hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "ramkrishna.s.vasudevan (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HBASE-19017) EnableTableProcedure is not retaining the assignments
Date Tue, 17 Oct 2017 05:32:00 GMT

    [ https://issues.apache.org/jira/browse/HBASE-19017?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16207035#comment-16207035
] 

ramkrishna.s.vasudevan commented on HBASE-19017:
------------------------------------------------

[~easyliangjob]
Thanks for the comment and review. So will commit this now. The failed test case is unrelated.

> EnableTableProcedure is not retaining the assignments
> -----------------------------------------------------
>
>                 Key: HBASE-19017
>                 URL: https://issues.apache.org/jira/browse/HBASE-19017
>             Project: HBase
>          Issue Type: Bug
>          Components: Region Assignment
>    Affects Versions: 2.0.0-alpha-3
>            Reporter: ramkrishna.s.vasudevan
>            Assignee: ramkrishna.s.vasudevan
>             Fix For: 2.0.0-beta-1
>
>         Attachments: HBASE-19017.patch
>
>
> Found this while working on HBASE-18946. In branch-1.4 when ever we do enable table we
try retain assignment. 
> But in branch-2 and trunk the EnableTableProcedure tries to get the location from the
existing regionNode. It always returns null because while doing region CLOSE while disabling
a table, the regionNode's 'regionLocation' is made NULL but the 'lastHost' is actually having
the servername where the region was hosted. But on trying assignment again we try to see what
was the last RegionLocation and not the 'lastHost' and we go ahead with new assignment.
> On region CLOSE while disable table
> {code}
> public void markRegionAsClosed(final RegionStateNode regionNode) throws IOException {
>     final RegionInfo hri = regionNode.getRegionInfo();
>     synchronized (regionNode) {
>       State state = regionNode.transitionState(State.CLOSED, RegionStates.STATES_EXPECTED_ON_CLOSE);
>       regionStates.removeRegionFromServer(regionNode.getRegionLocation(), regionNode);
>       regionNode.setLastHost(regionNode.getRegionLocation());
>       regionNode.setRegionLocation(null);
>       regionStateStore.updateRegionLocation(regionNode.getRegionInfo(), state,
>         regionNode.getRegionLocation()/*null*/, regionNode.getLastHost(),
>         HConstants.NO_SEQNUM, regionNode.getProcedure().getProcId());
>       sendRegionClosedNotification(hri);
>     }
> {code}
> In AssignProcedure
> {code}
>     ServerName lastRegionLocation = regionNode.offline();
> {code}
> {code}
> public ServerName setRegionLocation(final ServerName serverName) {
>       ServerName lastRegionLocation = this.regionLocation;
>       if (LOG.isTraceEnabled() && serverName == null) {
>         LOG.trace("Tracking when we are set to null " + this, new Throwable("TRACE"));
>       }
>       this.regionLocation = serverName;
>       this.lastUpdate = EnvironmentEdgeManager.currentTime();
>       return lastRegionLocation;
>     }
> {code}
> So further code in AssignProcedure
> {code}
>  boolean retain = false;
>     if (!forceNewPlan) {
>       if (this.targetServer != null) {
>         retain = targetServer.equals(lastRegionLocation);
>         regionNode.setRegionLocation(targetServer);
>       } else {
>         if (lastRegionLocation != null) {
>           // Try and keep the location we had before we offlined.
>           retain = true;
>           regionNode.setRegionLocation(lastRegionLocation);
>         }
>       }
>     }
> {code}
> Tries to do retainAssignment but fails because lastRegionLocation is always null.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

Mime
View raw message