Merge Queue Backlog
warningResource ContentionUpdated Feb 6, 2026
Growing merge queue depth indicates data ingestion is outpacing merge capacity, leading to increased part counts and eventual performance degradation.
Sources
Technologies:
How to detect:
Monitor system.merges for number of active merges and elapsed time. Alert when merge count remains high (e.g., >5 concurrent merges) or individual merges exceed 600 seconds.
Recommended action:
Increase background_pool_size for merge threads, adjust parts_to_throw_insert to prevent excessive part creation, reduce ingestion rate during peak merge times, or optimize table partitioning to reduce merge complexity.