MergeAllChunksOnShard runs optimistically and retries immediately if a conflicting moveChunk committed concurrently, up to 5 times. After that we fail the command with ConflictingOperationInProgress. This is detected by the AutoMergerPolicy which re-schedules the merge action immediately.
In a situation where moveChunks are constantly committing this drives up CPU utilization, specially while SERVER-78701 hasn't been fixed.
We should reduce the likelihood of merge chunks and move chunks being concurrently running for the same namespace.
- causes
-
SERVER-97681 maxTimeProcessingChunksMS for commitMergeAllChunksOnShard should default to INT_MAX
-
- Closed
-
- is related to
-
SERVER-97584 Remove the FCV check on merge_all_chunks_on_shard.js
-
- Blocked
-