jackrabbit-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Connor, Brett (LNG-TWY)" <Brett.Con...@lexisnexis.co.uk>
Subject NPE in VersionIteratorImpl.addVersion in cluster
Date Tue, 01 Jul 2008 15:21:27 GMT
I have a simple cluster with two nodes, bsc_1 and bsc_2, configured with
bundle.MSSqlPersistenceManager and datastore (on a shared filesystem).
This is using Jackrabbit 1.4.3.

In my test one node created a document, the other fairly eagerly spotted
the new document (another communication channel) and tried to find some
information about it. I had several callstacks as below for a few
seconds before it settled down and returned the informaion correctly:

	at
org.apache.jackrabbit.core.version.VersionIteratorImpl.addVersion(Versio
nIteratorImpl.java:147)
	at org.apache.jackrabbit.core.version.VersionIteratorImpl.
(VersionIteratorImpl.java:68)
	at
org.apache.jackrabbit.core.version.VersionHistoryImpl.getAllVersions(Ver
sionHistoryImpl.java:95)

Several other properties were read ok, it's not until it gets to the
versionhistory that it goes pear shaped.

The method in question traverses the version graph, internally adding
version node successors to a working queue to build the iteration order.
Somewhere a null item is being added to the work queue. I don't know
which because putting the debugger on it obviously changes the timing.
The method itself is synchronized, but my guess is that something (in
the journalling code?) that updates this isn't. Looks like it's pulling
null out of one of the node successors array most likely but I haven't
yet verified this.

Has anyone else seen this? I can't see a jira bug matching this.

Brett

REED ELSEVIER (UK) LIMITED - Registered office - 1-3 STRAND, LONDON WC2N 5JR
Registered in England - Company No. 02746621


Mime
View raw message