# Pd Tikv Related (Continue...)

+ `Grafana` Monitor TiDB —> KV Errors —> KV Backoff OPS


![left](./student-guide/diagram/09-TiDB-PDTiKV-03-01.png)
------------------------------------------------------------
- PD Client - PD TSO Wait/RPC Duration: These two indicators are related to TiDB getting TSO time from PD. If you find that get tso is slow, you can refer to the TSO Slow Troubleshooting Manual

- KV Request - KV Request Duration 99 by store/type: Response time of a request sent from TiDB to TiKV, by store node and by type request type, respectively; if it is high, the network latency between TiDB and TiKV may be high If some TiKV store nodes are relatively high, check if there are hot spots or load imbalance in the relevant stores. If some requests such as coprocessor time is high, check if some request such as coprocessor take high time, and further troubleshoot whether the SQL execution plan scans keys There are many factors such as thread pool bottlenecks such as unified read pool; if the commit is slow, such as kv_prewrite/kv_commit, etc., you can refer to TiKV write slow analysis TiKV related monitoring metrics.
- KV Errors - KV Backoff OPS: Counts the retryable error types and number of retries return by TiDB from TiKV/PD. Common error types are
  - PDRPC: PD access error, possibly due to network or PD Leader exception
  - RegionMiss: The visited region has been moved from the current TiKV
  - ServerBusy: TiKV is busy processing, for example, the TiKV read/write thread pool is full
  - TikVRPC: TiKV access error, possibly due to abnormal network outage or high TiKV load
  - TxNLock: write conflict
  - TxnLockFast: reads and writes conflicts
  - UpdateReader: TikV return the TiKV that is not the leader and the current leader is

- KV Errors - KV Backoff Duration: The total time for each KV request to be retried. The monitoring checks if the retry time is too long to increase the response time. If there are retries, the slow_query table also records the SQL specific retry time Backoff_time.