[
https://issues.apache.org/jira/browse/HBASE-27984?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17745093#comment-17745093
]
Duo Zhang commented on HBASE-27984:
-----------------------------------
ReplicationLogCleanerBarrier is just a simple data structure, so I think we
could put it in HMaster directly and create it at the very beginning, so there
is no problem to call its method in afterReplay.
Let me provide a patch and you can see if it is OK.
Thanks.
> NPE in MigrateReplicationQueueFromZkToTableProcedure recovery
> -------------------------------------------------------------
>
> Key: HBASE-27984
> URL: https://issues.apache.org/jira/browse/HBASE-27984
> Project: HBase
> Issue Type: Bug
> Components: Replication
> Affects Versions: 3.0.0-alpha-4
> Reporter: ruanhui
> Priority: Trivial
> Fix For: 4.0.0-alpha-1
>
>
> MigrateReplicationQueueFromZkToTableProcedure will restore the disabled state
> of replication log cleaner barrier when replayed in master recovery,
> {code:java}
> @Override
> protected void afterReplay(MasterProcedureEnv env) {
> if (getCurrentState() == getInitialState()) {
> // do not need to disable log cleaner or acquire lock if we are in the
> initial state, later
> // when executing the procedure we will try to disable and acquire.
> return;
> }
> if
> (!env.getReplicationPeerManager().getReplicationLogCleanerBarrier().disable())
> {
> throw new IllegalStateException("can not disable log cleaner, this should
> not happen");
> }
> } {code}
> however the replicationPeerManager has not been initialized when replay
> procedure, which will cause a nullPointerException and master to abort.
> Maybe better to add a check after the initialization of
> replicationPeerManager to determine whether replication log cleaner barrier
> needs to be disabled ?
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
