[LU-11527] sanity test_270a failed with O_DIRECT on ARM

• Status: Resolved
• Project: Lustre
• Fix Version/s: Lustre 2.12.0
• Type: Bug
• Priority: Major
• Reporter: Mikhail Pershin
• Assignee: Hongchao Zhang
• Resolution: Fixed
• Labels: DoM2, arm
• Severity: 3

Description

The issue was reported by John Hammond, in test 270a the writing to the DOM file may fail with -ENOSPC due to grants most probably. If test 270a is being ran in a loop then the first  several times it passes but all the other times seem to fail.

Issue Links

• is related to: LU-4664 sync write should consume grant on cl...
• is related to: LU-11597 sanityn test 16a failed with direct I/O
• is related to: LU-11200 Centos 8 arm64 server support

Comments

[Hongchao Zhang]

this issue does not only exist at DoM, it is the same behavior on OST.
the reason of this issue is the direct write (with "oflag direct") ignores the grant at client side, and the grant of the client
will be increased alongside the I/O, and if the remain disk space (subtracted the total grant to clients) at MDT or OST
is not enough (could be more easy to reach for local test on VMs), it will cause -ENOSPC

there is no error (ran 200 times in a loop) after deleting the "oflag direct" in test 270a of sanity.th
but it will trigger -ENSPC less than 5 times with the "oflag direct"
this error is also reproduced by writing the normal file with "dd" + "oflag direct"

[Mikhail Pershin]

So in context of this ticket I propose to modify test and remove 'oflag direct' which is not so needed. As for general issue with DIRECT IO vs grants I'd first ask around is there any ticket about that.

[Gerrit Updater]

Andreas Dilger ([e-mail]) uploaded a new patch: https://review.whamcloud.com/33633
Subject: LU-11527 tests: fix sanity 270a for ARM
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 8d83e2fef87c2668e03bd2d701eed7c9cc2d6970

[Andreas Dilger]

The above patch fixes the problem with O_DIRECT writes in test_270a for ARM, since it needs O_DIRECT to be aligned with PAGE_SIZE=64KB.  There is a separate problem with O_DIRECT and grant usage, which is being addressed by patch https://review.whamcloud.com/9454 "LU-4664 clio: consume grant for sync write".

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch https://review.whamcloud.com/33633/
Subject: LU-11527 tests: fix sanity 270a for ARM
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 26eb3969f10168314fccfaeff0bfa275f4ea1b90

[Peter Jones]

Landed for 2.12