Tanuj Khurana created PHOENIX-8054:
--------------------------------------
Summary: Fail replication stragglers back to the client instead of
aborting the RegionServer during a consistent failover
Key: PHOENIX-8054
URL: https://issues.apache.org/jira/browse/PHOENIX-8054
Project: Phoenix
Issue Type: Sub-task
Reporter: Tanuj Khurana
Assignee: Tanuj Khurana
During a consistent failover the standby fences the active cluster's
replication log writer, by recovering the lease on the log file before reading
it. A write whose replication sync fails after that point cannot fall back to
STORE_AND_FORWARD: the transition is rejected in the failover states. Today the
rejection is treated as fatal and the RegionServer aborts. One straggling write
therefore takes down the whole RS during a planned failover. It also forces a
crash recovery and WAL replay for every region on that RS while the failover is
in progress.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)