# Write Jitter Case Study Summary


![left](./student-guide/diagram/03-25-1.png)

------------------------------------------------------------

- Slow insert writes confirmed from the gRPC message Duration kv_prewrite delay jitter that an issue occurred in TiKV.

- Examine the relevant write latency monitoring under Raft IO/Raft Logs and found a significant increase in Raft log submission (copying logs to many replica) for a TiKV, indicating that the problem may be at the storage layer, or due to network or traffic flow control.

- Check RocksDB's write duration write latency is also significantly higher than normal, and the latency is mainly due to the disk itself, i.e. calling fsync to write the system cache buffer to disk.

- In summary, our judgment is that disk issues with the TiKV server are causing write jitter. The problem was resolved after we took the TiKV node offline and replaced it.