On Mon, 25 Jul 2022 08:34:04 -0700
Kent Borg <kentb...@borg.org> wrote:
> To check whether the data all reads back correctly do a scrub (takes
> a long time, and note the scrub happens on the underlying pool not
> the dataset,
> the volume does not have to be mounted to do a scrub, though the pool
> needs to be imported):
>
> zpool scrub extern_backup_pool
ZFS reads back every write immediately as part of the end to end data
validation. Frequent scrubbing is not necessary. Periodic scrubbing is
used to detect and repair errors that occour after the fact (cosmic ray
bit flips for example).
Here's a problem. Scrub can detect errors on this volume but it cannot
repair them because you have no data redundancy. I recommend enabling
copies=2 if you don't mirror your backup volumes:
zfs set copies=2 extern_backup_pool
While this will not save you from a device failure, it will allow zpool
scrub to repair data errors. At the cost of halving your backup storage
capacity and write performance. Note: this will not replicate existing
data; only new data will be replicated.
Scrubbing a production system is typically a weekly process. I scrub my
backup volumes roughly once a month.
--
\m/ (--) \m/
_______________________________________________
Discuss mailing list
Discuss@lists.blu.org
http://lists.blu.org/mailman/listinfo/discuss
