-
Type: Bug
-
Resolution: Unresolved
-
Priority: Major - P3
-
None
-
Affects Version/s: None
-
Component/s: Schema Management
-
Storage Engines
-
8
-
나비 (nabi) - 2024-04-16, Nick - 2024-04-30, Megabat - 2024-05-14, 2024-05-28 - FOLLOW ON SPRINT, 2024-06-11 - Dinosaurs go rawr, 2024-06-25 - Climbing BTrees, 2024-07-09 - Nyan Cat
There has been a vast amount of HELP tickets that deal with large contention around the schemaLock affecting checkpoint. HELP-52527, HELP-53809, HELP-57159 are all examples of tickets where we see this behaviour. Common symptoms of the problem all show sweep server being active and at the same time of the checkpoint. For example:
Note: There are missing FTDC data within the help tickets because to capture statistics, it will need to fetch the schemaLock too.
There has been a long history that deals with schemaLock contention with checkpoint and sweep server. This is because the start of checkpoint requires both the SCHEMA_LOCK and table lock to fetch the dhandle list. The sweep server acts on a regular interval basis and also requires the table lock. It is postulated that the sweep server constantly requires the table lock and contends with checkpoint making the large impact of schema lock.
This ticket aims to investigate into the potential reason why we have contention on the schema lock from the help tickets.