[LU-4799] Warning message in ll_dirty_page_discard_warn should not be rate limited

• Type: Improvement
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.6.0, Lustre 2.5.2
• Affects Version/s: Lustre 2.6.0
• Labels: patch
• Severity: 3
• Assignee: WC Triage
• Reporter: Ryan Haasken

Description

There is a warning message printed from ll_dirty_page_discard_warn(), meant to flag potentially corrupt files on client eviction, e.g.:

2014-01-03T06:50:10.880132-06:00 c3-0c2s7n0 Lustre: 5121:0:(llite_lib.c:2507:ll_dirty_page_discard_warn()) snx11029: dirty page discard: 10.10.100.3@o2ib6000:10.10.100.4@o2ib6000:/snx11029/fid: [0x215827c4c:0xca43:0x0]//filename may get corrupted (rc -108)
2014-01-03T06:50:10.905666-06:00 c3-0c2s7n0 Lustre: 5121:0:(llite_lib.c:2507:ll_dirty_page_discard_warn()) Skipped 1 previous similar message

This message should not be rate limited; the point is to identify data files that might have missing data as a result of client eviction. With the rate limiting, some suspect files won't be explicitly flagged.

Comments

[Nathaniel Clark]

Merged to master aacc286fe0575d455794121264a612c29ffb7740

[Ryan Haasken]

Oleg, vvp_vmpage_error() handles limiting dirty page discard warnings to one message per file. Isn't that right?

The patch worked correctly in my testing, printing only one message per file.

[Oleg Drokin]

Well, the thing is we definitely need to rate-limit this, but be more smart about it.
Multiple pages from teh same file warnings could be safely merged and rate-limited.
It's all the different files that we should print about for sure.

[Ryan Haasken]

Uploaded a patch here: http://review.whamcloud.com/#/c/9752/