[
https://issues.apache.org/jira/browse/HBASE-3291?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12965928#action_12965928
]
Jean-Daniel Cryans commented on HBASE-3291:
-------------------------------------------
Another issue, when restarting after the fixup happened the parent region is
offline and the daughters are never registered leaving a hole in .META. which
requires this patch:
{code}
--- a/src/main/java/org/apache/hadoop/hbase/regionserver/SplitTransaction.java
+++ b/src/main/java/org/apache/hadoop/hbase/regionserver/SplitTransaction.java
@@ -292,6 +292,7 @@ class SplitTransaction {
final RegionServerServices services, final HRegion daughter)
throws IOException, KeeperException {
if (server.isStopped() || services.isStopping()) {
+ MetaEditor.addDaughter(server.getCatalogTracker(),
daughter.getRegionInfo(), null);
LOG.info("Not opening daughter " +
daughter.getRegionInfo().getRegionNameAsString() +
" because stopping=" + services.isStopping() + ", stopped=" +
{code}
> If split happens while regionserver is going down, we can stick open.
> ---------------------------------------------------------------------
>
> Key: HBASE-3291
> URL: https://issues.apache.org/jira/browse/HBASE-3291
> Project: HBase
> Issue Type: Bug
> Reporter: stack
> Fix For: 0.90.0
>
> Attachments: closing.txt
>
>
> J-D found this one testing. He found that if a split comes in during
> shutdown of a regionserver, then the regionserver can stick open... and won't
> go down.
> We fixed a similar problem in the past where if balancer cut in during
> shutdown and assigned a regionserver an region during shutdown, we'd open it
> and it'd cause us again to stick open. We fixed that by introducing the
> 'closing' state.
> Fix for the issue j-d found is to do closing check when onlining daughters.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
