Question Regarding Mirrored Pools + Block Repair (maybe wrong words for it)

[xCatalystx]

Quick Question,

Can someone explain how when HDD's are in mirror pair (or multiple mirrors) how it determines which block is the correct non-corrupt block if damage occurs / change is detected / invalid or different checksum? I couldn't find anything on the forum explaining it in detail or i might just be using the wrong keywords.

I did find details explanations on raidz1-3 but not mirrors.

 

[danb35]

Every block of data written, on any device, has a checksum attached to it. If ZFS reads the data from one disk in a mirrored pair, and the data doesn't match its checksum, it reads the same data from the other disk. If that data matches its checksum, well and good, and ZFS fixes the bad data. If both disks come up with bad data, then you have a problem.

 

[xCatalystx]

Ok, how does it know which checksum is right though? I assume the checksum is kept on both mirrors, so in a pair setup how does it determine which checksum is right? In a 3 way mirror I guess it could be worked out, I'm just wondering in 2way.

 

[danb35]

If the checksum matches the data, it's "right". If the data and the checksum are both incorrect in such a way that they match, you have a hash collision, which is (by design) astronomically unlikely.

 

[Stux]

The checksums are checksummed all the way back to the uber block, which has many copies.

 

[Ericloewe]

If the checksum matches the data, it's "right". If the data and the checksum are both incorrect in such a way that they match, you have a hash collision, which is (by design) astronomically unlikely.

Not astronomically unlikely with the default fletcher checksum, but unlikely enough for most uses. If it's not high enough (encryption, dedup, ...), there's always SHA512 and skein.