Hai Zhou created RATIS-2684:
-------------------------------
Summary: Add first-class config and docs for Ratis
raft.server.close.threshold
Key: RATIS-2684
URL: https://issues.apache.org/jira/browse/RATIS-2684
Project: Ratis
Issue Type: Improvement
Reporter: Hai Zhou
`raft.server.close.threshold` (default 60s) controls when the Ratis
`JvmPauseMonitor`
shuts down the local raft server after a JVM/host pause, but in Celeborn
this knob is
effectively invisible: there is no dedicated `CelebornConf` entry (unlike
other Ratis
params such as `celeborn.master.ha.ratis.raft.rpc.request.timeout`), and
the only way to
tune it is the undocumented `celeborn.ratis.*` passthrough
(`celeborn.ratis.raft.server.close.threshold=180s`), discoverable only by
reading the
source. No doc mentions that a long JVM pause makes a master leave its
raft group until
the process restarts.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
