Scrubbing impractical for large volume?

[Hans Otto Lunde]

I'm running TrueNAS-12.0-U8.1 as a backup-server using rsync to run daily backups from many different, primarily linux-systems.
It has a large pool for this data, about 70 TB.
The system has only run for a few months and works absolutely great.
My question is simple:
Until now I've only done some S.M.A.R.T tests now and then on the 8 disks, that hold the backup-data. They are Western Digital Red Pro, WDC WD141KFGX-68FH9N0 disks using CMR technology, not SMR. They pass long S.M.A.R.T tests just fine until now.
I'm reluctant to do a scrub on the data-pool, although I know I should.
But won't it take forever to finish?
From what I can read in this forum this will take many days, and then I will have to reschedule my backup-scripts and all sorts of things, won't I?
What is your advice on this?

 

[ChrisRJ]

You should ABSOLUTELY do regular scrubs (at least monthly). Scrubs happen in the background, there is no need to reschedule backups. You may face a performance impact while the scrub is running, but that's about it. On my system a scrub takes less than 7 hours.

Without regular scrubs you would seriously compromise the point of having ZFS in the first place.

 

[Arwen]

Agree with @ChrisRJ, performing ZFS scrubs is highly suggested, and at least once a month.

One of the key features of ZFS, is bit rot detection AND correction. A scrub on a redundant pool will do both. When ZFS can no longer correct a block on a disk, it's likely the disk is out of spare blocks. At that point it needs to be replaced.

Seeing even a few correctable errors over time, (perhaps a few a year), may be normal. Just a day in the life of ZFS doing it's job.


On rare occasions, a non-redundant pool make sense. My miniature media server does not have redundancy for the Media pool. (It does have a Mirrored pool for the OS.) Since I have backups, I can recover any lost media, IF I KNOW ABOUT IT. Thus, the Media pool gets twice a month scrubs, same as the OS pool.


To be clear, I have personally lost data to bit rot before ZFS. Without having some kind of check, it was impossible to tell if a file was being corrupted. (Meaning if the file modification time did not change, but a block in the file went bad, how would you know?) ZFS has built in checks that solve that problem.