESF software image. a. Use the chroot shell to open a shell in the CLFS image and verify the lustre is enabled. If the lustre is not enabled, start it with the chkconfig lustre on command. esms1# chroot /cm/images/ESF-XX-2.1.0-201309252230 [root@esms1 /]# chkconfig --list lustre lustre 0:off 1:off 2:on 3:on 4:on 5:on 6:off [root@esms1 /]# exit esms1# b. Reboot the MDS and OSS nodes with the ESF software image. 5.14 Configure LDAP on MDS Nodes LDAP is configured in the category finalize script for the MDS node. 206 S–2327–B CLFS Administration Tasks [5] Important: Files created or modified by a finalize script must be listed in the excludelistupdate exclude list for the category. Software updates will over write customized files if the files are not specified in an exclude list for the category. Customized files must also be specified in the excludelistgrab, and excludelistgrabnew exclude lists to prevent customized files from being copied to the CIMS. Procedure 65. Configure LDAP on MDS nodes 1. Log in to the CIMS as root. 2. Edit the site.mds_finalize.sh script. esms1# cd /opt/cray/esms/cray-es-finalize-scripts-XX/etc esms1# vi site.mds_finalize.sh 3. Enable the name service caching daemon (nscd) and the LDAP name service daemon nslcd. Note: named, nslcd, nscd, and ldap are turned off in the default CLFS image. Add the following commands to the site.mds_finalize.sh file: chkconfig nscd on service nscd start chkconfig nslcd on service nslcd start 4. Save site.mds_finalize.sh file and exit the editor. 5. Use the chroot shell to edit the CLFS software image (in this example, the software image is named ESF-XX-2.1.0-201309252230): esms1# cd /cm/images esms1# chroot ESF-XX-2.1.0-201309252230 [root@esms1 /]# 6. Verify the LDAP service (ldap), is turned off in the CLFS software image. Note: The MDS should not act as an LDAP server and the following command disables the LDAP server instance. The LDAP client pieces are typically in /etc/passwd and /etc/nsswitch.conf, and in the pluggable authentication modules (PAM). [root@esms1 /]#chkconfig ldap off 7. Edit /etc/openldap/ldap.conf file, and uncomment (enable) the TIMELIMIT 15 line: [root@esms1 /]#cd /etc/openldap [root@esms1 /]#cp ldap.conf ldap.conf.orig [root@esms1 /]#vi ldap.conf TIMELIMIT 15 S–2327–B 207 Data Management Platform (DMP) Administrator's Guide 8. Add/edit the following lines and enter the specific IP and base value settings for your site's information structure: URI ldap://aaa.bbb.ccc.ddd/ BASE dc=somedomain,dc=somedomain,dc=com base ou=people,dc=somedomain,dc=somedomain,dc=com 9. Edit /etc/nslcd.conf file: [root@esms1 /]#cd /etc [root@esms1 /]#cp nslcd.conf nslcd.conf.orig [root@esms1 /]#vi nslcd.conf 10. Add/edit the following lines and enter the specific IP and base value settings for your site's information structure: uri ldap://aaa.bbb.ccc.ddd/ base dc=somedomain,dc=somedomain,dc=com base ou=people,dc=somedomain,dc=somedomain,dc=com 11. Exit the chroot shell. 12. Reboot all MDS nodes using the ESF-XX-2.1.0-201309252230 software image. 5.15 Configuring the Site User Network (for MDS Nodes Only) Note: Configure the external site user network (site-user-net) which is used by CLFS MDS nodes for authentication services (LDAP, for example) and file permissions (there are no user login accounts on CLFS nodes). Procedure 66. Configuring the site user network (for MDS nodes only) 1. Log in to the CIMS as root and run cmsh. esms1# cmsh [esms1]% 2. Switch to network mode: [esms1]% network [esms1->network]% 3. Display the existing networks. [esms1->network]% list Name (key) Type Netmask bits Base address Domain name IPv6 ---------------- ------------ ---------------- ---------------- ------------------ ---- esmaint-net Internal 16 10.141.0.0 esmaint-net.cluster no globalnet Global 16 0.0.0.0 cm.cluster no ib-net Internal 16 10.149.0.0 ib-net.cluster no ipmi-net Internal 16 10.148.0.0 ipmi-net.cluster no site-admin-net External 24 aaa.bbb.ccc.ddd your.domain.com no 4. Determine whether site-user-net exists on the CIMS a. If a site-user-net already exists on the CIMS, proceed to step 5. 208 S–2327–B CLFS Administration Tasks [5] b. If there is no site user network, create site-user-net by cloning the site-admin-net network. [esms1->network]% clone site-admin-net site-user-net [esms1->network*[site-user-net*]% c. The cloned network inherits the same settings as the original network (site-admin-net). You must change several settings for the new site-user-net: base address, broadcast address, domain name, gateway, and (if necessary) netmask bits. Display the existing settings. [esms1->network*[site-user-net*]% show Parameter Value ------------------------------ ---------------------- Base address aaa.bbb.ccc.ddd Broadcast address aaa.bbb.255.255 Domain Name your.domain.com Dynamic range end 0.0.0.0 Dynamic range start 0.0.0.0 Gateway aaa.bbb.ccc.ddd IPv6 no Lock down dhcpd no MTU 1500 Management allowed no Netmask bits 24 Node booting no Notes Revision Type External name site-user-net d. Change the base address. [esms1->network*[site-user-net*]% set baseaddress site-user-netBaseAddress e. Change the broadcast address. [esms1->network*[site-user-net*]% set broadcastaddress site-user-netBroadcastAddress f. Change the domain name. [esms1->network*[site-user-net*]% set domainname site-user-netDomainName g. Change the gateway. [esms1->network*[site-user-net*]% set gateway site-user-netGateway h. If necessary, change the netmask bits. [esms1->network*[site-user-net*]% set netmaskbits NN i. Commit your changes. [esms1->network*[site-user-net*]% commit 5. Switch to device mode. [esms1->network[site-user-net]% device S–2327–B 209 Data Management Platform (DMP) Administrator's Guide 6. Add an interface to the site-user-net network. This example shows the host name esfs-mds001, the Ethernet port eth1, and the example IP address aaa.bbb.ccc.ddd. Substitute your CLFS node's host name and IP address when configuring the eth1 interface. [esms1->device]% addinterface -n esfs-mds001 physical eth1 site-user-net aaa.bbb.ccc.ddd Note: You must repeat this step for each CLFS node that is added to the site-user-net network. 7. Commit your changes. [esms1->device*]% commit 8. Show the interfaces on esfs-mds001. [esms1->device]% use esfs-mds001 [esms1->device[esfs-mds001]% interfaces; list Type Network device name IP Network ------------ -------------------- ---------------- ---------------- bmc ipmi0 10.148.0.2 ipmi-net physical BOOTIF [prov] 10.141.0.2 esmaint-net physical eth1 aaa.bbb.ccc.ddd site-user-net physical ib0 10.149.0.2 ib-net 9. Display the existing networks. [esms1->device[esfs-mds001]->interfaces]% network list Name (key) Type Netmask bits Base address Domain name IPv6 ---------------- ------------ ---------------- ---------------- -------------------- ---- esmaint-net Internal 16 10.141.0.0 esmaint-net.cluster no globalnet Global 16 0.0.0.0 cm.cluster no ib-net Internal 16 10.149.0.0 ib-net.cluster no ipmi-net Internal 16 10.148.0.0 ipmi-net.cluster no site-admin-net External 24 aaa.bbb.0.0 your.domain.com no site-user-net External 24 aaa.bbb.0.0 your.domain.com no 10. Exit interface mode cmsh. [esms1->device[esfs-mds001]->interfaces]% exit [esms1->device[esfs-mds001] 11. Test boot the new node. [esms1->device[esfs-mds001] reboot 5.16 Mounting a CLFS Lustre® File System on a Cray CLE System Procedure 67. Mounting a Lustre file system on a CLE system 1. Add a node_class entry for LNET router nodes to CLEinstall.conf. The node_class entry will be added to /etc/hosts when CLEinstall is run. In this example, the LNET routers node IDs (NIDs) are 2 and 30. a. Log in to the SMW as root. 210 S–2327–B CLFS Administration Tasks [5] b. Edit the /home/crayadm/install.xtrel/CLEinstall.conf file so that there will be an lnet class created with nid 2 and 30 as members of that class. After running CLEinstall, the new class and its members will be in /etc/opt/cray/sdb/node_classes on the bootroot and sharedroot file systems. smw# vi /home/crayadm/install.xtrel/CLEinstall.conf node_class[1]=lnet 2 30 c. Save the file and exit. 2. Run CLEinstall using the same command line options as if you were performing a software update. Running CLEinstall makes the configuration change to the node_classes file. After running CLEinstall, boot the CLE system before continuing to step 3. 3. Make a mount point in the default view of the sharedroot. smw# ssh boot boot# xtopview default/:/ # mkdir -p /lus/scratch 4. Add modprobe.conf.local for login nodes and all service nodes. Note: Notice that the line containing options lnet networks=gni is commented. The setting of 10.149.1.* is the site IB network address to be used, but the network must be within 10.149.0.0/16. default:/ # vi /etc/modprobe.conf.local #options lnet networks=gni options lnet ip2nets="gni0 10.128.*.*; o2ib 10.149.*.*" options lnet routes="gni0 10.149.1.[3,31]@o2ib; o2ib [2,30]@gni0" ### LNET options options lnet check_routers_before_use=1 options lnet avoid_asym_router_failure=1 options lnet dead_router_check_interval=60 options lnet live_router_check_interval=60 options lnet router_ping_timeout=50 5. Exit xtopview. default:/ # exit 6. Create the lnet class-specialized files. boot-p1# xtopview -c lnet class/lnet:/ # ls -l /etc/modprobe.conf.local lrwxrwxrwx 1 root root 45 Feb 1 08:53 /etc/modprobe.conf.local -> /.shared/base/default/etc/ modprobe.conf.local class/lnet:/ # xtspec -c lnet /etc/modprobe.conf.local class/lnet:/ # ls -l /etc/modprobe.conf.local lrwxrwxrwx 1 root root 48 Feb 1 09:06 /etc/modprobe.conf.local -> /.shared/base/class/lnet/ etc/modprobe.conf.local class/lnet:/ # vi /etc/modprobe.conf.local S–2327–B 211 Data Management Platform (DMP) Administrator's Guide Add the local extensions to the /etc/modprobe.conf.local file. The two examples below assume that ib0 on the LNET router nodes is connected to the DMP IB switch. If ib1 is used, there is a different format. # # please add local extensions to this file # with ib0 ### LNET interfaces options lnet ip2nets="gni0 10.128.*.*; o2ib 10.149.*.*" ### LNET routes for esFS options lnet routes="gni0 10.149.1.[3,31]@o2ib; o2ib [2,30]@gni0" with ib1 ### LNET interfaces options lnet ip2nets="gni0 10.128.*.*; o2ib(ib1) 10.149.*.*" ### LNET routes for esFS options lnet routes="gni0 10.149.1.[3,31]@o2ib; o2ib(ib1) [2,30]@gni0" 7. Create node-specialized files for all LNET routers. Repeat steps step 7.a through step 7.d for each LNET router. a. Run xtopview to get the node view of the sharedroot for NID 2. boot# xtopview -n 2 b. Create and edit the interface file (example shows ifcfg-ib0) for NID 2 with IP address 10.149.0.2. This IP address is the IP address of the esfs-mds001 node on the ib-net. node/2:/ # touch /etc/sysconfig/network/ifcfg-ib0 node/2:/ # xtspec -n 2 /etc/sysconfig/network/ifcfg-ib0 node/2:/ # vi /etc/sysconfig/network/ifcfg-ib0 An example ifcfg-ib0 file follows: BOOTPROTO='static' IPADDR=10.149.1.3 NETMASK=255.255.0.0 STARTMODE='onboot' USERCONTROL='no' MTU=2044 IPOIB_MODE='connected' c. Create and edit the /etc/sysconfig/infiniband file. node/2:/ # touch /etc/sysconfig/infiniband node/2:/ # xtspec -n 2 /etc/sysconfig/infiniband node/2:/ # vi /etc/sysconfig/infiniband Change the following two variables: ONBOOT=no SRP_LOAD=yes to: ONBOOT=yes SRP_LOAD=no 212 S–2327–B CLFS Administration Tasks [5] d. Exit xtopview. node/2:/ # exit boot# 8. Update the bootimage so that compute nodes mount the Lustre file system. a. Update the bootimage /etc/modprobe.conf. This example uses the p1 partition of a CLE system. smw# cd /opt/xt-images/templates/default-p1 smw:/opt/xt-images/templates/default-p1 # vi etc/modprobe.conf Make the following changes to the modprobe.conf file. #options lnet networks=gni ### LNET options options lnet check_routers_before_use=1 options lnet avoid_asym_router_failure=1 options lnet dead_router_check_interval=60 options lnet live_router_check_interval=60 options lnet router_ping_timeout=50 ### LNET interfaces options lnet ip2nets="gni0 10.128.*.*; o2ib 10.149.*.*" ### LNET routes for esFS options lnet routes="gni0 10.149.1.2@o2ib; o2ib 1@gni0" b. Make a mount point for the bootimage. smw:/opt/xt-images/templates/default-p1 # mkdir -p lus/scratch c. Update /etc/fstab for bootimage. smw:/opt/xt-images/templates/default-p1 # cd etc smw:/opt/xt-images/templates/default-p1/etc # cp -p fstab fstab.orig smw:/opt/xt-images/templates/default-p1/etc # vi fstab Add the following line to /etc/fstab. 10.149.0.2@o2ib:/scratch /lus/scratch lustre rw,flock Exit and save the /etc/fstab file. d. Rebuild the bootimage. This example uses a script created for the BLUE system set. smw# /var/opt/cray/install/shell_bootimage_BLUE.sh -c 9. Update the esFS-MDS and esFS-OSS category finalize scripts. a. Log in to the CIMS as root. b. Edit the site.mds_finalize.sh script. esms1# cd /opt/cray/esms/cray-es-finalize-scripts-XX/etc esms1# vi site.mds_finalize.sh S–2327–B 213 Data Management Platform (DMP) Administrator's Guide c. Add the entry before the exit 0 line. This example uses the IP address 10.149.1.3 for NID 2 and 10.149.1.31 for NID 30. echo "options lnet routes="gni0 10.149.1.[3,31]@o2ib" >> /localdisk/etc/modprobe.d/cray.conf step 9.c requires that /etc/modprobe.d/cray.conf file exist in the CLFS node image. You may need to create (touch) the /etc/modprobe.d/cray.conf file to create it in the CLFS node image. d. Exit and save the site.mds_finalize.sh script. e. Update the esFS-MDS category with the new finalize script. esms1# cmsh [esms1]% category [esms1->category]% use esFS-MDS [esms1->category[esFS-MDS]]% set finalizescript /opt/cray/esms/cray-es-finalize-scripts-XX/ etc/site.mds_finalize.sh [esms1->category*[esFS-MDS*]]% commit f. Verify the finalize script. [esms1->category[esFS-MDS]]% get finalizescript g. Repeat step 9 to update the esFS-OSS category finalize script. h. Quit cmsh. [esms1->category[esfs-mds]]% quit 10. Unmount the Lustre clients and stop the Lustre file system. a. Unmount CDL Lustre clients. eslogin1# umount /lus/scratch b. Stop Lustre file system. esms1# lustre_control status -f scratch esms1# lustre_control stop -f scratch 11. Reboot MDS and OSS node(s). esms1# cmsh esms1# device esms1# foreach esfs-mds001,esfs-mds002,esfs-oss1,esfs-oss002,esfs-oss003,esfs-oss004 (reboot) esfs-mds001: Reboot in progress ... esfs-mds002: Reboot in progress ... Note: Wait for each all MDS and OSS nodes to reboot before continuing. Use Bright cmgui to open remote consoles for each node to monitor to reboot process. 214 S–2327–B CLFS Administration Tasks [5] 12. Exit cmsh and start the Lustre file system. [esms1]% exit esms1# lustre_control start -f scratch Performing 'start' from esms1 at Wed Apr 24 16:06:55 CDT 2013 Operating on file system - "scratch" Verifying network connectivity between "esms1" and "esfs-mds001"... . . . 13. Either configure LDAP on MDS nodes or execute the following command on the esfs-mds001 node after starting Lustre file system to disable upcall and make the MDS use the UID/GID supplied by the client. Refer to Configure LDAP on MDS Nodes on page 206 for more information about configuring LDAP after completing this procedure. esms1# ssh esfs-mds001 esf1-mds001# echo NONE > /proc/fs/lustre/mdt/scratch-MDT0000/identity_upcall 14. Exit to the CIMS. esf1-mds001# exit esms1# 15. Use ssh to log in to eslogin1 and mount the Lustre client. esms1# ssh eslogin1 eslogin1# mount -t lustre -o rw,flock,lazystatfs 10.149.0.2@o2ib:/scratch /lus/scratch 16. Reboot Cray CLE system. The compute nodes will mount the Lustre file system as specified in the boot image /etc/fstab file. 17. Mount the Lustre file system on the Cray CLE login node. Note: Include this mount command in the /etc/fstab file. smw# ssh boot boot# ssh login login# mount -t lustre -n -o rw,flock,lazystatfs 10.149.0.2@o2ib:/scratch /lus/scratch 5.17 Configuring Multipath on CLFS nodes All CLFS nodes configured for multipath have a connection to each storage array controller. Additional configuration is required to support multipath. Specific WWID values must be added to the multipaths {} section of the /etc/multipath.conf file to make sure that devices have the same name every time the node boots. Note: If the node is not cabled for multipath, then do not need to perform the multipath configuration described in Procedure 68 on page 216. S–2327–B 215 Data Management Platform (DMP) Administrator's Guide Procedure 68. Configuring multipath on CLFS nodes 1. After ESFinstall has completed, the multipathd service must be enabled for the software image. esms1# chroot /cm/images/ESFimage [root@esms1/]# chkconfig multipathd on 2. Copy multipath.conf.cray to /etc in the ESF software image. esms1# cp -p /opt/cray/esfs/cray-esf-multipath-XX /default/etc/multipath.conf.cray /etc/multipath.conf 3. Exit the chroot shell. esms1# exit esms1# 4. Identify the disk names for multipath on the MDS nodes. Boot the MDS 1 node and use SSH to login. esms1# ssh mds001 Last login: Tue Jun 11 15:42:52 2013 from esms1.cm.cluster [root@mds001 ~]# 5. Use multipath command to display the disk devices. mds001# multipath -ll mpatha (360080e50002f82ca000002ca5102bcc4) dm-0 LSI,INF-01-00 size=2.2T features='2 pg_init_retries 50' hwhandler='1 rdac' wp=rw |-+- policy='round-robin 0' prio=6 status=active | -` 0:0:0:0 sda 8:0 active ready running ` +- policy='round-robin 0' prio=1 status=enabled - ` 0:0:1:0 sdb 8:16 active ghost running - 6. In this example, the WWID of 360080e50002f82ca000002ca5102bcc4 is the device which we want to label as mdt0 instead of mpatha. 216 S–2327–B CLFS Administration Tasks [5] 7. On some systems, the disk device may need to be modified using the parted command to remove unneeded partitions. Identify which disk device (/dev/sda form) corresponds to this WWID. mds001# ls -l /dev/disk/by-id | grep scsi-360080e50002f82ca000002ca5102bcc4 lrwxrwxrwx 1 root root 9 Jun 10 09:28 scsi-360080e50002f82ca000002ca5102bcc4 -> ../../sda [root@mds001 ~]# parted /dev/sda GNU Parted 2.1 Using /dev/sda Welcome to GNU Parted! Type 'help' to view a list of commands. (parted) print Model: LSI INF-01-00 (scsi) Disk /dev/sda: 2398GB Sector size (logical/physical): 512B/512B Partition Table: gpt Number Start End Size File system Name Flags 1 17.4kB 20.5GB 20.5GB ext3 / 2 20.5GB 22.5GB 2048MB ext3 /var 3 22.5GB 24.6GB 2048MB ext3 /tmp 4 24.6GB 41.0GB 16.4GB linux-swap(v1) /dev/sda4 5 41.0GB 2398GB 2357GB ext3 /local 8. Remove all partitions so that Lustre® can use the entire device. This example pauses after removing partitions 2 through 5 to show that only one partition is left. (parted) rm 5 (parted) rm 4 (parted) rm 3 (parted) rm 2 (parted) p Model: LSI INF-01-00 (scsi) Disk /dev/sda: 2398GB Sector size (logical/physical): 512B/512B Partition Table: gpt Number Start End Size File system Name Flags 1 17.4kB 20.5GB 20.5GB ext3 / (parted) rm 1 (parted) p Error: /dev/sda: unrecognised disk label (parted) quit Information: You may need to update /etc/fstab. Note: The MDS nodes are ready for a reboot after step 16 below is complete. 9. Exit the MDS node SSH login and repeat step 4 for the other MDS node to identify the disk names for multipath. 10. Identify the disk names for multipath on the OSS nodes. Boot each OSS node and use SSH to login. esms1# ssh oss001 Last login: Tue Jun 11 15:45:32 2013 from esms1.cm.cluster [root@oss001 ~]# S–2327–B 217 Data Management Platform (DMP) Administrator's Guide 11. Display the disk devices with the multipath command. [root@oss001 ~]# multipath -llmpatha (360080e50001f8c64000000b4513098b2) dm-2 LSI,INF-01-00 size=7.3T features='3 queue_if_no_path pg_init_retries 50' hwhandler='1 rdac' wp=rw |-+- policy='round-robin 0' prio=6 status=active | -` 7:0:0:5 sdg 8:96 active ready running ` +- policy='round-robin 0' prio=1 status=enabled - ` 8:0:0:5 sdm 8:192 active ghost running - mpathb (360080e50001f87c8000000b951309893) dm-3 LSI,INF-01-00 size=7.3T features='3 queue_if_no_path pg_init_retries 50' hwhandler='1 rdac' wp=rw |-+- policy='round-robin 0' prio=6 status=active | -` 8:0:0:4 sdl 8:176 active ready running ` +- policy='round-robin 0' prio=1 status=enabled - ` 7:0:0:4 sdf 8:80 active ghost running - mpathc (360080e50001f8c64000000ae51309849) dm-1 LSI,INF-01-00 size=7.3T features='3 queue_if_no_path pg_init_retries 50' hwhandler='1 rdac' wp=rw |-+- policy='round-robin 0' prio=6 status=active | -` 7:0:0:3 sde 8:64 active ready running ` +- policy='round-robin 0' prio=1 status=enabled - ` 8:0:0:3 sdk 8:160 active ghost running - . . . 12. On some systems, the disk devices may need to be modified with the parted command to remove unneeded partitions. Identify which disk device (/dev/sda form) relates to this WWID. [root@oss001 ~]# ls -l /dev/disk/by-id | grep scsi-360080e50001f8c64000000b4513098b2 lrwxrwxrwx 1 root root 9 Jun 3 14:40 scsi-360080e50001f8c64000000b4513098b2 -> ../../sdm [root@oss001 ~]# parted /dev/sdm GNU Parted 2.1 Using /dev/sda Welcome to GNU Parted! Type 'help' to view a list of commands. (parted) print Model: LSI INF-01-00 (scsi) Disk /dev/sdm: 2398GB Sector size (logical/physical): 512B/512B Partition Table: gpt Number Start End Size File system Name Flags 1 17.4kB 20.5GB 20.5GB ext3 / 2 20.5GB 22.5GB 2048MB ext3 /var 3 22.5GB 24.6GB 2048MB ext3 /tmp 4 24.6GB 41.0GB 16.4GB linux-swap(v1) /dev/sda4 5 41.0GB 2398GB 2357GB ext3 /local 218 S–2327–B CLFS Administration Tasks [5] 13. Remove all partitions so that Lustre can use the entire device. This example pauses after removing partitions 2 through 5 to show that only one partition is left. (parted) rm 5 (parted) rm 4 (parted) rm 3 (parted) rm 2 (parted) p Model: LSI INF-01-00 (scsi) Disk /dev/sdm: 2398GB Sector size (logical/physical): 512B/512B Partition Table: gpt Number Start End Size File system Name Flags 1 17.4kB 20.5GB 20.5GB ext3 / (parted) rm 1 (parted) p Error: /dev/sdm: unrecognised disk label (parted) quit Information: You may need to update /etc/fstab. 14. Exit the OSS node SSH login and repeat step 10 for the other OSS nodes to identify the disk names for multipath. 15. The OSS nodes are ready for a reboot after step 16 below is complete. 16. After the disk names for multipath have been identified, edit the comment section near the end of the node finalize scripts for both esFS-MDS esFS-OSS categories. Adding specific WWID values to the multipaths {} section of the /etc/multipaths.conf file ensures that the device has the same name every time the node boots. Also, you can specify a label for each disk, such as /dev/mapper/ost5 or /dev/mapper/mdt0, then use that label for the Lustre configuration file. cat << EOMP >> /localdisk/etc/multipath.conf multipaths { multipath { wwid 360080e500017c6e40000093e51af554f alias mgt } multipath { wwid 360080e500017c57200000b0451af54ba alias mdt } } EOMP 17. Reboot all MDS and OSS nodes which were configured with multipath. 18. Confirm that the MDT device /dev/mapper/mdt0 is available on the MDS nodes. esms1# ssh mds001 fdisk -l /mapper/mdt0 S–2327–B 219 Data Management Platform (DMP) Administrator's Guide 19. Confirm that all of the OST devices /dev/mapper/ost* devices are available on the OSS nodes. esms1# ssh oss001 fdisk -l /dev/mapper/ost0 esms1# ssh oss001 fdisk -l /dev/mapper/ost1 esms1# ssh oss001 fdisk -l /dev/mapper/ost2 esms1# ssh oss001 fdisk -l /dev/mapper/ost3 esms1# ssh oss001 fdisk -l /dev/mapper/ost4 esms1# ssh oss001 fdisk -l /dev/mapper/ost5 20. When configuring the Lustre fs_defs file, the disk device name "/dev/mapper/mdt0" can be used for the MDT and MGT and "/dev/mapper/ost0" can be used for OST0. Similarly for the other OST devices, but the example below shows how to have /dev/mapper/ost[0-5] reference the appropriate devices with even numbered OST devices on oss002 and odd numbered OST devices on oss001. ## MDT ## MetaData Target mdt: node=mds001 dev=/dev/mapper/mdt0 fo_node=mds002 ## MGT ## Management Target mgt: node=mds001 dev=/dev/mapper/mdt0 fo_node=mds002 ## Object Storage Target(s) ost: node=oss00[2,1] dev=/dev/mapper/ost[0-5] fo_node=oss00[1,2] 5.18 Migrating from Lustre® 1.8.x to 2.4 ! Caution: The CIMS must be running the most recent released version of ESM software (ESM-XX-2.1.0) in order to migrate from Lustre 1.8 to 2.4. This document details the format differences between Lustre 1.8.x and 2.4, their purpose, and the process of upgrading from 1.8.x to 2.4. 5.18.1 Related Publications The following documents contain additional information that may be helpful: • Managing Lustre for the Cray Linux Environment (CLE) (S–0010) • Installing Lustre File System by Cray (CLFS) Software (S–2521) Note: The Bright administration guide and user guides are stored on the CIMS node (as Adobe® Acrobat® PDF files) in the /cm/shared/docs/cm directory. 220 S–2327–B CLFS Administration Tasks [5] 5.18.2 Introduction Lustre 2.4 represents a significant advance in Lustre design with the addition of many new features and support for future improvements. To accommodate the new features, 2.4 uses a somewhat different on-disk file system format than the one used by 1.8.x. The format differences are limited to Lustre's internal metadata about the file system. They do not affect the format or storage of user data. The format differences fall into two categories: those related to file identifiers (FIDs) that replace inodes in some cases and those related to quota support. Lustre 2.4 provides tools to add the new metadata structures used by 2.4 to an existing 1.8.x file system. Some of these tools run automatically when the 2.4 servers are started; some require the administrator to perform explicit upgrade operations at the time 2.4 is installed. 5.18.3 FIDS A file identifier (FID), is a unique identifier for a Lustre file or object. It is independent of the back end file system, for example, ldiskfs. In Lustre 1.8.x, inodes are used to uniquely identify the objects belonging to a file. In Lustre 2.4, FIDs replace inodes for this purpose. The drawback of using inodes is that a file's inode can change over the life of the file. For example, when a file is restored from a file level backup, it will be assigned a new inode/generation number. Afterwards, Lustre 1.8.x can no longer locate its objects. FIDs, on the other hand, never change once assigned. Following restore from a file level backup, the FIDs are still correct and Lustre 2.4 can locate its objects. Lustre 2.4, in addition to supporting the device level backup of 1.8.x, also supports file level backup and restore. Lustre 2.4 still uses inodes internally to interact with the ldiskfs back end file system. To facilitate this interaction, Lustre 2.4 maintains a map of FIDs to inodes. This map is called the Object Index (OI). When upgrading from 1.8.x to 2.4 the OI must be created through a process called OI Scrub. The OI Scrub occurs automatically when a 1.8.x formatted file system without an OI is mounted by 2.4 servers. An OI Scrub may also be triggered if Lustre discovers a missing or bad FID to inode mapping during normal operation. Finally, an OI Scrub can be started manually by running lfsck. lfsck checks and repairs errors in the OI. The OI maintains the FID to inode mapping, the inode to FID mapping is stored in the inode itself. The FID of the object identified by the inode is stored in the extended attributes area of the inode known as linkEA. The inode also contains the FID of the file's parent directory. This feature is known as FID-in-dirent. The linkEA and FID-in-dirent information enables Lustre to efficiently generate full path names from the inode. These names are used in POSIX style path name permission checks, to produce better error messages, and to support changelog applications like lustre_rsync. S–2327–B 221 Data Management Platform (DMP) Administrator's Guide Storing the parent FID in the inode also improves the performance of readdir and other directory operations. Unlike the OI, which Lustre 2.4 requires, the FID information in the inode is optional. If the FID information in the inode is missing, then directory lookup performance is affected and changelog features will not be fully supported, but otherwise Lustre will be totally functional. The FID information is optional, and the upgrade process to add FID information to the inode is also optional. To populate the inodes with the appropriate FIDs, the Lustre administrator must set the dirdata attribute on each MDT and then run lfsck with the -t namespace option. The lfsck process runs in the background; the rate at which it updates inodes can be tunable parameter under the control of the system administrator. After the dirdata attribute is set, the Lustre file system cannot be downgraded to work with 1.8.x servers. 5.18.4 Quota Support Lustre 2.4 addresses several limitations of the previous quota design. Among the improvements to quotas in 2.4 are: • Quota limits can be changed while slaves are offline • OSTs can be added and deleted without corrupting space usage statistics • Master recovery can be completed without all targets being online • A full quotacheck is no longer required following e2fsck • Quota enforcement is enabled/disabled by file system rather than per-target • Infrastructure is restructured for future growth, better performance, and improved functionality To support these improvements, Lustre 2.4 has changed both the on-disk format of quota information and the user interface to quota functionality. Quota specification has three components: space usage accounting, quota limit definition, and enabling enforcement. Space Usage Accounting: In previous versions of Lustre, the lfs quotacheck must be run to generate the database of space used on each target by each user and group. In Lustre 2.4, the quotacheck command is deprecated. Instead, newly formatted 2.4 file systems have space usage accounting enabled by default, and the statistics are automatically kept up to date. When upgrading from 1.8.x to 2.4, the usage statistics must be initialized for existing files before quota limits can be enforced. The statistics are generated by running tunefs.lustre --quota on each target. This tunefs command also sets the QUOTA attribute of the target, which enables automatic accounting when the target is mounted. 222 S–2327–B CLFS Administration Tasks [5] Initializing usage statistics is a one time operation. It can be done as part of the upgrade process or sometime later. Note however that quota enforcement and accounting are disabled until the tunefs.lustre --quota command is executed on all targets. Quota Limit Definition: The definition of user and group quota limits does not change with 2.4. The storage format does change. Prior to Lustre 2.4, the quota limits are stored in a file specific to the back end file system. With 2.4, this information is moved in a Lustre defined index along with other Lustre metadata. The quota limits are converted automatically to the new format and storage location when the MDT is upgraded to 2.4. Enabling Quota Enforcement: In Lustre 2.4, quota enforcement is independent of the space usage accounting. The accounting information is always maintained, even when enforcement is disabled. Enforcement is enabled/disabled for the entire file system. The command is: lctl _param fsname/quota/ost | mdt=u | g | ug | none The lfs quotaon|off command and per-target quota_type parameter are no longer used in Lustre 2.4. 5.18.5 Performance Expectations For optimum long term performance and functionality, all of the disk format changes described above are recommended. However, each of the upgrade processes has a performance cost. 5.18.5.1 Object Index Creation and Repair OI Scrub is designed to have minimal impact on system performance. It runs in the background while the file system remains online. Clients can continue to access files while it runs. The system administrator can control the overhead of the scrubbing process by tuning the maximum number objects examined per second. If no limit is set, OI Scrub will run as fast as possible. On an unloaded system, with no limit set, experiments have shown OI Scrub to process in excess of 100,000 objects per second. OI Scrub status can be monitored through the /proc file: osd-ldisk/mdt_device/oi_scrub 5.18.5.2 Adding FIDs to inodes Updating all the inode extended attributes to include the related FID and parent FID information is a one time operation. (Note, updates to individual inodes may occur when lfsck repairs file system corruption.) The process is similar to an explicitly invoked OI Scrub and has similar performance characteristics. S–2327–B 223 Data Management Platform (DMP) Administrator's Guide Progress of the inode update can be monitored with the /proc file: mmd/mdt_device/lfsck_namespace 5.18.5.3 Space Usage Statistics When upgrading a 1.8.x formatted file system, a database of the space usage statistics must be created. The usage data is gathered and stored when the quota flag is set on each target. The speed of the data collection in 2.4 is similar to the speed of an lfs quotacheck in earlier Lustre versions. After the initial creation, Lustre updates the space usage statistics automatically as files change. Updating the statistics does impose overhead and has been reported to affect metadata performance by as much as 5%. Cray internal testing has shown that the accounting overhead has no measurable effect on performance. 5.18.6 Upgrade Procedure This procedure is scripted, and assumes the person performing this update has experience with the Cray Linux Environment (CLE) and Lustre administration. Procedure 69. Upgrading Lustre 1.8.x to Lustre 2.4 1. If you do not already have an .fs_defs file, create one that defines the structure of the 1.8.x file system. See Managing Lustre for the Cray Linux Environment (CLE), Section 2.2. a. Use umount to unmount Lustre file system on all clients. b. Stop Lustre servers. c. Install Lustre 2.4 RPMs on all CLFS servers. d. Log in to the CLFS node. Do NOT start Lustre yet. e. If the .fs_defs file has not been installed, do so now, being very careful not to start Lustre. From the CIMS, enter: esms1# lustre_control install filename.fs_defs ! Caution: The following upgrade procedures require e2fsprogs version 1.42.3.wc1 or later. Check the installed version by running dumpe2fs without any parameters. Do NOT proceed with these instructions if the version is not at least 1.42.3.wc1. 2. Initial mount: Create Object Index (OI). 224 S–2327–B CLFS Administration Tasks [5] a. Regenerate configuration logs on all targets. Note: If the write_conf operation reports that a disk device was not available, repeat the write_conf operation. esms1# lustre_control write_conf -f fsname esms1# lustre_control start -p -f fsname b. Mount clients. esms1# lustre_control mount_clients -f fsname esms1# lustre_control mount_clients -c -f fsname c. Verify file system can be accessed from clients. Use ls, touch, cat, etc. of Lustre files or another procedure for verifying that a Lustre file system is operational. 3. lfsck: Add FIDs to inode attributes. ! Caution: After the dirdata attribute is set on the MDT, 1.8.x servers will no longer be able to mount the file system. Cray strongly recommends that you set the dirdata attribute to get the best performance and complete functionality from the Lustre 2.4 file system. a. Shutdown Lustre esms1# lustre_control umount_clients -c -f fsname esms1# lustre_control umount_clients -f fsname esms1# lustre_control stop -f fsname b. Set dirdata attribute on the MDT. esms1# ssh MDS_node mds# tune2fs -O dirdata MDS_node mds# exit esms1# Note: The state of the dirdata attribute can be checked before and after the tune2fs command by dumping the superblock of the MDT. mds# dumpe2fs -h MDS_node | grep 'Filesystem features' An example of attributes before the tune2fs command: dumpe2fs 1.42.7.wc1 (12-Apr-2013) [Filesystem features: has_journal ext_attr resize_inode dir_index filetype sparse_super large_file uninit_bg quota] c. Start Lustre servers. esms1# lustre_control start -f fsname S–2327–B 225 Data Management Platform (DMP) Administrator's Guide d. Update inode attributes. Note: The lfsck process performs periodic checkpoints so it can resume from where it left off if it is stopped or interrupted. lfsck progress can be monitored by watching the lfsck_namespace /proc file: esms1# ssh MDS_node mds# lctl lfsck_start -M fsname-MDT0000 -t namespace mds# lctl get_param mdd/fsname-MDT0000/lfsck_namespace mds# exit esms1# Before doing lfsck -t namespace command, the output looks like this: mdd.fsname-MDT0000.lfsck_namespace= name: lfsck_namespace magic: 0xa0629d03 version: 2 status: init flags: param: time_since_last_completed: N/A time_since_latest_start: N/A time_since_last_checkpoint: N/A latest_start_position: N/A, N/A, N/A last_checkpoint_position: N/A, N/A, N/A first_failure_position: N/A, N/A, N/A checked_phase1: 0 checked_phase2: 0 updated_phase1: 0 updated_phase2: 0 failed_phase1: 0 failed_phase2: 0 dirs: 0 M-linked: 0 nlinks_repaired: 0 lost_found: 0 success_count: 0 run_time_phase1: 0 seconds run_time_phase2: 0 seconds average_speed_phase1: 0 items/sec average_speed_phase2: 0 objs/sec real-time_speed_phase1: N/A real-time_speed_phase2: N/A current_position: N/A These values will be updated while lfsck is running. Once it has completed, the success_count value will go up by 1. 4. (Optional) Enable Quotas. ! Caution: Some small regressions in metadata performance have been attributed to the automatic usage accounting. If quota enforcement is not needed, step 4 can be skipped. Be aware however that the default configuration for Lustre 2.4 enables automatic accounting. Disabling automatic accounting, although possible, is not a documented feature. 226 S–2327–B CLFS Administration Tasks [5] a. Stop the Lustre servers. esms1# lustre_control stop -f fsname b. Enable accounting and create space usage database. Note: The lustre_control set_quota_flag operation executes tunefs.lustre --quota on each Lustre target in the specified file system. The tunefs.lustre command sets the QUOTA feature flag in the superblock of the target and runs e2fsck to build the per-UID/GID disk usage database. esms1# lustre_control set_quota_flag -f fsname 5. (Optional) OI Scrub. Note: The upgrade process performs an OI Scrub to create the FID to inode mappings automatically. The following commands to start and stop an OI Scrub are not needed to complete the upgrade, but are included here for reference. These commands are run on the MDS node. a. Start OI Scrub. mds# lctl lfsck_start -M fsname-MDT0000 b. Stop OI Scrub. mds# lctl lfsck_stop -M fsname-MDT0000 c. Tune OI Scrub speed. mds# lctl lfsck_start -M fsname-MDT0000 -s Max_Objects_Persecond d. Monitor OI Scrub status. mds# lctl get_param osd-ldiskfs/fsname-MDT0000/oi_scrub S–2327–B 227 Data Management Platform (DMP) Administrator's Guide Example of output from a small test system after the scrub has completed: name: OI_scrub magic: 0x4c5fd252 oi_files: 64 status: completed flags: param: time_since_last_completed: 711866 seconds time_since_latest_start: 711916 seconds time_since_last_checkpoint: 711866 seconds latest_start_position: 12 last_checkpoint_position: 268435457 first_failure_position: N/A checked: 628157 updated: 504 failed: 0 prior_updated: 0 noscrub: 0 igif: 0 success_count: 1 run_time: 49 seconds average_speed: 12819 objects/sec real-time_speed: N/A current_position: N/A 5.19 SCSI RDAC Driver Kernel Parameters for Fibre Channel Storage OSS and MDS slave nodes implementing Fibre Channel (FC) host bus adapters (HBAs) should use a different boot image than the OSS and MDS using SAS or IB HBAs. When booting an OSS or MDS with the CLFS software image, the scsi_dh_rdac driver is not loaded at the correct time. This causes nodes that are attached to storage via FC HBAs to encounter I/O errors, which (if enough LUNs are present) can significantly slow boot times. This condition can be corrected by pre-loading the RDAC module using a kernel parameter before the system starts the qla2xxx FC module. To create the FC software image, clone the exiting ESF software image an add the kernel parameter. The CLFS software image can be modified using the Settings tab from the cmgui. Enter rdloaddriver=scsi_dh_rdac in the Kernel Parameters field. Cray recommends this kernel parameter for servers with Fibre Channel attached storage. Procedure 70. Adding SCSI RDAC kernel parameter to ESF software image 1. Log in to the CIMS as root. 2. From a UNIX® shell, copy the functional ESF image (ESF-XX-2.1.0-201309252230), and wait for the copy operation to complete. esms1 # cd /cm/images esms1 # cp -pr ESF-XX-2.1.0-201309252230 ESF-XX-2.1.0-FC 228 S–2327–B CLFS Administration Tasks [5] 3. Start cmsh and clone the functional CLFS software image. esms1 # cmsh [esms1]% softwareimage [esms1->softwareimage]% listName (key) Path -------------------------- ---------------------------------------- ------------------------- ESF-XX-2.1.0-201309252230 /cm/images/ESF-XX-2.1.0-201309252230 2.6.32-279.14.1.el6.x86_64 ESL-XC-1.3.0 /cm/images/ESL-XC-1.3.0 3.0.74-0.6.8-default ESL-XE-2.1.0-201309042109 /cm/images/ESL-XE-2.1.0-201309042109 2.6.32.59-0.7-default default-image /cm/images/default-image 3.0.80-0.5-default default-image.previous /cm/images/default-image.previous 3.0.80-0.5-default [esms1->softwareimage]% clone ESF-XX-2.1.0-201309252230 ESF-XX-2.1.0-FC [esms1->softwareimage*[ESF-XX-2.1.0-FC*]]% commit [esms1->softwareimage[ESF-XX-2.1.0-FC]]% 4. Set kernel parameters to rdloaddriver=scsi_dh_rdac. [esms1->softwareimage[ESF-XX-2.1.0-FC]]% set kernelparameters rdloaddriver=scsi_dh_rdac [esms1->softwareimage*[ESF-XX-2.1.0-FC*]]% commit The cmgui can also be used to set kernel parameters. Select the software image from the Resources tree, then select Settings, and enter the kernel parameter rdloaddriver=scsi_dh_rdac in the Kernel Parameters field as shown in Figure 45. Figure 45. Setting RDAC Kernel Parameters for Fibre Channel 5.20 Configuring Lustre® Monitoring Tool (LMT) and Cerebro on the CIMS Note: Refer to the man pages for lmtinit(8), lmtsh(8), lmtop(1), and lmt.conf(1) for more information. S–2327–B 229 Data Management Platform (DMP) Administrator's Guide 5.20.1 Configuring Cerebro This configuration is explained in terms of the LMT server (CIMS node) and the LMT agents (CLFS nodes). The lmt-server package is installed on the CIMS node. The lmt-server-agent package is installed in CLFS nodes. The Cerebro configuration file, /etc/cerebro.conf, contains comments describing the settings and their defaults. You can also view this information by viewing the man page for cerebro.conf. On CIMS node, set up the configuration file as follows: LMT_SERVER_IP=$(hostname -i) echo "cerebrod_speak off cerebrod_listen on cerebrod_listen_message_config $LMT_SERVER_IP" > /etc/cerebro.conf This configuration makes the CIMS node listen for Cerebro messages on its IP address without sending any of its own. On the CLFS nodes, assuming LMT_SERVER_IP contains the IP address of the CIMS node, set up the configuration file as follows: echo "cerebrod_speak on cerebrod_speak_message_config $LMT_SERVER_IP cerebrod_listen off" > /etc/cerebro.conf This causes the CLFS nodes to send Cerebro messages to the IP address of the CIMS node without listening for messages. The Cerebro daemon (cerebrod) can be set up to start automatically on the CIMS node and on the Lustre servers using chkconfig. To do so, run the following command on the CIMS node and on all Lustre servers. esms1# chkconfig --level 235 cerebrod on This causes cerebrod to start whenever init is run in run levels 2, 3, and 5. These are the run levels for multi-user mode, multi-user mode with networking, and multi-user mode with networking and X11, respectively. Turning on cerebrod with chkconfig will not start cerebrod immediately, but it will configure it to start whenever the system is booted (in run levels 2, 3, and 5). After the chkconfig command, you can start cerebrod manually as described below in Starting Cerebro and LMT on page 230. 5.20.2 Starting Cerebro and LMT The Cerebro daemon (cerebrod) can be started manually, if chkconfig has not been used to configure cerebrod to start on boot or if the system has not been restarted since the chkconfig was issued. To begin sending data into LMT, start the cerebrod on all CLFS nodes and the CIMS node as follows: esms1# pdsh -w NodeList "/sbin/service cerebrod start" /sbin/service cerebrod start 230 S–2327–B CLFS Administration Tasks [5] Example 15. Starting cerebrod manually esms1# pdsh -w esfs-mds[1,2],esfs-oss[1,2,3,4] "/sbin/service cerebrod start" esms1# /sbin/service cerebrod start 5.20.3 Configuring the MySQL Database for Cerebro and LMT After LMT and Cerebro have been installed, the MySQL database must be configured and the cerebrod on the CIMS node must be restarted before data will be added to the database. Bright Cluster Manager® (Bright) uses a MySQL database on the CIMS node. Procedure 71. Configuring the MySQL database for Cerebro and LMT 1. Log in to the CIMS as root. 2. Edit the mkusers.sql file to change the password from mypass to the site password. esms1# chmod 600 /usr/share/lmt/mkusers.sql esms1# vi /usr/share/lmt/mkusers.sql • Edit the GRANT statements to grant privileges on only filesystem_FileSystemName.* where FileSystemName is the name of your file system. This will only grant permissions on the database for the file system being monitored. • Edit the password for lwatchadmin by changing mypass to the desired password. Also add a password for the lwatchclient user. Here is an example mkusers.sql script where the file system is named scratch, and the desired passwords for lwatchclient and lwatchadmin are foo and bar. CREATE USER 'lwatchclient'@'localhost' IDENTIFIED BY 'foo'; GRANT SELECT ON filesystem_scratch.* TO 'lwatchclient'@'localhost'; CREATE USER 'lwatchadmin'@'localhost' IDENTIFIED BY 'bar'; GRANT SELECT,INSERT,DELETE ON filesystem_scratch.* TO 'lwatchadmin'@'localhost'; GRANT CREATE,DROP ON filesystem_scratch.* TO 'lwatchadmin'@'localhost'; FLUSH PRIVILEGES; 3. Enter the following command and type root password when prompted to verify that there are no errors in themkusers.sql script. esms1# mysql -u root -p < /usr/share/lmt/mkusers.sql 4. Change the permissions of /etc/lmt/lmt.conf so that only root has read access, then edit the file. esms1# chmod 600 /etc/lmt/lmt.conf esms1# vi /etc/lmt/lmt.conf 5. Edit the LMT configuration file (/etc/lmt/lmt.conf) to add the passwords S–2327–B 231 Data Management Platform (DMP) Administrator's Guide for the users lwatchclient and lwatchadmin. Replace nil in the following line with the lwatchclient password in double quotation marks (foo in the example): esms1# lmt_db_ropasswd = "foo" 6. Create the /etc/lmt/rwpasswd file, type the lwatchadmin password in the file (lmt.conf reads the password from this file), and save the file. It should be accessible only by the root user. esms1# touch /etc/lmt/rwpasswd esms1# chmod 600 /etc/lmt/rwpasswd esms1# vi /etc/lmt/rwpasswd 7. (Optional) Note that a similar password file scheme can be used for the lwatchclient user if desired. To do so, replace the lmt_db_ropasswd line with the following: f = io.open("/etc/lmt/ropasswd") if (f) then lmt_db_ropasswd = f:read("*l") f:close() else lmt_db_ropasswd = nil end 8. Create the /etc/lmt/ropasswd file, type the lwatchclient password in the file, and save the file. It should be accessible only by the root user. esms1# touch /etc/lmt/ropasswd esms1# chmod 600 /etc/lmt/ropasswd esms1# vi /etc/lmt/ropasswd Note: Creating separate files for both passwords enables /etc/lmt/lmt.conf to be readable by anybody, while protecting the passwords for lwatchadmin and lwatchclient. 9. Create the database for the file system being monitored. esms1# lminit -a FileSystemName 10. Restart cerebrod on the CIMS. esms1# /sbin/service cerebrod restart 11. Verify that LMT is adding data to its MySQL database by using lmtsh to bring up the LMT shell. Then enter t to list tables. If the row count increases when you enter t again, LMT is configured properly. esms1# lmtsh -f FileSystemName 12. Use the ltop command to display real time information about a Lustre file system. esms1# ltop -f FileSystemName 232 S–2327–B CLFS Administration Tasks [5] 13. The cerebrod service will continue to gather data for LMT until the CIMS and CLFS nodes are rebooted. Reboot the CIMS and CLFS nodes, and repeat steps step 13.a through step 13.b to gather fresh data. Note: Use the cmsh reboot -c category command from cmsh device mode to reboot the CLFS nodes. a. Restart the cerebrod daemon on each Lustre server. Substitute the node names of your MDS and OSS nodes in the following command. esms1# pdsh -w esfs-mds00[1,2],esfs-oss00[1,2,3,4] "/sbin/service cerebrod restart" b. Restart the cerebrod on the CIMS. esms1# /sbin/service cerebrod restart 5.20.4 Managing the Data There are two ways to view data provided by LMT. You can view live data with ltop, or you can view historical data from the MySQL database with lmtsh. Refer to the man pages for each command using the --help option for more information. Note: You can also access the MySQL database directly to view the data if you need more control over how the data is presented. LMT provides scripts which aggregate data into the aggregate tables in the MySQL database. To run the aggregation scripts, enter the following: esms1# /usr/share/lmt/cron/lmt_agg.cron This command may take some time to complete, but subsequent executions will be much faster. To see the tables which were populated by the aggregation scripts, use lmtsh. The aggregation script can be set up to run as a cron job if you would like the aggregated tables to be populated on a regular basis. Use the following commands to set up a cron job: esms1# crontab -e 0 * * * * /usr/share/lmt/cron/lmt_agg.cron LMT does not provide an automated utility for clearing old data from the MySQL database; this must be done manually using MySQL commands. For example, to clear all data from the MDS_OPS_DATA table which is older than October 4th at 15:00:00, run the following mysql command: esms1# mysql -p -e "use filesystem_FileSystemName; delete MDS_OPS_DATA from MDS_OPS_DATA inner join TIMESTAMP_INFO on MDS_OPS_DATA.TS_ID=TIMESTAMP_INFO.TS_ID where TIMESTAMP < '2013-10-04 15:00:00';" S–2327–B 233 Data Management Platform (DMP) Administrator's Guide 5.20.5 Stopping the Cerebro Service To stop Cerebro from sending data to LMT, stop the Cerebro daemon from running on all Lustre servers and the LMT server. NodeList in the following command can be esfs-mds[1,2],esfs-oss[1,2,3,4]. esms1# pdsh -w NodeList "/sbin/service cerebrod stop" esms1# /sbin/service cerebrod stop If cerebrod has been turned on with chkconfig, it can also be turned off with chkconfig so that it does not start every time the system is booted. To turn off cerebrod, use: esms1# chkconfig --level 235 cerebrod off 5.20.6 Deleting the LMT MySQL Database To delete the LMT MySQL database, enter the following command where FileSystemName is the name of the file system you would like to remove. esms1# lmtinit -d FileSystemName To remove the MySQL users added by LMT, run the following MySQL command: esms1# mysql -u root -p -e "drop user 'lwatchclient'@'localhost'; drop user 'lwatchadmin'@'localhost';" 5.20.7 Managing Cerebro with Bright The cerebrod service can be started, stopped, and monitoring using Bright. The following procedures adds the Cerebro service to the esfs-mds1 slave node. This same procedure can be used to monitor the cerebrod service on the CIMS. Procedure 72. Managing Cerebro on a slave node with Bright 1. Log in to the CIMS as root and start cmsh. esms1# cmsh [esms1]% 2. Switch to device services mode and select the esfs-mds1 slave node. [esms1]% device services esfs-mds1 [esms1->device[esfs-mds1]->services]% 234 S–2327–B CLFS Administration Tasks [5] 3. Add and configure the cerebrod service. [esms1->device[esfs-mds1]->services]% add cerebrod [esms1->device*[esfs-mds1*]->services*[cerebrod*]]% show Parameter Value ------------------------------ -------------------------- Autostart no Belongs to role no Monitored no Revision Run if ALWAYS Service cerebrod [esms1->device*[esfs-mds1*]->services*[cerebrod*]]% set monitored on [esms1->device*[esfs-mds1*]->services*[cerebrod*]]% set autostart on [esms1->device*[esfs-mds1*]->services*[cerebrod*]]% commit [esms1->device[esfs-mds1]->services[cerebrod]]% 4. To monitor the status of cerebrod, enter: [esms1->device[esfs-mds1]->services[cerebrod]]% status cerebrod [DOWN] Procedure 73. Managing Cerebro for a category 1. Log in to the CIMS as root and start cmsh. esms1# cmsh [esms1]% 2. Switch to category services mode and select the esFS-MDS category. [esms1]% category services esFS-MDS [esms1->category[esFS-MDS]->services]% 3. Configure the cerebrod services for the esFS-MDS category. [esms1->category[esFS-MDS]->services]% add cerebrod [esms1->category*[esFS-MDS*]->services*]% set autostart yes [esms1->category*[esFS-MDS*]->services*]% set monitored yes [esms1->category*[esFS-MDS*]->services*]% commit esms1->category[esFS-MDS]->services]% Refer to the Bright Cluster Manager 6.0 Administrator Manual for more information about configuring services in Bright. S–2327–B 235 Data Management Platform (DMP) Administrator's Guide 236 S–2327–B Monitoring and Troubleshooting [6] Bright Cluster Manager® (Bright) software enables administrators to monitor health check information and metrics. Health checks run periodically from the cluster management daemon (cmd or CMDaemon) and may run on either the CIMS or slave node or both. Health checks return a PASS, FAIL or UNKNOWN condition, and actions can be taken based on the return value. Metrics also run periodically from CMDaemon (cmd) and may run on either the CIMS, slave node, or both. Metrics return a numeric value, and actions can be taken based on crossing a threshold value. Refer to Failover Features and Bright Monitoring on page 167 for information about how to install esfsmon_healthcheck and esfsmon_action to monitor failover conditions. Refer to the Bright Cluster Manager 6.0 Administrator Manual for more information about Cray Data Management Platform (DMP) monitoring capabilities. The monitoring system: • Inspects monitoring data at preset levels to preserve resources • Configures and gathers monitoring data for new resources • Identifies current and past problems or abnormal behavior • Analyzes trends that help an administrator predict likely future problems • Handles problems by triggering alerts • Taking action, if necessary, to improve the situation or to investigate further Bright uses the term Metric to describe a property of a device that can be monitored and has a numeric value. Examples are 45.2  C, any number like 1.23, or a value in bytes such as 12322343. Thresholds can be set so that when a the threshold is crossed (ion either direction), Actions are taken. Actions are stand-alone shell scripts that run when a monitoring condition is met. Health checks are device states, that are returned when a health check script is periodically run on a node. Return values for the example are, PASS, FAIL, or UNKNOWN. An example of health the healthcheck feature follows: • Determine if a hard drive has enough space available and returning a PASS status if it does • Determine if an NFS® mount is accessible, and returning FAIL if it is not S–2327–B 237 Data Management Platform (DMP) Administrator's Guide • Determine if CPUUser is below 50%, and returning PASS if it is • Determine if the cmsh binary is found, and returning UNKNOWN if it is not Refer to Table 7 for alter level descriptions. The cmsh monitoring mode is separated into 4 sections: actions, healthchecks, metrics, and setup. The monitoring actions mode enables control over the actions you have defined in shell scripts, or the built in actions such as power off. Example 16. cmsh monitoring actions Mode esms1# cmsh [esms1]% monitoring actions [esms1->monitoring->actions]% list Name (key) Command ------------------------ ------------------------------------------------ Drain node Power off Power on Power reset Reboot SendEmail Shutdown Undrain node killprocess /cm/local/apps/cmd/scripts/actions/killprocess.pl remount /cm/local/apps/cmd/scripts/actions/remount testaction /cm/local/apps/cmd/scripts/actions/testaction 238 S–2327–B Monitoring and Troubleshooting [6] Example 17. cmsh monitoring healthchecks mode esms1# cmsh [esms1->monitoring->healthchecks]% list name (key) command ------------------ ------------------------------------------------------- DeviceIsUp ManagedServicesOk chrootprocess /cm/local/apps/cmd/scripts/healthchecks/chrootprocess cmsh /cm/local/apps/cmd/scripts/healthchecks/cmsh diskspace /cm/local/apps/cmd/scripts/healthchecks/diskspace exports /cm/local/apps/cmd/scripts/healthchecks/exports failedprejob /cm/local/apps/cmd/scripts/healthchecks/failedprejob failover /cm/local/apps/cmd/scripts/healthchecks/failover hardware-profile /cm/local/apps/cmd/scripts/healthchecks/node-hardware-+ hpraid /cm/local/apps/cmd/scripts/healthchecks/hpraid interfaces /cm/local/apps/cmd/scripts/healthchecks/interfaces ipmihealth /cm/local/apps/cmd/scripts/metrics/sample_ipmi ldap /cm/local/apps/cmd/scripts/healthchecks/ldap lustre /cm/local/apps/cmd/scripts/healthchecks/lustre mounts /cm/local/apps/cmd/scripts/healthchecks/mounts mysql /cm/local/apps/cmd/scripts/healthchecks/mysql ntp /cm/local/apps/cmd/scripts/healthchecks/ntp oomkiller /cm/local/apps/cmd/scripts/healthchecks/oomkiller portchecker /cm/local/apps/cmd/scripts/healthchecks/portchecker rogueprocess /cm/local/apps/cmd/scripts/healthchecks/rogueprocess schedulers /cm/local/apps/cmd/scripts/healthchecks/schedulers smart /cm/local/apps/cmd/scripts/healthchecks/smart ssh2node /cm/local/apps/cmd/scripts/healthchecks/ssh2node swraid /cm/local/apps/cmd/scripts/healthchecks/swraid testhealthcheck /cm/local/apps/cmd/scripts/healthchecks/testhealthcheck Refer to the Bright Cluster Manager 6.0 Administrator Manual for more information about each metric and descriptions for each. Example 18. cmsh monitoring metrics Mode esms1# cmsh [esms1->monitoring]% metrics [esms1->monitoring->metrics]% list Name (key) Command ---------------------------- ------------------------------------------------ AlertLevel Ambient_Temp /cm/local/apps/cmd/scripts/metrics/sample_ipmi . . . S–2327–B 239 Data Management Platform (DMP) Administrator's Guide 6.1 Check Device Status Use the following cmsh commands to check node and other device status in Bright. esms1# cmsh [esms1]% device status KVM ................. [ UP ] esms ................ [ UP ] eth-01 .............. [ UP ] ib-01 ............... [ UP ] ib-02 ............... [ UP ] mds01 ............... [ UP ] mds02 ............... [ UP ] oss01 ............... [ UP ] oss02 ............... [ UP ] oss03 ............... [ UP ] oss04 ............... [ UP ] [esms]% 6.2 Check Power Status Use the following cmsh commands to check node power status, and the power status of other devices in Bright. esms# cmsh [esms]% device power status No power control .... [ UNKNOWN ] KVM No power control .... [ UNKNOWN ] esms No power control .... [ UNKNOWN ] eth-01 No power control .... [ UNKNOWN ] ib-01 No power control .... [ UNKNOWN ] ib-02 ipmi0 ............... [ ON ] mds01 ipmi0 ............... [ ON ] mds02 ipmi0 ............... [ ON ] oss01 ipmi0 ............... [ ON ] oss02 ipmi0 ............... [ ON ] oss03 ipmi0 ............... [ ON ] oss04 [esms]% 6.3 Check Node Health Status Use the following cmsh commands to check node health status, and the health status of other devices in Bright. esms1# cmsh [esms1]% device [esms1->device]% showhealth Device AlertLevel Failed Thresholds Unknown ------------------- ---------- -------------------- -------------- --------------- esmaint-net-switch 0 esms1 30 ManagedServicesOk ib-switch-1 0 ipmi-net-switch 0 lake-esl 0 mds001 10 mounts, rogueprocess 240 S–2327–B Monitoring and Troubleshooting [6] mds002 40 DeviceIsUp node001 no data oss001 10 mounts, rogueprocess oss002 10 mounts, rogueprocess eslogin1 0 The AltertLevel entries are defined in Table 7: Table 7. healthcheck Alert Levels Value Name Description 0 Info Informational Message 10 Notice Normal, but significant condition 20 Warning Warning conditions 30 Error Error conditions 40 alert take immediate action esms1# cmsh [esms1]% device [esms1->device]% latesthealthdata mds001 Health Check Severity Value Age (sec.) Info Message ---------------------- -------- ------- --------- ------------------------- DeviceIsUp 0 PASS 45 ManagedServicesOk 0 PASS 165 mounts 10 FAIL 645 defined mountpoint /proc has different+ rogueprocess 10 FAIL 645 /usr/sbin/sendmail.sendmail (smmsp) /*6+ ssh2node 0 PASS 645 [esms1->device]% 6.4 Monitoring Configuration with cmgui The monitoring configuration for health checks, metrics, and triggered actions is configured in the Monitoring Configuration section under resources section of cmgui. S–2327–B 241 Data Management Platform (DMP) Administrator's Guide Figure 46. Bright Monitoring Configuration The Monitoring option in the menu bar of cmgui starts a visualization tool that enables you to monitor system behavior over periods of time. The monitoring framework enables you to monitor a condition, add an Action (an executable shell script), and then set up a threshold level that triggers the action. Procedure 74. Monitoring configuration setup This procedures configures a monitoring rule that alerts the administrator if an CDL node goes down for any reason. 1. In cmgui, select Monitoring Configuration from the resource tree (refer to Figure 46.) 2. From the Threshold Wizard, select a device category (this example selects the esLogin-XE category). 242 S–2327–B Monitoring and Troubleshooting [6] Figure 47. Monitoring Configuration Category 3. Click Next, then scroll down and locate the DeviceIsUp health check metric from the menu, and click Next again. S–2327–B 243 Data Management Platform (DMP) Administrator's Guide Figure 48. Monitoring Configuration Metric 4. In the Set Actions pulldown menu, select Power Reset, and enter the name of the slave node to reset (this example shows eslogin1). You can also include a SendEmail action, and enter an email address for the administrator. 244 S–2327–B Monitoring and Troubleshooting [6] Figure 49. Monitoring Configuration Action 5. Click Finish to save the monitoring rule. 6.5 Check System Status The CM software enables you to display power and device status. The following states indicate a normal operational status: • CLOSED — The node is not being monitored by Bright Cluster Manager (Bright). • OPEN — The node is being monitored by the Bright and is not in one of the following states. • DOWN — The operating system is down and/or the CIMS cannot communicate S–2327–B 245 Data Management Platform (DMP) Administrator's Guide with CMDaemon (cmd) running on the node. This can be an expected state if it is intentionally entered by the administrator. If it in not intentionally in a DOWN state, expect a failure. Note: In cases of high load on a CLFS node during normal operation, the CLFS node state may toggle between UP and DOWN within 2-3 second intervals. This is not a failure, but indicates that the node was busy handling file system traffic during the time the CIMS was requesting its state. • INSTALLING — The Bright node-installer is provisioning the node during the boot process. • INSTALLER_CALLINGINIT — The CM node-installer has handed over control to the local init process. • UP — The OS is up and the CIMS is able to manage the node. The following states indicate a problem has occurred during the boot process: • INSTALLER_FAILED — The Bright node-installer has detected an unrecoverable problem during the boot process or has taken too long to enter the UP state. Possible reasons for this state include: – Local hard disk not found. – Failure to start a network interface. – Previous state was INSTALLER_REBOOTING and the reboot took too long. Possible reasons for failure to reach the UP state in time include: • Failure to hand over control from the Bright node-installer to the local init process. • The local init process failed to start CMDaemon (cmd) or the cmd took too long to start if the latter, this state will go to UP when cmd starts. • INSTALLER_UNREACHABLE — The CIMS CMDaemon (cmd) can no longer ping the node. The node may have crashed while running the CM node-installer. • INSTALLER_REBOOTING — The Bright node-installer may need to reboot a node to install a new kernel. • INSTALLER_FAILED — The Bright node-installer failed or it took too long to long to enter the UP state. Use the examples in the following procedures to check status of individual nodes or a group of nodes. You can check the power status of nodes individually, by list, range, category, or node group as shown in Procedure 75. 246 S–2327–B Monitoring and Troubleshooting [6] Procedure 75. Check power status 1. Log in to the CIMS as root and run cmsh. esms1# cmsh [esms1]% 2. Type device at the cmsh prompt to enter device mode. [esms1]% device [esms1->device]% a. To check power status of an individual node such as, eslogin01, type: [esms1->device]% power -n eslogin01 status b. To check power status of a list of nodes, such as eslogin01 to eslogin04 plus eslogin06, type: [esms1->device]% power -n eslogin01..eslogin04,eslogin06 status c. To check power status of all nodes in the eslogin category, type: [esms1->device]% power -c eslogin status d. To check power status of all nodes in the dm node group, type: [esms1->device]% power -g dm status 3. You can check the device status of nodes individually, by list and range, by category or node group. a. To check device status of an individual node such as, eslogin01, type: [esms1->device]% status -n eslogin01 b. To check device status of a list of nodes, such as eslogin01 to eslogin04 plus eslogin06, type: [esms1->device]% status -n eslogin01..eslogin04,eslogin06 c. To check device status of all nodes in the eslogin-XE category, type: [esms1->device]% status -c eslogin-XE d. To check device status of all nodes in the datamover node group, type: [esms1->device]% status -g datamover 6.6 Monitoring Health and Metrics • Refer to Switches and PDUs on page 34 for information about monitoring switches, RAID controllers, and other devices using Bright. • Refer to Configuring the LSI® MegaCLI™ RAID Utility on page 69 for information about monitoring CIMS, CDL, or CLFS local RAID systems. S–2327–B 247 Data Management Platform (DMP) Administrator's Guide • Refer to Configuring CLFS Failover (esfsmon) on page 165 for information about configuring esfsmon to monitor file system failover. • Set Email Alerts when a Node Goes Down on page 250 describes how to configure a healthcheck to send an Email when a node goes down. Procedure 76. Monitor system health and metrics 1. Log in to the CIMS as root and start cmsh. esms1# cmsh 2. Type device at the cmsh prompt to enter device mode and display the device mode prompt: [esms1]% device [esms1->device]% 3. Type latesthealthdata device -v to monitor system health and metrics: [esms1->device]% latesthealthdata esfs-oss001 -v Health Check Severity Value Age (sec.) Info Message ---------------------------- -------- ---------------- ---------- ---------------------------------------- DeviceIsUp 0 PASS 111 ManagedServicesOk 0 PASS 111 mounts 0 PASS 1551 rogueprocess 0 PASS 351 smart 0 PASS 1551 sda: Smart command failed sdb: Smart command failed sdc: Smart command failed sdd: Smart command failed sde: Smart command failed sdf: Smart command failed sdg: Smart command failed sdh: Smart command failed ssh2node 0 PASS 1551 interfaces 0 PASS 1551 oomkiller 0 PASS 1551 diskspace:2% 10% 20% 0 PASS 1551 ntp 0 PASS 351 schedulers 0 PASS 1551 ipmihealth 0 PASS 111 Procedure 77. Display the metric status 1. Log in to the CIMS as root and start cmsh. esms1# cmsh 2. Type device at the cmsh prompt to enter device mode and display the device mode prompt: [esms1]% device [esms1->device]% 248 S–2327–B Monitoring and Troubleshooting [6] 3. Type latestmetricdata device -v: [esms1->device]% latestmetricdata esfs-oss001 -v Metric Value Age (sec.) Info Message ---------------------------- ---------------- ---------- ---------------------------------------- AlertLevel:max 0 106 AlertLevel:sum 0 106 BytesRecv:BOOTIF 332.35 166 BytesRecv:eth1 0 166 BytesRecv:eth2 0 166 BytesRecv:eth3 0 166 BytesRecv:ib0 8.4 166 BytesRecv:ib1 0 166 ... Procedure 78 shows you how to dump the health check data for a node. You can dump the collected health check data for a node for a specified period of time. Most health checks are logging the past 3000 samples. Available health checks for a node are given by the latesthealthdata command described above. Procedure 78. Dump health check data 1. Log in to the CIMS as root and start cmsh. esms1# cmsh 2. Type device at the cmsh prompt to enter device mode and display the device mode prompt: [esms1]% device [esms1->device]% 3. Type dumphealthdata start-time end-time healthcheck device. The following example dumps the deviceisup data for the past hour: [esms1->device]% dumphealthdata -1h now deviceisup esfs-oss001 # From Wed Sep 11 13:32:10 2013 to Wed Sep 11 14:32:10 2013 Time Value Info Message -------------------------- ---------------- ------------------------------------------ Wed Sep 11 13:32:10 2013 PASS Wed Sep 11 14:32:00 2013 PASS Procedure 79 shows you how to dump the metric data for a node. You can dump the collected metric data for a node for a specified period of time. Most metrics are logging the past 3000 samples. Available metrics for a node are given by the latestmetricdata command described above. Procedure 79. Dump metric data 1. Log in to the CIMS as root and start cmsh. esms1# cmsh 2. Type device at the cmsh prompt to enter device mode and display the device mode prompt: [esms1]% device [esms1->device]% S–2327–B 249 Data Management Platform (DMP) Administrator's Guide 3. Type dumpmetricdata start-time end-time healthcheck device. The following example dumps the deviceisup data for the past hour: [esms1->device]% dumpmetricdata -1h now FreeSpace:/var/crash eslogin1 # From Wed Sep 11 13:38:58 2013 to Wed Sep 11 14:38:58 2013 Time Value Info Message -------------------------- ---------------- ------------------------------------------ Wed Sep 11 13:38:58 2013 1.64826e+10 Wed Sep 11 14:38:58 2013 1.64826e+10 6.6.1 Set Email Alerts when a Node Goes Down Procedure 80. Set email alerts when a node goes down 1. Log in to the CIMS as root and start cmsh. esms1# cmsh [esms1]% 2. Switch to monitoring mode. [esms1]% monitoring [esms1->monitoring]% 3. Switch to setup mode. [esms1]% setup [esms1->monitoring->setup]% 4. Enter healthconf and the node category name for the devices that you want to configure. [esms1]% healthconf esLogin-XC [esms1->monitoring->setup[esLogin-XC]->healthconf]% 5. Enter add and press the tab key to see the available command line options. [esms1->monitoring->setup[esLogin-XC]->healthconf]% add chrootprocess failedprejob ipmihealth mysql schedulers cmsh failover ldap ntp smart aeviceisup hardware-profile lustre oomkiller ssh2node diskspace hpraid managedservicesok portchecker swraid exports interfaces mounts rogueprocess testhealthcheck 6. Enter add deviceisup. Note: Pressing the Tab key displays a list of valid options. [esms1->monitoring->setup[esLogin-XC]->healthconf]% add Tab chrootprocess failedprejob ipmihealth mysql schedulers cmsh failover ldap ntp smart deviceisup hardware-profile lustre oomkiller ssh2node diskspace hpraid managedservicesok portchecker swraid exports interfaces mounts rogueprocess testhealthcheck 7. Add the SendEmail action to DeviceIsUp [esms1->monitoring->setup[esLogin-XC]->healthconf]% set DeviceIsUp failactions SendEmail 250 S–2327–B Monitoring and Troubleshooting [6] 8. Enter commit to store the changes. [esms1->monitoring->setup[esLogin-XC]->healthconf]% commit 6.7 Log Files All syslog traffic from the DMP slave nodes is forwarded to /var/log/messages on the CIMS. In addition to syslog, Bright maintains an event log in its database (refer to View the Event Log on page 251.) Ensure that CMDaemon (cmd) is running on the CIMS and on the suspect node if cluster management operations are malfunctioning. Other log files of interest are: • /var/log/messages — Slave node syslog messages. • /var/log/cmdaemon — Bright CMDaemon (cmd) messages. Check this log if there are system management problems. • /var/log/node-installer — Node Installer messages. Check this log if there are boot problems. • /var/log/conman/ — Slave node console messages. • /var/adm/cray/logs — Software installation logs. • Bright event log. Stored in the Bright database and accessed using events command in cmsh or event viewer when using the cmgui. 6.8 Bright Logs 6.8.1 View the Event Log Procedure 81 shows you how to view the event log. Procedure 81. View the event log 1. Log in to the CIMS as root and start cmsh. esms1# cmsh S–2327–B 251 Data Management Platform (DMP) Administrator's Guide 2. Type events followed by the number of events to display. Use events details eventnum to display details for a specific event. [esms1]% events 50 Thu Aug 29 16:30:43 2013 [notice] esms1: Starting image directory removal: /cm/images/ESL-XE-1.1.1-kdump Thu Aug 29 16:31:00 2013 [notice] esms1: Check 'chrootprocess' is in state PASS on esms1 Thu Aug 29 16:31:47 2013 [notice] esms1: Image directory removal succeeded for: /cm/images/ESL-XE-1.1.1-kdump Fri Aug 30 08:29:14 2013 [notice] esms1: Service named was restarted Fri Aug 30 08:42:04 2013 [warning] esms1: Service nfs died Fri Aug 30 08:42:05 2013 [notice] esms1: Service nfs was restarted Fri Aug 30 09:18:03 2013 [warning] lake-esl: Check 'ntp' is in state FAIL on lake-esl For details type: events details 3118 Fri Aug 30 09:18:03 2013 [warning] oss001: Check 'ntp' is in state FAIL on oss001 ... esms1]% events details 3118 ntpd not synchronized to a time server 6.8.2 View the rsync Log Procedure 82 shows you how to view the rsync log which records which changes were stored to a particular device during the last image update operation. Procedure 82. View the rsync log 1. Log in to the CIMS as root and start cmsh. esms1# cmsh 2. Type device at the cmsh prompt to enter device mode and display the device mode prompt: [esms1]% device [esms1->device]% 3. Type synclog device: [esms1->device]% synclog lustre1-oss001 252 S–2327–B Changing BIOS for a DELL™ R720 Managed CDL Node [A] Before you can configure a managed CDL node with Bright Cluster Manager® (Bright), you must change the BIOS and Dell remote access controller (iDRAC) settings. The following configuration is required for each CDL node. • A CDL node must be cabled to both the CIMS administration network (esmaint-net) and IPMI network (ipmi-net). • A CDL node using a workload manager such as Moab® or TORQUE must be cabled directly to the SDB on the Cray system over the wlm-net. • A CDL node must be configured to provide IPMI Serial Over LAN (SOL) for remote console support. • A CDL node must be configured to PXE boot from the CIMS (embedded NIC) before attempting to boot from the local disk. Use the following procedure to change the BIOS and iDRAC settings for a CDL node. Procedure 83. Changing a R720 slave node's BIOS and iDRAC settings Note: This procedure shows specific steps for a Dell R720 system. Dell R815 BIOS setup procedures are also in the Appendix. 1. Power up the slave node. When the BIOS power-on self-test (POST) process begins, quickly press the F2 key after the following messages appear in the upper-right of the screen. F2 = System Setup F10 = System Services F11 = BIOS Boot Manager F12 = PXE Boot When the F2 keypress is recognized, the F2 = System Setup line changes to Entering System Setup. S–2327–B 253 Data Management Platform (DMP) Administrator's Guide After the POST process completes and all disk and network controllers have been initialized, the Dell System Setup screen appears. The following submenus are available: System BIOS iDRAC Settings Device Settings Note: In this utility, use the Tab key to move to different areas on the screen. To select an item, use the up-arrow and down-arrow keys to highlight the item, then press the Enter key. Press the Escape key to exit a submenu and return to the previous screen. 2. Change the system BIOS settings. a. Select System BIOS, then press Enter. See Figure 50. Figure 50. Dell 720 System BIOS Settings b. Select Boot Settings, then press Enter. c. Select BIOS Boot Settings, then press Enter. d. Select Boot Sequence, then press Enter to view the boot settings. e. In the pop-up window, change the boot order so that the integrated NIC appears first, before the optical (DVD) drive. The hard drive should be last on the list. See Figure 51. Tip: Use the up-arrow or down-arrow key to highlight an item, then use the + and - keys to move the item up or down. 254 S–2327–B Changing BIOS for a DELL™ R720 Managed CDL Node [A] Figure 51. Dell 720 Boot Sequence BIOS Settings f. Be sure that Hard Drive C: is enabled under the Boot Option Enable/Disable section. See Figure 51. g. Press Enter to return to the BIOS Boot Settings screen. h. Press Escape to exit BIOS Boot Settings. i. Press Escape to exit Boot Settings and return to the System BIOS Settings screen. 3. Change the serial communication settings. a. On the System BIOS Settings screen, select Serial Communication. b. On the Serial Communication screen, select Serial Communication. A pop-up window displays the available options. c. Select On with Console Redirection via COM2, then press Enter. d. Verify that Serial Port Address is set to Serial Device1=COM1, Serial Device2=COM2. Note: This setting enables the remote console. If this setting is incorrect, you cannot use a remote console to access the CDL node. 1) If necessary, press Enter to display the available options. 2) Change the setting to Serial Device1=COM1, Serial Device2=COM2. See Figure 52. S–2327–B 255 Data Management Platform (DMP) Administrator's Guide Figure 52. Dell 720 Serial Device BIOS Settings 3) Press Enter to return to the Serial Communication screen. e. Select External Serial Connector. A pop-up window displays the available options. f. In the pop-up window, select Remote Access Device, then press Enter to return to the previous screen. g. Select Failsafe Baud Rate. A pop-up window displays the available options. Figure 53. Dell 720 Serial Communication BIOS Settings h. In the pop-up window, select 115200, then press Enter to return to the previous screen. 256 S–2327–B Changing BIOS for a DELL™ R720 Managed CDL Node [A] i. Press the Escape key to exit the Serial Communication screen. j. Press the Escape key to exit the System BIOS Settings screen. k. Press the Escape key to exit the BIOS Settings screen. l. A "Settings have changed" message appears. Select Yes to save your changes. m. A "Settings saved successfully" message appears. Select OK. 4. On the System Setup Main Menu, select iDRAC Settings, then press Enter. See Figure 54. Figure 54. Dell 720 iDRAC BIOS Settings 5. Select Network, then press Enter. A long list of network settings is displayed. 6. Change the IPMI settings to enable the Serial Over LAN (SOL) console. a. Use the down-arrow key to scroll to the IPMI SETTINGS list. b. Ensure that IPMI over LAN (or Enable IPMI over LAN) is enabled. 1) If necessary, select IPMI over LAN, then press Enter. 2) In the pop-up window, select . 3) Press Enter to return to the previous screen. c. Press the Escape key to exit the Network screen and return to the iDRAC Settings menu. 7. Change the LCD configuration to show the host name in the LCD display. a. On the iDRAC Settings screen, use the down-arrow key to scroll down and highlight LCD (or Front Panel Security), then press Enter. S–2327–B 257 Data Management Platform (DMP) Administrator's Guide b. Select Set LCD message. A pop-up window opens. c. In the pop-up window, select User-Defined String, then press Enter. d. Select User-Defined String (again), then press Enter. A text pop-up window opens for entering the new string. See Figure 55. Figure 55. Dell 720 iDRAC BIOS LCD Settings e. In the text pop-up window, enter the CDL host name (such as eslogin1), then press Enter. f. Press the Escape key to exit the LCD screen. g. Press the Escape key to exit the Network screen. h. Press the Escape key to exit the iDRAC Settings screen. i. A "Settings have changed" message appears. Select Yes, then press Enter to save your changes. j. A "Settings saved successfully" message appears. Select OK, then press Enter. 8. Change the device settings so that the CDL node can PXE boot on the CIMS administration network (esmaint-net). a. On the System Setup Main Menu, select Device Settings, then press Enter. b. In the Device Settings window, select Integrated NIC 1 Port N ..., then press Enter. The Main Configuration Page opens. 258 S–2327–B Changing BIOS for a DELL™ R720 Managed CDL Node [A] Tip: Choose the NIC port number that corresponds to the Ethernet port for the esmaint-net network. • If esmaint-net uses the first Ethernet port (eth0), select Integrated NIC 1 Port 1 ... • If esmaint-net uses the third Ethernet port (eth2), select Integrated NIC 1 Port 3 ... Note: PXE booting must be disabled for the other three Ethernet ports. c. On the Main Configuration Page screen, select MBA Configuration Menu, then press Enter. See Figure 56. Figure 56. Dell 720 MBA Configuration Menu BIOS Settings d. On the MBA Configuration Menu screen, select Legacy Boot Protocol, then press Enter. A pop-up window displays the available options. e. In the pop-up window, use the down-arrow key to highlight PXE, then press Enter. See Figure 57. S–2327–B 259 Data Management Platform (DMP) Administrator's Guide Figure 57. Dell 720 Legacy Boot Protocol BIOS Settings f. Press the Escape key to exit the MBA Configuration Menu screen. g. Press the Escape key to exit the Main Configuration Page screen. h. Verify that Legacy Boot Protocol is set to None for the other three Ethernet ports. If necessary, repeat step 8.b through step 8.g to change the setting for these three ports. i. Press the Escape key to exit the Device Settings screen. j. A "Settings have changed" message appears. Select Yes, then press Enter to save your changes. k. A "Settings saved successfully" message appears. Select OK, then press Enter. The main screen (System Setup Main Menu) appears. 9. Save your changes and exit. a. Press Escape to exit the System Setup Main Menu. b. The utility displays the prompt "Are you sure you want to exit and reboot?" Select Yes. Next, you will use Bright on the CIMS to create an CDL node. 260 S–2327–B Changing BIOS for a DELL™ R815 Managed CDL Node [B] Before configuring a managed CDL node with Bright Cluster Manager® (Bright), you must change the BIOS and Dell remote access controller (iDRAC) settings. The following configuration is required for each CDL node. • A CDL node must be cabled to both the CIMS administration network (esmaint-net) and IPMI network (ipmi-net). • A CDL node using a workload manager such as TORQUE or Moab®, must be cabled directly to the SDB on the Cray system over the wlm-net. • A CDL node must be configured to provide IPMI Serial Over LAN (SOL) for remote console support. • A CDL node must be configured to PXE boot from the CIMS (embedded NIC) before attempting to boot from the local disk. Use the following procedure to change the BIOS and iDRAC settings for a CDL node. Procedure 84. Changing a R815 slave node's BIOS and iDRAC settings Note: This procedure shows specific steps for a Dell R815 system. See Procedure 83 for Dell R720 BIOS set up procedure. 1. Power up the slave node. When the BIOS power-on self-test (POST) process begins, quickly press the F2 key after the following messages appear in the upper-right of the screen. F2 = System Setup F10 = System Services F11 = BIOS Boot Manager F12 = PXE Boot When the F2 keypress is recognized, the F2 = System Setup line changes to Entering System Setup. 2. Select Boot Settings, then press Enter. S–2327–B 261 Data Management Platform (DMP) Administrator's Guide Figure 58. Dell 815 Boot Settings Menu a. Select Boot Sequence, then press Enter to view the boot settings. b. In the pop-up window, change the boot order so that the integrated NIC appears first, before the optical (DVD) drive. The hard drive should be last on the list. Figure 59. Dell 815 Boot Sequence Menu 262 S–2327–B Changing BIOS for a DELL™ R815 Managed CDL Node [B] Figure 60. Dell 815 Boot Sequence Settings c. Press Enter to return to the BIOS Boot Settings screen. 3. Press Esc to return to the System Setup Menu, scroll down and select Integrated Devices. Figure 61. Dell 815 Integrated Devices (NIC) Settings a. Set Embedded NIC 1 to Enabled with PXE. b. Set Embedded Gb NIC 2 to Enabled. c. Scroll down and set Embedded NIC 3 to Enabled. S–2327–B 263 Data Management Platform (DMP) Administrator's Guide d. Set Embedded Gb NIC 4 to Enabled. e. Press Esc to return to the System Settings Menu. 4. Change the serial communication settings. Figure 62. Dell 815 Serial Communication BIOS Settings a. Select Serial Communication. b. Select Serial Communication and set it to On with Console Redirection via COM2. c. Select Serial Port Address and set it to Serial Device=COM1, Serial Device2=COM2. d. Set External Serial Connector, and set it to Remote Access Device. e. Set Failsafe Baud Rate to 115200. f. Press Esc to return to the System Setup Menu. 5. Select Embedded Server Management. 264 S–2327–B Changing BIOS for a DELL™ R815 Managed CDL Node [B] Figure 63. Dell 815 Embedded Server Management Settings a. Set Front-Panel LCD Options to User-Defined LCD String. b. Set User-Defined LCD String to your login host name, such as eslogin1. Figure 64. Dell 815 User-defined LCD String Settings 6. Save your changes and exit. a. Press Escape to exit the System Setup Main Menu. b. The utility displays the prompt "Are you sure you want to exit and reboot?" Select Yes. S–2327–B 265 Data Management Platform (DMP) Administrator's Guide 7. When the system reboots, press Ctrl-E to configure the iDRAC port settings. www.dell.com iDRAC6 Configuration Utility 1.60 Copyright 2011 Dell Inc. All Rights Reserved Four 2.10 GHz Twelve-core Processors, L2/L3 Cache: 6 MB/10 MB iDRAC6 FirmwareaRevisionHversion: 1.70.21 . . .' IPv4 Stack : Enabled IP Address : 10.148. 0 . 2 Subnet mask : 255.255. 0 . 0 Default Gateway : 0 . 0 . 0 . 0 Press for Remote Access Setup within 5 sec...... a. Set the iDRAC6 LAN to ON. b. Set IPMI Over LAN to ON. Figure 65. Dell 815 DRAC LAN Parameters Settings c. Select LAN Parameters and press Enter. Set the IPv4 address to next available IP address on the esmaint-net network (10.148.0.x). d. Press Esc to return to the iDRAC6 menu, and Esc to exit and save. 266 S–2327–B Changing BIOS for a DELL™ R815 Managed CDL Node [B] Figure 66. Dell 815 DRAC IPv4 Parameter Settings S–2327–B 267 Data Management Platform (DMP) Administrator's Guide 268 S–2327–B Changing BIOS for a DELL™ R720 Managed CLFS Node [C] Before you can configure a managed CLFS node with Bright Cluster Manager® (Bright), you must change the BIOS and Dell remote access controller (iDRAC) settings. The following configuration is required for each CLFS node. • A CLFS node must be cabled to both the CIMS administration network (esmaint-net) and IPMI network (ipmi-net). • A CLFS node using a workload manager such as TORQUE or Moab®, must be cabled directly to the SDB on the Cray system over the wlm-net. • A CLFS node must be configured to provide IPMI Serial Over LAN (SOL) for remote console support. • Processor hyper-threading (the "Logical Processor" setting) must be disabled on CLFS nodes. Use the following procedure to change the BIOS and iDRAC settings for a CLFS node. Procedure 85. Changing a R720 CLFS node's BIOS and iDRAC settings Note: This procedure shows specific steps for a Dell 720 system. 1. Power up the slave node. When the BIOS power-on self-test (POST) process begins, quickly press the F2 key after the following messages appear in the upper-right of the screen. F2 = System Setup F10 = System Services F11 = BIOS Boot Manager F12 = PXE Boot When the F2 keypress is recognized, the F2 = System Setup line changes to Entering System Setup. S–2327–B 269 Data Management Platform (DMP) Administrator's Guide After the POST process completes and all disk and network controllers have been initialized, the Dell System Setup screen appears. The following submenus are available: System BIOS iDRAC Settings Device Settings Note: In this utility, use the Tab key to move to different areas on the screen. To select an item, use the up-arrow and down-arrow keys to highlight the item, then press the Enter key. Press the Escape key to exit a submenu and return to the previous screen. 2. Change the system BIOS settings. a. Select System BIOS, then press Enter. See Figure 67. Figure 67. Dell 720 System BIOS Settings b. Select Boot Settings, then press Enter. c. Select BIOS Boot Settings, then press Enter. d. Select Boot Sequence, then press Enter to view the boot settings. e. In the pop-up window, change the boot order so that the integrated NIC appears first, before the optical (DVD) drive. The hard drive should be last on the list. See Figure 68. Tip: Use the up-arrow or down-arrow key to highlight an item, then use the + and - keys to move the item up or down. 270 S–2327–B Changing BIOS for a DELL™ R720 Managed CLFS Node [C] Figure 68. Dell 720 Boot Sequence BIOS Settings f. Be sure that Hard Drive C: is enabled under the Boot Option Enable/Disable section. g. Press Enter to return to the BIOS Boot Settings screen. h. Press Escape to exit BIOS Boot Settings. i. Press Escape to exit Boot Settings and return to the System BIOS Settings screen. 3. On the System BIOS Settings screen, select Processor Settings and press Enter. a. Select Logical Processor, and press enter. Verify that Logical Processor is set to Disabled. b. Press Escape to exit Processor Settings. 4. Change the serial communication settings. a. On the System BIOS Settings screen, select Serial Communication. b. On the Serial Communication screen, select Serial Communication. A pop-up window displays the available options. c. Select On with Console Redirection via COM2, then press Enter. d. Verify that Serial Port Address is set to Serial Device1=COM1, Serial Device2=COM2. Note: This setting enables the remote console. If this setting is incorrect, you cannot use a remote console to access the CLFS node. 1) If necessary, press Enter to display the available options. S–2327–B 271 Data Management Platform (DMP) Administrator's Guide 2) Change the setting to Serial Device1=COM1, Serial Device2=COM2. See Figure 69. Figure 69. Dell 720 Serial Device BIOS Settings 3) Press Enter to return to the Serial Communication screen. e. Select External Serial Connector. A pop-up window displays the available options. f. In the pop-up window, select Remote Access Device, then press Enter to return to the previous screen. g. Select Failsafe Baud Rate. A pop-up window displays the available options. Figure 70. Dell 720 Serial Communication BIOS Settings 272 S–2327–B Changing BIOS for a DELL™ R720 Managed CLFS Node [C] h. In the pop-up window, select 115200, then press Enter to return to the previous screen. i. Press the Escape key to exit the Serial Communication screen. j. Press the Escape key to exit the System BIOS Settings screen. k. Press the Escape key to exit the BIOS Settings screen. l. A "Settings have changed" message appears. Select Yes to save your changes. m. A "Settings saved successfully" message appears. Select OK. 5. On the System Setup Main Menu, select iDRAC Settings, then press Enter. See Figure 71. Figure 71. Dell 720 iDRAC BIOS Settings 6. Select Network, then press Enter. A long list of network settings is displayed. 7. Change the IPMI settings to enable the Serial Over LAN (SOL) console. a. Use the down-arrow key to scroll to the IPMI SETTINGS list. b. Ensure that IPMI over LAN is enabled. 1) If necessary, select IPMI over LAN, then press Enter. 2) In the pop-up window, select . 3) Press Enter to return to the previous screen. c. Press the Escape key to exit the Network screen and return to the iDRAC Settings menu. 8. Change the LCD configuration to show the host name in the LCD display. S–2327–B 273 Data Management Platform (DMP) Administrator's Guide a. On the iDRAC Settings screen, use the down-arrow key to highlight LCD, then press Enter. b. Select Set LCD message. A pop-up window opens. c. In the pop-up window, select User-Defined String, then press Enter. d. Select User-Defined String (again), then press Enter. A text pop-up window opens for entering the new string. See Figure 72. Figure 72. Dell 720 iDRAC BIOS LCD Settings e. In the text pop-up window, enter the CLFS host name (such as esfs-mds001), then press Enter. f. Press the Escape key to exit the LCD screen. g. Press the Escape key to exit the Network screen. h. Press the Escape key to exit the iDRAC Settings screen. i. A "Settings have changed" message appears. Select Yes, then press Enter to save your changes. j. A "Settings saved successfully" message appears. Select OK, then press Enter. 9. Change the device settings so that the CLFS node can PXE boot on the CIMS administration network (esmaint-net). a. On the System Setup Main Menu, select Device Settings, then press Enter. b. In the Device Settings window, select Integrated NIC 1 Port N ..., then press Enter. The Main Configuration Page opens. 274 S–2327–B Changing BIOS for a DELL™ R720 Managed CLFS Node [C] Tip: Choose the NIC port number that corresponds to the Ethernet port for the esmaint-net network. • If esmaint-net uses the first Ethernet port (eth0), select Integrated NIC 1 Port 1 .... • If esmaint-net uses the third Ethernet port (eth2), select Integrated NIC 1 Port 3 .... Note: PXE booting must be disabled for the other three Ethernet ports. c. On the Main Configuration Page screen, select MBA Configuration Menu, then press Enter. See Figure 73. Figure 73. Dell 720 MBA Configuration Menu BIOS Settings d. On the MBA Configuration Menu screen, select Legacy Boot Protocol, then press Enter. A pop-up window displays the available options. e. In the pop-up window, use the down-arrow key to highlight PXE, then press Enter. See Figure 74. S–2327–B 275 Data Management Platform (DMP) Administrator's Guide Figure 74. Dell 720 Legacy Boot Protocol BIOS Settings f. Press the Escape key to exit the MBA Configuration Menu screen. g. Press the Escape key to exit the Main Configuration Page screen. h. Verify that Legacy Boot Protocol is set to None for the other three Ethernet ports. If necessary, repeat step 9.b through step 9.g to change the setting for these three ports. i. Press the Escape key to exit the Device Settings screen. j. A "Settings have changed" message appears. Select Yes, then press Enter to save your changes. k. A "Settings saved successfully" message appears. Select OK, then press Enter. The main screen (System Setup Main Menu) appears. 10. Save your changes and exit. a. Press Escape to exit the System Setup Main Menu. b. The utility displays the prompt "Are you sure you want to exit and reboot?" Select Yes. Next, you will use Bright on the CIMS to create a CLFS node. 276 S–2327–B

Écrire un commentaire

Optionnel