tikv-performance

To troubleshoot TiKV latency spikes and ServerIsBusy errors, check for gRPC thread hotspots, disk IOPS exhaustion from internal SQL, or flow-control throttling. Apply metric checklists to quickly isolate whether the bottleneck is connection imbalance or scheduler saturation.

TiKV flow control throttling is typically caused by write surges and level-0 SST file accumulation. You can resolve this performance issue by diagnosing the write load, checking scheduler backlogs, and applying workarounds to alleviate the immediate disk pressure and stabilize cluster throughput.

Fix TiKV disk IOPS exhaustion by identifying internal SQL commands like ANALYZE or large cold-data scans consuming disk bandwidth. Apply diagnostic guidance to throttle or reschedule these operations, preventing resource saturation and restoring normal IOPS availability for TiKV.

A large region causes TiKV performance degradation through hot-row updates and load-based split failures. Mitigate this scheduling side effect by identifying oversized regions and applying targeted split strategies to balance the load and eliminate the resulting RPC long-tail latency.

Yes, this guidance applies to on-call engineering scenarios involving TiDB-to-TiKV RPC long-tail latency. It provides battle-tested workarounds and version-specific known issue references to accelerate root cause identification for unexplained performance bottlenecks without obvious crash symptoms.

Diagnosing TiKV performance bottlenecks without crash symptoms is limited by the need to correlate multiple metrics like gRPC latency, disk IOPS, and flow control. Complex interactions like scheduling-induced side effects require structured checklists to avoid missing subtle root causes.