Following a storage subsystem failure, it may be necessary to recover file systems.
File system corruption due to an event (such as RAID controller crash, storage system component failure, or power loss) often affects objects that were being modified around the time of the event.
The file system is configured to keep up to 128 checkpoints. The maximum number of checkpoints supported is 1024. The number of checkpoints preserved is configurable when the file system is formatted, but, once set, the number of checkpoints cannot be changed.
When a checkpoint completes, rather than immediately freeing the storage used for the previous checkpoint, the file system maintains a number of old checkpoints. As each new checkpoint completes, the oldest checkpoint is overwritten. This means that there can be multiple checkpoints on-disk, each of which is complete and internally consistent point-in-time view of the file system. If necessary, the file system can be restored to any of these checkpoints.
In the case of file system corruption, if there are enough checkpoints on disk, it may be possible to roll back to a previous checkpoint, pre-dating the event that caused the corruption and restoring the file system using the uncorrupted checkpoint. This may be possible even if this event occurred up to a few minutes before the file system was taken offline.
To restore a file system to a previous checkpoint, use the fs-checkpoint-health and the fs-checkpoint-select commands. Refer to the Command Line Reference for more information about these commands.
- Restoring a file system using a checkpoint does not affect snapshots taken prior to the checkpoint being restored, but, like any other file system update, snapshots taken after that checkpoint are lost.
- After restoring to a checkpoint, it is possible to restore again, to an older checkpoint and, if the file system has not been modified, restore again, to a more recent checkpoint. So, for example, it is possible to mount the file system in read only mode, check its status, and then decide whether to re-mount the file system in normal (read/write) mode or to restore to a different checkpoint.