hbase-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Jim Kellerman (JIRA)" <j...@apache.org>
Subject [jira] Commented: (HBASE-543) A region's state is kept in several places in the master opening the possibility for race conditions
Date Mon, 22 Dec 2008 19:06:44 GMT

    [ https://issues.apache.org/jira/browse/HBASE-543?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12658626#action_12658626

Jim Kellerman commented on HBASE-543:


> There is a bunch of redoing of state processing but no explanation
> as to why? For example: 
+   * Remove a region from the region state map.
+   * 
+   * @param info
+   */
+  public void removeRegion(HRegionInfo info) {
> Under what circumstance would I remove a region from state map?
> The region state map itself has no explanation:
+  // Needs to be SortedMap so we can specify a comparator
+  private final SortedMap<byte[], RegionState> regionState =
+    Collections.synchronizedSortedMap(
+        new TreeMap<byte[], RegionState>(Bytes.BYTES_COMPARATOR));
> The maps it would replace tried to explain what they were about.
> Nor does the new RegionState map.

Good points. The name of the map should really be regionsInTransition,
so removeRegion is called when a region completes the transitions from

unassigned -> assigned -> pending -> open

closing -> closed -> offline

closing -> closed -> unassigned -> assigned -> pending -> open

I will update the documentation and change the name of the map.

> With above said, looks like this could be an improvement in that
> state is all in one place. 

Yes, it is no longer possible for a region to be in multiple states at
once as state transitions always clear all the other states.

> Should RegionState be looking for illegal states? It doesn't seem to
> do any checking. This would be a good place to check we're doing
> transitions properly.

Good point. It should at least check for valid state transitions to
make sure that they happen in the proper order. Will update to include

> Should resetting of connection root region be done inside
> unsetRootRegion in below so the two actions are tied together: 
+      master.connection.setRootRegionLocation(null);
+      master.regionManager.unsetRootRegion();

Yes. Good idea. The connection is visible inside this method.

> Does unsetRootRegion set root region to null in regionManger?


> Maybe connection and regionManager both need an unsetRootRegion
> method (or both a setRootRegionLocation that takes null) so same
> action in two places uses similarily named methods (This stuff
> preexisted your patch). 

Yes. I think connection should have an unsetRootRegion instead of
setRootRegion(null). setRootRegion should only take a valid

> In RegionState, data members are named isUnassigned, isClosing,
> etc. Usual javabean convention is that data members are named
> unassigned, closing, etc. and that isClosing is one of the names
> getters have.

Ok, no problem.

> Why dumpState and not toString?

No good reason. I was using it for debugging, and was going to remove
it, but having a toString is a good idea.


Making the above edits now. Do you want to re-review the patch when I'm done, or should I

> A region's state is kept in several places in the master opening the possibility for
race conditions
> ----------------------------------------------------------------------------------------------------
>                 Key: HBASE-543
>                 URL: https://issues.apache.org/jira/browse/HBASE-543
>             Project: Hadoop HBase
>          Issue Type: Bug
>          Components: master
>    Affects Versions: 0.1.0, 0.1.1, 0.2.0
>            Reporter: Jim Kellerman
>            Assignee: Jim Kellerman
>             Fix For: 0.19.0
>         Attachments: 543.patch, 543.patch, recent-changes.patch, regionstate.txt
> A region's state exists in multiple maps in the RegionManager: unassignedRegions, pendingRegions,
regionsToClose, closingRegions, regionsToDelete, etc.
> One of these race conditions was found in HBASE-534.
> For HBase-0.1.x, we should just patch the holes we find.
> The ultimate solution (which requires a lot of changes in HMaster) should be applied
to HBase trunk.
> Proposed solution:
> Create a class that encapsulates a region's state and provide synchronized access to
the class that validates state changes.
> There should be a single structure that holds regions in these transitional states and
it should be a synchronized collection of some kind.

This message is automatically generated by JIRA.
You can reply to this email to add a comment to the issue online.

View raw message