Tanuj Khurana created PHOENIX-8030:
--------------------------------------
Summary: Do not close replication log groups on RegionServer stop
Key: PHOENIX-8030
URL: https://issues.apache.org/jira/browse/PHOENIX-8030
Project: Phoenix
Issue Type: Sub-task
Reporter: Tanuj Khurana
Assignee: Tanuj Khurana
_*PhoenixRegionServerEndpoint.stop()*_ calls ReplicationLogGroup.closeAll().
HBase runs it in preStop, before it closes the regions. A live write already
committed to the WAL can then fail replication with IOException("Closed"). On a
clean region close the WAL is never replayed, so the edit is never shipped to
the standby. This can happen on any graceful RegionServer stop under HA load,
for example a rolling restart.
Fix: don't close the groups on stop. HBase's region close refuses new writes
and drains in-flight ones, including their replication sync, so every committed
write is replicated. The writer is left without a trailer at process exit, as
after a crash, which replay already handles.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
