[jboss-jira] [JBoss JIRA] Updated: (JGRP-259) Unidentified errors with TCP/TCPPING/MERGE2 protocol stack.

Unidentified errors with TCP/TCPPING/MERGE2 protocol stack. ----------------------------------------------------------- Key: JGRP-259 URL: http://jira.jboss.com/jira/browse/JGRP-259 Project: JGroups Issue Type: Bug Affects Versions: 2.2.9.2 Environment: Debian GNU/Linux 2.4.27, JDK 1.5.0_05 Reporter: Tomasz Skutnik Assigned To: Bela Ban Fix For: 2.4 Attachments: bla.java, Test1.java I've encountered connect/disconnect problems during my experiments with TCP/TCPPING/MERGE2 protocol stack. I run client program (see attachment) on 2 different hosts (witnin two VMWare virtual hosts). One of the hosts (h1) plays role of well-known host from which initial membership will be retrieved, the other plays role of plain client. I use following command lines to run client programs: - on host h1: $ java -cp \ jgroups-all-2.2.9.2.jar:concurrent.jar:commons-logging-1.1.jar:\ classes test.Test1 "h1" "h1[7800]" - on host h2 $ java -cp \ jgroups-all-2.2.9.2.jar:concurrent.jar:commons-logging-1.1.jar:\ classes test.Test1 "h2" "h1[7800]" When I run client programs in h1, h2 order everyting works fine. But if I run them in h2, h1 order, or if I kill h1 and then rerun it something weird happens: merge of two disjoint groups happens (MergeView is transmitted), but then followin exception occurs: 2006-07-10 12:01:32 org.jgroups.Message readHeader SEVERE: magic number 306192 is not available in magic map 2006-07-10 12:01:32 org.jgroups.protocols.TP handleIncomingPacket SEVERE: failed unmarshalling message java.io.IOException: failed read header: java.lang.NullPointerException at org.jgroups.Message.readHeader(Message.java:697) at org.jgroups.Message.readFrom(Message.java:614) at org.jgroups.protocols.TP.bufferToMessage(TP.java:974) at org.jgroups.protocols.TP.handleIncomingPacket(TP.java:830) at org.jgroups.protocols.TP.receive(TP.java:781) at org.jgroups.protocols.TCP.receive(TCP.java:226) at org.jgroups.blocks.ConnectionTable.receive(ConnectionTable.java:471) at org.jgroups.blocks.ConnectionTable$Connection.run(ConnectionTable.java:813) at java.lang.Thread.run(Thread.java:595) Because there's no magic number in magic map (whatever that means) it causes NPE within org.jgroups.Message class (line 687). >From now on everything is going downhill - jgroups logs infinite stream of retransmition errors, e.g.: SEVERE: magic number 306192 is not available in magic map 2006-07-10 12:01:54 org.jgroups.protocols.pbcast.NAKACK handleXmitRsp SEVERE: message did not contain a list (LinkedList) of retransmitted messages: java.io.IOException: failed read header: java.lang.NullPointerException This flow of exceptions does not stop until all clients (h1 and h2) are killed - stopping h1 does not stop exceptions on h2 and vice versa.