Replication to Single drive Bit Rot protected?

[Catsrules]

I have a Raidz1 volume on my main Freenas server. That volume is replicated another freenas server to a Single large drive. Oviously the Raidz1 volume is protected againced bit rot but what type of protection (if any) is on that single drive?

I am just wondering if somehow the replication could detect/fix the bit rot.

Thanks.

 

[Stux]

Scrub after replication will detect bitrot.

 

[Catsrules]

Scrub after replication will detect bitrot.

Ahh, Cool so once the bitrot is detected would another replication fix the error?

 

[Stux]

Not necessarily. If you rolled back to the previous snapshot as part of the replication, it would.

If you're using ECC the bitrot should not happen as part of replication

 

[rs225]

It manifests in two ways, practically speaking. Corruption in the copying process (most commonly via USB), and slow degradation of the sector which eventually shows up as a read error when the built in drive error correction can't correct.

The 'bit-rot' that people imagine when you say 'bit-rot' doesn't really exist. If the bits are rotten, you get a read error (not corruption). And if the bits are wrong, it's because they were copied that way. (and ZFS should let you know.... with an I/O error.)

To address a corruption in your example, the easiest way would be to re-copy the source data on the source, so that it would be replicated again in the next snapshot.

 

[Stux]

ZFS gives you 'bit rot' aka data corruption detection, and if there is available redundancy, then correction when the data is read again, and that happens as part of a scrub.

 

[Catsrules]

ZFS gives you 'bit rot' aka data corruption detection, and if there is available redundancy, then correction when the data is read again, and that happens as part of a scrub.

And my primary server counts as redundancy in the case of my backup server. Correct?

 

[Ericloewe]

No, but you can transfer the data to the backup again.

Now that I think about it, it'd be nice if zfs recv could automagically do this, assuming send on the other side sent the relevant snapshot. Still a pain in the ass, but on average it's half the pain...