-
Type: Improvement
-
Resolution: Unresolved
-
Priority: Major - P3
-
None
-
Affects Version/s: None
-
Component/s: None
-
(copied to CRM)
-
StorEng - Refinement Pipeline
Summary
In WT there's the notion of choosing the block allocation algorithm to use for a given table. However this can only be set during creation in the WT_SESSION::create method. We wish to be able to set it afterwards with a call to WT_SESSION::alter.
Motivation
- Does this affect any team outside of WT?
Yes, it affects MongoDB. - How likely is it that this use case or problem will occur?
Only in cases where we want to change the block allocation algorithm after creating the table.
- If the problem does occur, what are the consequences and how severe are they?
In cases with a very high Oplog churn rate the Oplog can cause file extensions. WT can then proceed to select pages near the end of the file even if there are large chunks of empty blocks earlier in the file. This blocks file truncation as there's always data near the end. Selecting a different block allocation algorithm would help in those cases.
- Is this issue urgent?
No
Acceptance Criteria (Definition of Done)
When we can select the block allocation algorithm to use for a table at runtime.{}
- related to
-
SERVER-70884 Investigate using "first" fit block allocation for the Oplog
- Open