recoverability and multiple storage pools in a volume

[jdub]

Hello,

I'm going to be adding an other storage pool to an xsan and am trying to figure out if I should expand the existing volume with a new storage pool, or to create a new volume from the pool.

The main question I'm trying to resolve involves recoverability. Backups aside, I've been asked to create a setup where some set of files are on both pools so that the data is still immediately available in case one of the LUNs has a failure.

Some things I've read indicate that if any LUN in any of the pools that are part of a volume fails, the entire volume is destroyed (presumably including the data on the unaffected storage pool.

However, the training docs say: "Each storage pool is assigned an affinity tag according to the pool's performance and recoverability", which seems to indicate that recoverability is specific to the underlying pool, not the entire volume.

I would rather expand the existing volume rather than make a new one so that users don't need to keep track of 2 separate volumes. But I don't want the new (larger) pools to be vulnerable to a failure in the older pool.

Does anyone have experience with losing a LUN from a volume with multiple pools? Does it render the entire volume unusable? Is there any other reason why I may want to create a new volume?

Thanks

[singlemalt]

You can down a data stripe group and have data on other stripe groups available. see man cvadmin and the verb down. Or mark the stripe as down
in the .cfg file. Obviously you can't mark the meta data stripe group as
down and start the volume. I think your management may not be thinking this
through though. What happens if the irrecoverable lun is the meta data lun?
Or if the file system gets corrupt to the point that you can't start the volume
at all? Wouldn't matter then if you did have multiple copies of the file on
different stripe groups since you can't get the volume to start.

[jdub]

Thanks, that helps clarify things a lot.

So there's some isolation to the storage pools in a volume. But creating a new volume with its own metadata lun would add extra redundancy.

[singlemalt]