Details
-
Improvement
-
Status: Closed
-
Major - P3
-
Resolution: Fixed
-
None
-
None
-
None
-
Fully Compatible
-
v6.0, v5.0, v4.4
-
Repl 2022-05-02, Repl 2022-05-16
Description
The combination of ending oplog batches when the commit point moves and oplog exhaust results in very small oplog batches when the secondaries are nearly caught up. This results in more iops on the secondary. This ticket will add a knob to delay oplog batching when we're caught up, which will improve batching at the expense of some w:majority latency. Experimentally a modest delay of 20ms was enough to restore full-sized batches when the oplog is being continuously written (e.g. multi-update). The default for this parameter will be 0.
Attachments
Issue Links
- is documented by
-
DOCS-15338 Investigate changes in SERVER-65723: Add tunable parameter to improve batching on secondaries
-
- Closed
-
- is related to
-
SERVER-54939 Investigate secondary batching behavior in v4.4
-
- Closed
-
- related to
-
SERVER-53667 High rate of journal flushes on secondary in 4.4
-
- Closed
-