Should we shutdown the bookie or retry when registration failure

Open hezhangjian opened this issue 3 years ago • 1 comments

about the flaky test https://github.com/apache/bookkeeper/issues/3206

I have figured out why testBookieServerZKExpireBehaviour falky.

It's due to Registration task, bookie will retry to registration when zk session expires. If the registration fail, it will cause the bookie shutdown ExitCode.ZK_REG_FAIL.

@dlg99 @eolivelli @merlimat PTAL

Apr 28 '22 06:04 hezhangjian

de-facto doc: https://github.com/apache/bookkeeper/blob/6b03b012f731536af56cc63ba29511f09994d11b/conf/bk_server.conf#L804-L809

BK will exit on SessionExpiredException from zk client; it is hard to recover from it (have to re-create the zk client + redo all initialization steps of all subcomponents) so better to exit and let k8s/supervisord restart the service.

Jul 21 '22 03:07 dlg99