Michael Stack created HBASE-24070:
-------------------------------------
Summary: Master shutdown rpc call may not complete
Key: HBASE-24070
URL: https://issues.apache.org/jira/browse/HBASE-24070
Project: HBase
Issue Type: Sub-task
Components: master
Reporter: Michael Stack
Debugging the parent issue and discussion over in the adjacent test fix issue
HBASE-24052 and in its PR exposes a problem calling shutdown asynchronously.
The shutdown runs inline with the rpc call which should change (as suggested by
[~bharathv]). Quoting from a Bharath comment in the the HBASE-24052 PR:
bq. I think internally it is masking the problem of "shutdown()" implementation
being synchronous on the server side. Ideally, I'd think it should be async on
the server side too, meaning the rpc call just sets a flag (startShuttingDown =
true) which is read by multiple running threads and they begin the tear down
rather than doing it in the context of the rpc thread. We already do this
partially but the main driver of master shutdown still runs in the rpc context.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
