[LU-19620] MDS can't resume llog processing if a plain llog is found broken at start

• Type: Bug
• Resolution: Unresolved
• Priority: Medium
• Severity: 3
• Assignee: Alex Zhuravlev
• Reporter: Alex Zhuravlev

Description

if llog is truncated for a reason (power loss, corruption, etc), then MDS breaks llog record processing:

[   45.564829] LustreError: 8282:0:(llog_osd.c:1243:llog_osd_next_block()) lustre-MDT0000-osd: invalid llog tail at log id [0x1:0x3:0x0] offset 8492 last_rec idx 65536 tail idx 5 lrt len 2 read_size 300
[   45.565304] LustreError: 8282:0:(osp_sync.c:1406:osp_sync_thread()) lustre-OST0000-osc-MDT0000: llog process with osp_sync_process_queues failed: -22

and corresponding osp thread exits:

# ps ax|grep osp
   8283 ?        I      0:00 [osp-pre-0-0]
   8284 ?        I      0:00 [osp-syn-1-0]
   8285 ?        I      0:00 [osp-pre-1-0]

this leads lost space on OST, missing credential updates.