[jboss-jira] [JBoss JIRA] (JGRP-1707) MergeView: marshalling fails when subgroups are not a subset of members

[Bela Ban (JIRA)]

    [ https://issues.jboss.org/browse/JGRP-1707?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12807867#comment-12807867 ] 

Bela Ban commented on JGRP-1707:
--------------------------------

The second exception:
{noformat}
version = 6400 (3.4.0)
flags: OOB
java.lang.RuntimeException: class for magic number 9625 not found
    at org.jgroups.util.Util.readOtherAddress(Util.java:967)
    at org.jgroups.util.Util.readAddress(Util.java:911)
    at org.jgroups.util.Util.readAddresses(Util.java:1038)
    at org.jgroups.View.readFrom(View.java:190)
    at org.jgroups.MergeView.readFrom(MergeView.java:108)
    at org.jgroups.protocols.pbcast.GMS$GmsHeader.readFrom(GMS.java:1285)
    at org.jgroups.Message.readHeader(Message.java:889)
    at org.jgroups.Message.readFrom(Message.java:803)
    at org.jgroups.protocols.TP.readMessageBatch(TP.java:1796)
    at org.jgroups.tests.ParseMessages.parse(ParseMessages.java:56)
    at org.jgroups.ParserPcap.main(ParserPcap.java:37)
{noformat}
is caused by a view of a subgroup not being read correctly. The code is here:
{code}
    public void readFrom(DataInput in) throws Exception {
        super.readFrom(in);
        short len=in.readShort();
        if(len > 0) {
            subgroups=new View[len];
            for(int i=0; i < len; i++) {
                int index=in.readShort();
                long id=in.readLong();
                Address creator=get(index);
                Address[] mbrs=new Address[in.readShort()];
                for(int j=0; j < mbrs.length; j++)
                    mbrs[j]=get(in.readShort());
                View view=View.create(creator, id, mbrs);
                subgroups[i]=view;
            }
        }
    }
{code}
When {{View.create()}} throws an exception, and there are more views in the subgroups, then we have unread data left, and parsing of the next header will fail.
SOLUTION: wrap {{View.create()}} in a try-catch clause and simply skip the view if an exception is thrown
                
> MergeView: marshalling fails when subgroups are not a subset of members
> -----------------------------------------------------------------------
>
>                 Key: JGRP-1707
>                 URL: https://issues.jboss.org/browse/JGRP-1707
>             Project: JGroups
>          Issue Type: Bug
>            Reporter: Bela Ban
>            Assignee: Bela Ban
>             Fix For: 3.4
>
>
> When we have 2 partitions \{A,B,C\} and \{X,Y,Z\} and they merge, but one of the members (e.g. Z) is excluded from the merge (e.g. because it is involved in a different merge already), we create the following MergeView:
> {noformat}
> MergeView:
> view={A,B,C,X,Y}
> subgroups={A,B,C} {X,Y,Z}
> {noformat}
> Because the members of the subgroups refer to the merge view via index, the index for Z is -1, and thus - when unmarshalled - Z points to a null creator, which triggers the following stack trace:
> {noformat}
> 00631037-49819: failed handling incoming message
> java.lang.IllegalArgumentException: creator cannot be null
>         at org.jgroups.ViewId.<init>(ViewId.java:32)
>         at org.jgroups.ViewId.<init>(ViewId.java:42)
>         at org.jgroups.View.create(View.java:90)
>         at org.jgroups.MergeView.readFrom(MergeView.java:119)
>         at org.jgroups.protocols.pbcast.GMS$GmsHeader.readFrom(GMS.java:1285)
>         at org.jgroups.Message.readHeader(Message.java:889)
>         at org.jgroups.Message.readFrom(Message.java:803)
>         at org.jgroups.protocols.TP.readMessageBatch(TP.java:1796)
>         at org.jgroups.protocols.TP.receive(TP.java:1463)
>         at org.jgroups.protocols.UDP$PacketReceiver.run(UDP.java:683)
>         at java.lang.Thread.run(Unknown Source) 
> {noformat}
> SOLUTION:
> * When marshalling a MergeView, and a subgroup member has no corresponding member in {{members}}, then send the *address* instead of the index
> * OR: remove all members from subgroups which are not in {{members}}

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira