You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@lucene.apache.org by "Shai Erera (JIRA)" <ji...@apache.org> on 2011/01/18 12:13:43 UTC
[jira] Commented: (LUCENE-2584) Concurrency issues in
SegmentInfo.files() could lead to ConcurrentModificationException
[ https://issues.apache.org/jira/browse/LUCENE-2584?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12983142#action_12983142 ]
Shai Erera commented on LUCENE-2584:
------------------------------------
On one hand, it's good to add the files to a Set, so that we can be sure they are added uniquely. On the other hand though, if we expect files are added properly, then adding to the set is redundant. Since this code is executed once per SI instance, I think explicitly adding to a Set is better.
Note that while the assert you added will work, if someone runs without assertions he may get duplicate file names, if indeed they are added twice. I think that it's not so crucial to know that the same files was added twice, it's a very unlikely bug, but it is crucial that files() return unique names.
So can you please use a Set in the method instead of the assert (like it's done on trunk). Also, while you're at it, the method doesn't have javadocs - they appear in regular comments. Can you convert them to javadocs (there is a warning there about not modifying the returned List, but it's not visible as javadocs :).
> Concurrency issues in SegmentInfo.files() could lead to ConcurrentModificationException
> ---------------------------------------------------------------------------------------
>
> Key: LUCENE-2584
> URL: https://issues.apache.org/jira/browse/LUCENE-2584
> Project: Lucene - Java
> Issue Type: Bug
> Components: Index
> Affects Versions: 2.9, 2.9.1, 2.9.2, 2.9.3, 3.0, 3.0.1, 3.0.2
> Reporter: Alexander Kanarsky
> Priority: Minor
> Fix For: 3.1, 4.0
>
> Attachments: LUCENE-2584-branch_3x.patch, LUCENE-2584-lucene-2_9.patch, LUCENE-2584-lucene-3_0.patch
>
>
> The multi-threaded call of the files() in SegmentInfo could lead to the ConcurrentModificationException if one thread is not finished additions to the ArrayList (files) yet while the other thread already obtained it as cached (see below). This is a rare exception, but it would be nice to fix. I see the code is no longer problematic in the trunk (and others ported from flex_1458), looks it was fixed while implementing post 3.x features. The fix to 3.x and 2.9.x branches could be the same - create the files set first and populate it, and then assign to the member variable at the end of the method. This will resolve the issue. I could prepare the patch for 2.9.4 and 3.x, if needed.
> --
> INFO: [19] webapp= path=/replication params={command=fetchindex&wt=javabin} status=0 QTime=1
> Jul 30, 2010 9:13:05 AM org.apache.solr.core.SolrCore execute
> INFO: [19] webapp= path=/replication params={command=details&wt=javabin} status=0 QTime=24
> Jul 30, 2010 9:13:05 AM org.apache.solr.handler.ReplicationHandler doFetch
> SEVERE: SnapPull failed
> java.util.ConcurrentModificationException
> at java.util.AbstractList$Itr.checkForComodification(AbstractList.java:372)
> at java.util.AbstractList$Itr.next(AbstractList.java:343)
> at java.util.AbstractCollection.addAll(AbstractCollection.java:305)
> at org.apache.lucene.index.SegmentInfos.files(SegmentInfos.java:826)
> at org.apache.lucene.index.DirectoryReader$ReaderCommit.<init>(DirectoryReader.java:916)
> at org.apache.lucene.index.DirectoryReader.getIndexCommit(DirectoryReader.java:856)
> at org.apache.solr.search.SolrIndexReader.getIndexCommit(SolrIndexReader.java:454)
> at org.apache.solr.handler.SnapPuller.fetchLatestIndex(SnapPuller.java:261)
> at org.apache.solr.handler.ReplicationHandler.doFetch(ReplicationHandler.java:264)
> at org.apache.solr.handler.ReplicationHandler$1.run(ReplicationHandler.java:146)
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org