-
Type: Bug
-
Resolution: Fixed
-
Priority: Major - P3
-
Affects Version/s: None
-
Component/s: None
-
3
-
Storage - Ra 2022-02-21
Summary
The issue was identified when investigating WT-8690 in this comment. The error signature is the following:
[2022/01/28 02:48:57.954] 0x2064530:[1643338011:161][65029:0x7ff662ffd700], t, file:F00006, WT_SESSION.checkpoint: [WT_VERB_GENERATION][NOTICE]: evict generation drain waited 1 minutes [2022/01/28 02:48:57.954] 0x2064530:[1643338071:8][65029:0x7ff662ffd700], t, file:F00006, WT_SESSION.checkpoint: [WT_VERB_GENERATION][NOTICE]: evict generation drain waited 2 minutes [2022/01/28 02:48:57.954] 0x2064530:[1643338131:155][65029:0x7ff662ffd700], t, file:F00006, WT_SESSION.checkpoint: [WT_VERB_GENERATION][NOTICE]: evict generation drain waited 3 minutes [2022/01/28 02:48:57.954] [1643338131:184][65029:0x7ff662ffd700], t, file:F00006, WT_SESSION.checkpoint: [WT_VERB_DEFAULT][ERROR]: __wt_gen_drain, 169: minutes < 4 [2022/01/28 02:48:57.954] [1643338131:207][65029:0x7ff662ffd700], t, file:F00006, WT_SESSION.checkpoint: [WT_VERB_DEFAULT][ERROR]: __wt_abort, 28: aborting WiredTiger library
The issue can be reproduced using the attached test/format configuration: reproducer_config.rtf.
After some investigation (this comment and the following ones), it seems that it is linked to overflow items and direct_io being enabled.
Acceptance Criteria (Definition of Done)
We need to decide what to do in the case where we have overflow items and direct_io enabled. It is expected that it makes read/write operations slower to a point that we can reach our timeout thresholds.
Suggested solution(s) so far are:
- Increase the page size when direct_io is enabled in test/format so we are less likely to have overflow items
- is duplicated by
-
WT-8356 Fix cache stuck with dirty content during checkpoint at conn_close.
- Closed