Document recovery behavior after interrupted multi-parity sync in Known Issues

This commit is contained in:
Andrea Mazzoleni 2026-09-08 09:10:20 +02:00
parent 5beb465d42
commit d6b0786eca
42 changed files with 1196 additions and 0 deletions

View file

@ -2045,6 +2045,36 @@ To avoid this limitation, avoid modifying protected files through hard
links that are excluded from SnapRAID, or ensure that all hard links to a
file are included in the scan so SnapRAID can normalize them.
16.8 Recovery after an interrupted multi-parity sync
----------------------------------------------------
In an array with multiple parity disks, an interrupted `sync` can leave
parity levels temporarily out of sync with each other. For example,
parity 1 might already contain newly updated data, while parity 2 still
contains the old data.
During `check` or `fix`, SnapRAID tries different parity combinations
and validates reconstructed data against the known block hashes (covering
the full block, typically 256 KiB).
If SnapRAID attempts a reconstruction using an inconsistent mix of old
and new parity, the linear RAID equations will produce incorrect data.
For this invalid combination to be falsely accepted, the equations would
have to accidentally reconstruct at least one complete 256 KiB block so
perfectly that its hash matches, while other unsynchronized blocks in the
same stripe are reconstructed with incorrect data.
This scenario does not depend on a hash collision; rather, the mixed
parity equations themselves would have to coincidentally yield the exact
original 256 KiB block contents. While mathematically possible with
specially crafted or highly repetitive data, it is practically impossible
with real-world data.
If all disks are healthy, completing the interrupted `sync` before making
further changes removes this ambiguity. If a disk has failed and recovery
is required, avoid modifying any data disks before running `check` or
`fix`.
17 COPYRIGHT
============