TiKV - v5.4.3

Improvements

• Support configuring the RocksDB write stall settings to a value smaller than the flow control threshold #13467
• Support configuring the unreachable_backoff item to avoid Raftstore broadcasting too many messages after one peer becomes unreachable #13054

Bug fixes

• Fix the issue of continuous SQL execution errors in the cluster after the PD leader is switched or PD is restarted #12934
• Cause: This issue is caused by a TiKV bug that TiKV does not retry sending heartbeat information to PD client after heartbeat requests fail, until TiKV reconnects to PD client. As a result, the Region information on the failed TiKV node becomes outdated, and TiDB cannot get the latest Region information, which causes SQL execution errors.
• Affected versions: v5.3.2 and v5.4.2. This issue has been fixed in v5.3.3 and v5.4.3. If you are using v5.4.2, you can upgrade your cluster to v5.4.3.
• Workaround: In addition to upgrade, you can also restart the TiKV nodes that cannot send Region heartbeat to PD, until there is no Region heartbeat to send.
• Fix the issue that causes permission denied error when TiKV gets an error from the web identity provider and fails back to the default provider #13122
• Fix the issue that the PD client might cause deadlocks #13191
• Fix the issue that Regions might be overlapped if Raftstore is busy #13160