[LU-14860] setting obdfilter.*.access_log_size=0 fails Created: 16/Jul/21 Updated: 20/Jul/21 |
|
| Status: | Open |
| Project: | Lustre |
| Component/s: | None |
| Affects Version/s: | None |
| Fix Version/s: | None |
| Type: | Bug | Priority: | Minor |
| Reporter: | Andreas Dilger | Assignee: | WC Triage |
| Resolution: | Unresolved | Votes: | 0 |
| Labels: | None | ||
| Issue Links: |
|
||||||||
| Severity: | 3 | ||||||||
| Rank (Obsolete): | 9223372036854775807 | ||||||||
| Description |
|
While the code in access_log_size_store() appears to allow size=0 to be used to disable the OAL on a system, this actually fails with -EBUSY. Setting it to any valid size after it has been created fails with -EEXIST. It should be possible to (at least) disable the OAL after it was enabled without an unmount/remount, accepting that any entries in the log will be lost. |
| Comments |
| Comment by John Hammond [ 19/Jul/21 ] |
|
This was designed in to make it faster. Under the original implementation there were no shared locks between readers and writers. I guess it could be done under the semaphore we added but what is the use case? |
| Comment by Andreas Dilger [ 20/Jul/21 ] |
|
I enabled OAL on to see if I could use it for tracking hot object IO on the OSTs with a user script instead of via lamigo, but wasn't able to turn OAL off afterward. I think having a simple way to aggregate data from OAL for this would be useful for debugging but it should be possible to turn it off again once that is finished. |