Question about PowerPath Migration Enabler for Windows 2008 Clusters
I have been using PowerPath Migration Enabler Host Copy to perform migrations from a CX3-20 to a CX4-120. It has worked great on my windows servers enabling me to do host-based migrations with zero downtime.
Now it comes to migrating the Windows 2008 clusters. I'm reading through the PowerPath Migration Enabler 5.5 User Guide and on pages 51-52 it discusses using PPME for clusters. The steps seem the same until it comes to step 8.
It says:
On Windows 2003 clusters when ready to commit the migration run "powermig commit" and you are finished.
On Windows 2008 clusters it says to first "bring the failover groups offline, stopping the applications", then run "powermig commit", and finally bring the failover groups back online.
Am I reading this correctly that I must bring the groups down in order to commit the migration, but I did not have to do so on 2003? If so, that is disappointing, because PPME has been great for all of my other migrations. I wonder why the groups have to be shut down in 2008 when they did not need to be in 2003.
Hello. PowerPath Migration Enabler (PPME) is a non-cluster application that under limited conditions can be used to migrate in clusters. It is not aware of the persistent reservation commands. PPME does not transfer the registration from source to target during the "commit" process.
Once a migration is committed and the cluster starts using the target/new device, it expects to find a persistent reservation registration on that device. However, it does not have the expected registration. For that reason we need to off-line the resource group, commit the migration, then restart the groups. This will initialize the new disk with the appropriate registration on the new device. This doesn't apply to Windows 2003 because it doesn't use the SCSI-3 persistent reservation that Windows 2008 uses. There will be enhancements to PPME in a coming version that will not require this step. I hope this helps.
You are reading it correctly. The basic difference between Windows 2003 Clusters and Windows 2008 Clusters is that Windows 2003 Clusters are dependent on disk signatures - whereas Windows 2008 are dependent on GUIDs. This necessitates the difference between the 2 ways of commiting the changeover.
Another quick question for you. In the same document it says to "move all failover groups to the active node" and says to shut down the other nodes in the cluster.
If my cluster has multiple cluster services running and I plan on doing a migration on only one, do I still have to fail all groups to a single node and shut the other nodes down? And if I do, is it still required to restart all services during powermig commit, or just the one where the migration took place?
I think I can sell taking the momentary outage to restart the individual cluster service, but if I have to shut down every single one of them in this cluster just to migrate one, I don't think that will fly. It would also be beneficial for me if I could leave the remaining services running on the other nodes while the single service is running on the node that is performing the migration.
That is helpful information. The only other thing I need to know is if just the resource group being migrated needs to be restarted or if all resource groups must be restarted.
If you have multiple applications running across the cluster hardware and you only want to migrate one set of devices for one application, you "could" unmask just those devices from all but one of the cluster nodes. That way, you wouldn't necessarily have to shut everything down. Today, that is not an officially supported migration method for PPME. I would recommend speaking with your EMC representative about an RPQ for this. The RPQ is a request for a non-standard configuration.
Thanks for the info on the unsupported method, Brion. In this case we're working on a Clariion where all three cluster nodes are in the same storage group, so it would be tricky to remove the LUNs only from two of the nodes, but I will keep that in mind for future reference.
Everything was successful with our PPME migration on a SQL 2008 cluster. The only outage was the momentary shutdown of the failover groups to run powermig commit.
We did have an issue on a subsequent migration on a non-cluster machine. We decided to try a migration from a thick LUN to a thin LUN (Clariion thin pool). When we ran the powermig setup command to put the source and target into a session, the target LUNs (the thin ones) went into an "unreadable" state according to Windows Disk Management.
So this new problem is probably for a new post, but the original SQL 2008 migrations were successful thanks to all of your help.
Brion2
154 Posts
2113
0
Posted November 15th, 2010 11:00
Hello. PowerPath Migration Enabler (PPME) is a non-cluster application that under limited conditions can be used to migrate in clusters. It is not aware of the persistent reservation commands. PPME does not transfer the registration from source to target during the "commit" process.
Once a migration is committed and the cluster starts using the target/new device, it expects to find a persistent reservation registration on that device. However, it does not have the expected registration. For that reason we need to off-line the resource group, commit the migration, then restart the groups. This will initialize the new disk with the appropriate registration on the new device. This doesn't apply to Windows 2003 because it doesn't use the SCSI-3 persistent reservation that Windows 2008 uses. There will be enhancements to PPME in a coming version that will not require this step. I hope this helps.