[LU-15009] precreate should cleanup orphans upon error

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.12.9, Lustre 2.15.0
• Severity: 3
• Assignee: Lai Siyao
• Reporter: Lai Siyao

Description

Sometimes OST object precreate succeeds, but the reply get lost or timed out, in this case MDT doesn't update the known last_id, and just retries. And then subsequent precreate will fail because the last_id known on MDT is less than that on OST. Instead, upon each failure, it should restart as a reconnect and cleanup orphans first, and then continue subsequent precreate.

Issue Links

• is related to: LU-5648 corrupt files contain extra data

Comments

[Gerrit Updater]

"Oleg Drokin <[e-mail]>" merged in patch https://review.whamcloud.com/45930/
Subject: LU-15009 ofd: continue precreate if LAST_ID is less on MDT
Project: fs/lustre-release
Branch: b2_12
Current Patch Set:
Commit: 62d4e1d5f1be33abd3ee9f58cc09d471e22ee7ad

[Gerrit Updater]

"Andreas Dilger <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/45930
Subject: LU-15009 ofd: continue precreate if LAST_ID is less on MDT
Project: fs/lustre-release
Branch: b2_12
Current Patch Set: 1
Commit: ac5034013c97f0e458252b7a337d3f184bd065ba

[Peter Jones]

Landed for 2.15

[Gerrit Updater]

"Oleg Drokin <[e-mail]>" merged in patch https://review.whamcloud.com/44984/
Subject: LU-15009 ofd: continue precreate if LAST_ID is less on MDT
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 1711e26ae861c28829870c2433caf7ee232909cf

[Gerrit Updater]

"Lai Siyao <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/44984
Subject: LU-15009 ofd: continue precreate if LAST_ID is less on MDT
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: bb30ada9d8f086bc3c2f97450df0fbb68d35f746

[Lai Siyao]

The description is updated, the issue is objects are created on OST, but MDT get error, IMO this is no different from a reconnect, and can be handled the same.

[Andreas Dilger]

In this case, the MDT should handle this by skipping OST objects below the LAST_ID value, rather than deleting OST objects back to the MDT last_id value. That would leave these objects available for recovery via LFSCK, if the problem was on the MDT (e.g. restored from backup), rather than deleting them.