[LU-14522] missing ldlm lock processing causes timeouts in racer

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.15.0
• Severity: 3
• Assignee: Alex Zhuravlev
• Reporter: Alex Zhuravlev

Description

post-mortem analysis for locks may look like the following:

R: 240000402:35e1:0 in mdt-lustre-MDT0001_UUID
  W: 480963a72f1c4f5e in #09443 --/EX IB:22 40210400000020 in mdt-lustre-MDT0001_UUID
  W: 480963a72f1c4f9d in #08398 --/PR IB:13 40210000000000 in mdt-lustre-MDT0001_UUID
  W: 480963a72f1c54a5 in #11284 --/PR IB:13 40210000000000 in mdt-lustre-MDT0001_UUID
    #11284 hold 240000402:1:0 PR/PR IB:2 1/0 in mdt-lustre-MDT0001_UUID
    #11284 hold 240000402:1:0 CR/CR IB:2 1/0 in mdt-lustre-MDT0001_UUID
  W: 480963a72f1c89f7 in #11331 --/PR IB:13 40210000000000 in mdt-lustre-MDT0001_UUID
    #11331 hold 240000402:1:0 PR/PR IB:2 1/0 in mdt-lustre-MDT0001_UUID
    #11331 hold 240000402:1:0 CR/CR IB:2 1/0 in mdt-lustre-MDT0001_UUID

i.e. there are number of the locks in the waiting queue, but none is granted.

Issue Links

• duplicates: LU-14551 MDT hangs on ldlm_expired_completion_wait()