[LU-7706] add dmesg -c to conf-sanity_83

• Status: Resolved
• Project: Lustre
• Type: Bug
• Priority: Minor
• Reporter: Sergey Cheremencev
• Assignee: WC Triage
• Resolution: Won't Fix
• Labels: patch
• Severity: 3

Description

conf sanity 83 may fail with message:

conf-sanity test_83: @@@@@@ FAIL: VFS: Busy inodes after unmount of vdb. Self-destruct in 5 seconds. Have a nice day... 

Dmesg clearing is needed in conf-sanity 83 because previous tests
(for example conf-sanity_62) may also be the results of message "VFS: Busy inodes after unmount of ...".

Issue Links

• is related to: LU-7227 Add a check to t-f that dmesg doesn't...

Comments

[Gerrit Updater]

Sergey Cheremencev ([e-mail]) uploaded a new patch: http://review.whamcloud.com/18122
Subject: LU-7706 tests: add dmesg -c to conf-sanity_83
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: dca4bf26e5177ece36d6d528106f1b033424c6d3

[Andreas Dilger]

It seems bad that any test causes the "VFS: Buay inodes ..." message to be hit. It would be better to find and fix that problem rather than working around this to allow the test to pass.

[Sergey Cheremencev]

The problem is fixed right now(in seagate) - it needs a kernel fix. But I didn't send it upstream yet.
If you also face this problem("busy inodes" after conf-sanity 62) I can send it faster or make a kernel patch for lustre.

conf-sanity_83 checks that there is no "busy inodes" messages caused by ENOSPACE on OST.
Why Does it need to fail when message is caused by another issue(theoretically not lustre related issue) ?
Even if the "busy inodes" after conf-sanity 62 will disappear there is no guarantee that another test will not cause the same message.

Earlier in https://jira.hpdd.intel.com/browse/LU-7227 I suggested to fail tests that lead to kernel errors like "VFS: Busy inodes ..."
New lustre code or fix may introduce "missed iput" and "VFS: busy inodes" as result. And it is better to find this error in testing if possible.

[Andreas Dilger]

New lustre code or fix may introduce "missed iput" and "VFS: busy inodes" as result. And it is better to find this error in testing if possible.

Right, so we should detect this failure even if it happened in a previous test, so erasing the previous messages isn't helpful. The fix in LU-7227 seems a lot better, since it catches the failure right away. I don't recall seeing that bug/patch, but I'd think it better to abandon this patch and ticket in favor of that one.

[Andreas Dilger]

Landed patch from LU-7227 instead.