Recovering from loss of production SDs
To recover production SDs which are lost while file systems are mounted
Procedure
If you do not wish to mount file systems at once, use the 'automount' command to disable automount.
Use your storage configurator to break the mirror relationships if the production SDs no longer exist, or to promote the backup SDs to primaries if the production SDs exist but have failed. If, as a result, mirror relationships move into the SSWS state and the surviving SDs remain as S-Vols, use the sd-peg command to make the server treat them as primaries. Remember to unpeg them (by pegging them back to the default role) as soon as the SSWS state is resolved.
Run the pn all span-list command to confirm that the span is healthy. If automount is enabled, file systems mount automatically. If they do not, or if automount is disabled, use the mount command to mount them. Before doing so, you can use the check-directory command to confirm that the backup file systems are intact.
This step depends on whether the mirrors are synchronous:
- If mirrors are synchronous then, as soon as you mount the file systems on your backup SDs, the server replays any uncommitted writes held in NVRAM, and the data on the production SDs must be considered out of date. If at all possible, avoid manually discarding NVRAM data during your recovery procedure, because it results in the loss of writes that the server has acknowledged to clients but not yet written to disk.
- If mirrors are near-synchronous or asynchronous, the data on disk is too old to permit NVRAM data to be replayed. In addition, recent changes to on-disk configuration (such as filesystem expansions) may not have been copied to the backup SDs. Start by rescanning the on-disk configuration using the sd-rescan-cod <span-instance-name> command. Next, try to mount the filesystems. If they do not mount in the usual way, discard the NVRAM data and then mount by force. Some recent writes will be lost.
Once file systems are mounted, either repair your existing production SDs or create new mirrored SDs for eventual use as production SDs; synchronize, then schedule an orderly failback. Re-enable automount.