[LU-15548] Interop: sanity-quota test 40b fails with ‘project id expected 2 not 1'

• Status: Resolved
• Project: Lustre
• Affects Version/s: Lustre 2.15.0
• Fix Version/s: Lustre 2.15.0
• Type: Bug
• Priority: Minor
• Reporter: James Nunez (Inactive)
• Assignee: Dongyang Li
• Resolution: Fixed
• Labels: quota
• Severity: 3

Description

sanity-quota test_40b started failing with ‘project id expected 2 not 1' on January 10, 2022 with 2.14.56.67 servers and 2.14.0 clients with logs at https://testing.whamcloud.com/test_sets/673d6bdc-3a8b-470e-998a-2cfe113ac0a5.

On the same day, we see a filure between 2.14.56.67 servers and 2.12.7 clients with logs at https://testing.whamcloud.com/test_sets/e8909fc4-8359-452e-adf8-2c02d8ae09c6.

Looking at the suite_log, we see

== sanity-quota test 40b: Mv across different project ID ============================================= 05:45:15 (1641793515)
CMD: trevis-48vm4 /usr/sbin/lctl get_param -n version 2>/dev/null
Waiting for local destroys to complete
Creating test directory
CMD: trevis-48vm3,trevis-48vm4 lctl set_param fail_val=0 fail_loc=0
fail_val=0
fail_loc=0
fail_val=0
fail_loc=0
lfs project -sp 1 /mnt/lustre/d40b.sanity-quota/dir1
lfs project -sp 2 /mnt/lustre/d40b.sanity-quota/dir2
 sanity-quota test_40b: @@@@@@ FAIL: project id expected 2 not 1 
  Trace dump:
  = /usr/lib64/lustre/tests/test-framework.sh:6273:error()
  = /usr/lib64/lustre/tests/sanity-quota.sh:3489:test_40b()

There were a few patches that landed to master on January 6 that modify the quota/project id code:
https://review.whamcloud.com/45256 - LU-15097 quota: stop pool_recalc before killing pool
https://review.whamcloud.com/45032 - LU-15031 quota: reseed glbe in qmt_lvbo_udate
https://review.whamcloud.com/43987 - LU-13587 quota: protect qpi in proc
https://review.whamcloud.com/45679 - LU-12056 ldiskfs: add trusted.projid virtual xattr

Comments

[Peter Jones]

Dongyang

Does this seem related to LU-12056?

Peter

[Dongyang Li]

LU-12056 added project id xattr as an interface to check and set project id.
on client we will hide the project id xattr if parent has inherit flag set, meaning all the dir/files under parent should inherit the project id from parent.
older client doesn't do the hiding, so if we have 2 dirs dir1 and dir2, having project id 1 and 2, both have inherit flag set,
mv dir1/file to dir2/file will trigger the xattr set on dir2/file, with project id xattr being 1.

So the impact is on the old client, if we are moving a file to a different dir, and the src dir has project id inherit xattr set, then the file will retain the old project id after it's in the dest dir.

[Andreas Dilger]

Should we have a parameter to disable the project xattr exporting to clients, since I think it will be a case of some users will find this helpful, and other users will not. The virtual xattr should probably be disabled by default for clients, but enabled for ldiskfs.

[Gerrit Updater]

"Li Dongyang <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/46900
Subject: LU-15548 osd-ldiskfs: hide virtual projid xattr
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 58d18aa4a8382954c5c1b65c7bc8e5cbd08e1112

[Gerrit Updater]

"Oleg Drokin <[e-mail]>" merged in patch https://review.whamcloud.com/46900/
Subject: LU-15548 osd-ldiskfs: hide virtual projid xattr
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: ef826db1f43b2849af26491028f4cfc25595faf6

[Peter Jones]

Landed for 2.15

[Gerrit Updater]

"Andreas Dilger <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/48151
Subject: LU-15548 tests: skip conf-sanity/131 for older servers
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: a901bc0e0b4bf81b25be7680b07f813af0e23a11

[Gerrit Updater]

"Oleg Drokin <[e-mail]>" merged in patch https://review.whamcloud.com/48151/
Subject: LU-15548 tests: skip conf-sanity/131 for older servers
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: ac6528af8dc5961a6331d2b12a9f20a17ac0b605