Details
-
Bug
-
Status: Closed (View Workflow)
-
Blocker
-
Resolution: Fixed
-
10.6, 10.11, 11.1(EOL), 11.2, 11.4, 11.0(EOL), 11.3(EOL)
Description
While testing MDEV-32050 and its follow-up changes MDEV-33009 and MDEV-33112, axel pointed out that the InnoDB history list is not being shrunk during a workload.
Some initial debugging indicates that the purge_truncation_task (which was added in MDEV-32050) is being invoked but it is not doing anything. Also, the purge_sys.view is advancing; there are no old read views that would block the purge of history. Throttling the Sysbench workload for a few seconds would allow the history list to shrink immediately.
Attachments
Issue Links
- causes
-
MDEV-33464 Crash when innodb_max_undo_log_size is set to innodb_page_size*4294967296
- Closed
- relates to
-
MDEV-34259 Optimization in row_purge_poss_sec Function for Undo Purge Process
- Closed
-
MDEV-30628 10.6 performance regression with sustained high-connection write-only OLTP workload (55-80% degradation)
- Closed
-
MDEV-31676 Innodb history length keeps growing
- Closed
-
MDEV-32050 UNDO logs still growing for write-intensive workloads
- Closed
-
MDEV-33009 Server hangs for a long time with innodb_undo_log_truncate=ON
- Closed
-
MDEV-33112 innodb_undo_log_truncate=ON is blocking page writes
- Closed
-
MDEV-33315 InnoDB history length and undo tablespace files keep growing
- Closed