khimaros / raid-explorations Goto Github PK

I did test the dm-integrity with raid too (my blog post) and have come to a completely opposite conclusion: that dm-integrity does help to protect against silent corruption.

As far as I can tell from the errors you quoted, the issue may be in the test damaging dm-integrity metadata and/or journal (or mdadm superblocks). In such cases, the recovery is much more complex than just "assemble and resync", but it's not any different than without dm-integrity. If you damage the mdadm superblock, the array won't assemble either. 10GiB partitions also make it particularly likely that the corruption will hit the dm-integrity journal instead of the data or checksum block (for a 1.5TiB partition integritysetup used a 64MiB journal).

What happens when in case of issues mounting the dm-integrity you open it with --integrity-no-journal option?

(Side note: having dealt with multiple failing hard drives, I've never seen a failure mode in which the read would change just one byte in a sector. When silent corruption occurred the whole sector contained more-or-less random data).

Hello,

Regarding metadata level on btrfs it's complicated. When you have 3 or more disks it makes sense to have raid1c3 regardless of data level. Only if your kernel does not support it, or you have only 2 disks it must be raid1.
That could explain why raid6 fails:

dm-crypt + btrfs: 30KB corruption on 2/4 raid6 disks
btrfs fails catastrophically in this case despite the corruption being within expected parameters for a raid6 array.

Of course raid5/raid6 is still problematic on btrfs.