|< Day Day Up >|| |
This appendix provides the steps to migrate to GPFS Version 1.3 from earlier versions of GPFS.
All nodes in a GPFS fileset must be running the same level of GPFS. If you want to test the newer revision of GPFS on a small subset of your cluster before installing it on the bulk of the nodes, you must remove some nodes from the GPFS nodeset. Details of how to do this are in the IBM General Parallel File System for Linux: Concepts, Planning, and Installation Guide, GA22-7844, although the procedure is a little involved. A potentially simpler approach is to install a small isolated cluster for testing. Here we cover simply the main migration of a cluster.
If you are changing kernel versions as well as GPFS levels, it is strongly recommend that you perform at least rudimentary testing.
To upgrade GPFS on an entire cluster:
Shut down GPFS on all the nodes.
GPFS cannot be running on the nodes when the software is updated.
Ensure no applications are running that access GPFS, un-export any NFS exported GPFS file systems, and then run, from the GPFS primary node:
# mmshutdown -a
This will shut down GPFS on all the nodes in your GPFS cluster.
Install the new level of code.
Section 7.4.1, "Installing the source files" on page 205 details the procedure for installing GPFS RPMs with SMS. This same procedure can be used to upgrade the nodes to a later revision of code. The basic steps are:
Copy all GPFS and bundled RSCT RPMs to /csminstall/Linux/RedHat/7.3/i386/RedHat/updates.
Run cd /csminstall/Linux/RedHat/7.3/i386/RedHat/updates.
Use smsupdatenode to install the new versions of the code. Ensure that your gpfs wildcard only include the very latest release, for example:
# smsupdatenode -ai gpfs*1.3.0-0* rsct*126.96.36.199-0*
If the wildcard expand to multiple levels (1.3.0-0 and 1.3.0-1, for example), the smsupdatenode will fail.
Re-build and re-distribute the GPFS portability layer.
Although the GPFS portability layer source code (gpfs.gpl.*) need not be installed on all the nodes in the cluster, the binary modules built from it must.
Before installing the new modules, we recommend that the ones from the previous version are backed up, in case you want to revert back to the old level of code. The files are mmfslinux, lxtrace, tracedev and dumpconv, and are located in /usr/lpp/mmfs/bin; If you are using CFM to distribute these files, the master copies reside in /cfmroot/usr/lpp/mmfs/bin on the management node.
The procedure for building and distributing the portability layer is documented in 7.4.2, "Building the GPFS open source portability layer" on page 206.
If you are upgrading your Linux kernel at the same time as GPFS, be sure to build the portability layer (as well as additional drivers, such as gm) against the new version of the kernel source.
Start the new software.
In order to start the new version of GPFS, the newly built kernel modules must be loaded. In order to load the new modules, the old ones must first be unloaded. If possible, we recommend that this is performed with a reboot of all the nodes in the GPFS cluster; this allows GPFS startup on boot to be verified and provides a known clean starting point in the event of any problems.
If the nodes cannot be rebooted, run the following from the master node to unload the modules:
# dsh -av /sbin/rmmod mmfs mmfslinux tracedev
If this command fails, the nodes must be rebooted anyway.
If the nodes were not rebooted (or if GPFS is not configured to start on boot), start it by running the following command on the GPFS primary node:
# mmstartup -a
Verify the correct operation.
At this point, you should verify that GPFS is functioning correctly at the latest level. Check that the file systems mounted correctly and that you can perform normal file system operations (cd, ls, cat, and so on). Run a "real" job and ensure it runs correctly to completion.
Migrate the file system.
You may not be taking advantage of all the new features of your new version of GPFS until you migrate the file system format to the latest level. However, once you migrate the file system it becomes permanently incompatible with previous versions of GPFS. It is not recommended that you migrate immediately; instead, continue running without migration until you are satisfied that the new release operates correctly.
When you are ready, format migration can be performed by running the following command for each of your file systems:
# mmchfs -V <gpfsdevice>
|< Day Day Up >|| |