Hello all,
I'm working to fix a two-node SysMirror cluster that uses NFS mounts from a NetApp appliance as the data repository. Currently all the NFS mounts/unmounts are called from the application controller scripts, and since file collection isn't currently working, (One fight at at time here..) the SAs need to remember to update the start/stop scripts on each node anytime changes are made.
My question is, is there a way to maintain and update the required NFS mounts for the resource group similar to how the filesystems and VGs are maintained by using the CSPOC commands to synchronize the changes across the cluster?
There's likely a cleaner way to implement this, but I haven't had enough practical interaction with HACMP/Sysmirror to really understand it.
Yes, there is. But before trying to administrate a HACMP-cluster you should really REALLY know what you are doing. This can end up in disaster pretty fast.
In the details of of the resource groups definition (the same SMIT panel where you also state the VGs which belong to it) you can also define NFS-mounts to be part of the RG. Note, that you will need some non-NFS file system for NFS-4 mounts. (Because, unlike NFS3, NFS4 is state-aware. Locked files stay locked.)
Notice also that for NFS4 to work you need different services started, not the same as for NFS3. You also need to configure the NFS-domain before, which i suggest to be the same as the DNS-domain of the system (as this is easier to manage).
Something else to look into (bakunin please verify still possible - I do not have a PowerHA/SystemMirror config to review) would be to remove the mount/unmount from the start/stop scripts.
Instead, make the mounts part of the resource group - and the resource group makes sure the mounts are active before calling the start script and the unmounts are called after the stop script.
Or, create a resource group parent-child relationship (e.g., you want the NFS mounts active across the cluster, but the application is active on only one node.) With this setup - to activate the current application (call this appA) it would only start if the NFS mounts are already active (appMounts).
As Bakunin says - PowerHA/SystemMirror can make things worse if you do not understand it. If you need short term assistance I would recommend contacting IBM Lab Services for a healthcheck and/or workshop (on-site)
Or, contact an education center for SystemMirror/PowerHA class schedule and prices - and go with at least two admins (or you become the SPOF - single point of failure)
Right now i don't have a HACMP-system at hand to check the details myself, but: yes, that is (still) possible and in fact is the recommended way to do it.
Mounts will be done by a sort-of start/stop-Script HACMP provides for itself, its name is "clas-nfsv4" or something similar (again, i can't check and am working from memory) which will be entered automatically when you configure an NFS-mount and/or -export. I am not sure if one can configure the order in which this and your own application-server is called but if you want to be perfectly sure configure the NFS-services in its own RG, then configure a start before/stop after-dependency with your application RG.
