Too many failed slow query during TiKV graceful shutdown

[crazycs520]

Bug Report

Please answer these questions before submitting your issue. Thanks!

1. Minimal reproduce step (Required)

1. Deploy a TiDB cluster.
2. Run sysbench oltp_read_write workload.
3. Stop 1 TiKV store.

2. What did you expect to see? (Required)

Query execution time should less than 30s.

3. What did you see instead (Required)

Many slow query execution time more than 60s.

4. What is your TiDB version? (Required)

nightly: ce74191