Details
-
Sub-task
-
Status: Resolved
-
Major
-
Resolution: Later
-
None
-
None
-
None
-
None
Description
Servers that are participating in the LogService's quorum will die unexpectedly.
While RAFT/Ratis would be capable of recovering from this scenario, we likely do not want to do this because of the associated cost of shipping the edits for a LogStream to a new peer.
Instead, the simple solution would be for a RegionServer to create a new LogStream. This is analogous to us rolling an (hdfs file-backed) WAL when we have errors writing/syncing it.