[
https://issues.apache.org/jira/browse/HBASE-19415?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16276627#comment-16276627
]
Chia-Ping Tsai commented on HBASE-19415:
----------------------------------------
The story in branch-1 is similar with master. The reassign will generate the
put request to update user table info in meta table. The put request about the
user table info will be sent to the suspended rs (the wal sync make rs kill
itself) since the rs taking over the region from suspend rs doesn't update it's
region cache before submitting the request.
{code:title=HRegionServer.java}
@Override
public void postOpenDeployTasks(final PostOpenDeployContext context)
throws KeeperException, IOException {
...
// Update ZK, or META
if (r.getRegionInfo().isMetaRegion()) {
MetaTableLocator.setMetaLocation(getZooKeeper(), serverName,
r.getRegionInfo().getReplicaId(),
State.OPEN);
} else if (useZKForAssignment) {
MetaTableAccessor.updateRegionLocation(getConnection(), r.getRegionInfo(),
this.serverName, openSeqNum, masterSystemTime);
}
if (!useZKForAssignment && !reportRegionStateTransition(new
RegionStateTransitionContext(
TransitionCode.OPENED, openSeqNum, masterSystemTime,
r.getRegionInfo()))) {
throw new IOException("Failed to report opened region to master: "
+ r.getRegionInfo().getRegionNameAsString());
}
...
}
{code}
Looped the tests with [~tedyu]'s solution 50 times. All pass. [~tedyu] Do you
want to take this over? Or I can do the backport if you have no free cycle.
> Backport HBASE-19056 to branch-1.x
> ----------------------------------
>
> Key: HBASE-19415
> URL: https://issues.apache.org/jira/browse/HBASE-19415
> Project: HBase
> Issue Type: Bug
> Components: test
> Affects Versions: 1.4.0, 1.5.0
> Reporter: Duo Zhang
>
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
