Details
-
Improvement
-
Status: Closed
-
Major
-
Resolution: Fixed
-
None
-
None
-
None
Description
For 3C3D 2 replicas or 3 replicas dataregion, if we kill one datanode, client should keep inserting data successfully without any error.
However, currently, there is no retry logic in dispatch for write, so if the leader of one DataRegionGroup is down, the dispatch phase will fail and will return a 302 status code to Session, and meet this kind of StatusCode, Session won't retry this write request, it will throw this exception to the upper layer. You will get the error msg like the following:
more details about this improvement can be found in this feishu doc(https://timechor.feishu.cn/docx/WuVJdNpGOowEGIxvaOlcwt9mngg)