[
https://issues.apache.org/jira/browse/HBASE-14275?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14876415#comment-14876415
]
Andrew Purtell commented on HBASE-14275:
----------------------------------------
Go to this commit in 0.98, run TestAssignmentManagerOnCluster a few times, we
will eventually get stuck cleaning up at the end of
testSSHWhenDisablingTableRegionsInOpeningOrPendingOpenState.
The deleteTable at line 1581 of HTU called from line 678 of the test gets
stuck. This only happens when above in the test, at line 639, META is moved
when colocated with the test table on the victim regionserver. If I add this:
{noformat}
@@ -645,6 +645,8 @@ public class TestAssignmentManagerOnCluster {
.move(HRegionInfo.FIRST_META_REGIONINFO.getEncodedNameAsBytes(),
Bytes.toBytes(rs.getServerName().getServerName()));
am.waitForAssignment(HRegionInfo.FIRST_META_REGIONINFO);
+ // Drop the cached location for META
+ TEST_UTIL.getHBaseAdmin().getConnection().clearCaches(metaServerName);
}
am.regionOffline(hri);
{noformat}
then the test passes reliably.
Looks like when using MetaReader to wait until all regions are deleted,
deleteTable won't attempt to relocate META should its cached location be stale.
> Backport to 0.98 HBASE-10785 Metas own location should be cached
> ----------------------------------------------------------------
>
> Key: HBASE-14275
> URL: https://issues.apache.org/jira/browse/HBASE-14275
> Project: HBase
> Issue Type: Improvement
> Reporter: Jerry He
> Assignee: Jerry He
> Fix For: 0.98.14
>
> Attachments: HBASE-14275-0.98.patch
>
>
> We've seen similar problem reported on 0.98.
> It is good improvement to have.
> This will cover HBASE-10785 and the a later HBASE-11332.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
