-
Type:
Improvement
-
Resolution: Unresolved
-
Priority:
Major - P3
-
None
-
Affects Version/s: None
-
Component/s: None
-
Networking & Observability
-
N&O Prioritized List
-
None
-
3
-
None
-
None
-
None
-
None
-
None
-
None
-
None
A reactor thread can become overloaded if too many tasks are scheduled on it, leading to latency between scheduling and the task actually running. Sometimes, this can lead to a positive feedback loop wherein tasks begin to time out, leading to more cancellation tasks being scheduled on the reactor, leading to the reactor becoming more overwhelmed. We should short circuit in these situations to allow the reactor to continue making progress. We should also add some alerting if a reactor thread reaches a state like this.
- is related to
-
-
- Closed
-
- related to
-
SERVER-102410 Investigate performance regressions on large sharded clusters with taskExecutorPoolSize=1
-
- Open
-
-
-
- Closed
-