[3.3.1] OutOfMemory in client when slice definition modified

[joshmoore]

The addition of an object to our EventContext type definition:

diff --git a/components/blitz/resources/omero/System.ice b/components/blitz/resources/omero/System.ice
index bd84c53..9ccd744 100644
--- a/components/blitz/resources/omero/System.ice
+++ b/components/blitz/resources/omero/System.ice
@@ -40,6 +40,7 @@ module omero {
       LongList memberOfGroups;
       LongList leaderOfGroups;
       omero::model::Permissions umask;
+      omero::model::Permissions groupPermissions;
     };

lead to the following failure in a client which had not yet updated its jar:

Data Retrieval Failure: java.lang.OutOfMemoryError: Java heap space
    at IceInternal.BasicStream.readString(BasicStream.java:1301)
    at IceInternal.BasicStream.readTypeId(BasicStream.java:665)
    at Ice.ObjectImpl.__read(ObjectImpl.java:246)
    at omero.sys.EventContext.__read(EventContext.java:176)
    at IceInternal.BasicStream.readObject(BasicStream.java:1590)
    at IceInternal.BasicStream.readPendingObjects(BasicStream.java:1739)
    at omero.api._IAdminDelM.getEventContext(_IAdminDelM.java:985)

Cheers,
~Josh.

[bernard]

Hi Josh,

If you set a large Ice.MessageSizeMax for your application, this is not unexpected.

See http://www.zeroc.com/faq/requestSizeLimit.html

Best regards,
Bernard

[joshmoore]

We do increase MessageSizeMax on all our clients & servers, but only to 64MB. Would this be enough to cause an OOM in such a situtation? I ask because this is the among the first method in our handshake as well as the keepalive ping method. Having it fail due to a stale jar is certainly unfortunate.

Thanks,
~Josh

[benoit]

Hi Josh,

After having a closer look, I think it's a bug in Ice for Java. Thanks for the report, we will fix this for 3.4.

Cheers,
Benoit.

[joshmoore]

Thanks, guys.

And, of course, I meant "the addition led to ... "

~J.