[jboss-jira] [JBoss JIRA] Commented: (JBMESSAGING-1763) Bisocket connection won't be closed if pulling out the ethernet cable between client and server. The failure detection code won't close the failure connection, as a result, the subsequent requests will hang after connection account exceeds the threshold

Thu Apr 22 12:07:10 EDT 2010

    [ https://jira.jboss.org/jira/browse/JBMESSAGING-1763?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12527091#action_12527091 ] 

mingjun jiang commented on JBMESSAGING-1763:
--------------------------------------------

https://jira.jboss.org/jira/browse/JBMESSAGING-1268 
https://jira.jboss.org/jira/browse/JBREM-947 
https://jira.jboss.org/jira/browse/JBREM-949 

In the JBREM-947 and JBREM-949, the remoting wrongly depend on the socket timeout for failure detection, the bug reporter said 'Remoting should not be dependent on the socket timeout for failure detection, the connetion validation and socket timeout should be possible to be configured separately'. According to the JIRA, this issue had been fixed in Remoting 2.2.2. SP8 and 2.4.0.CR2 (Pinto), since it is said 'JBREM-947 - is the proper fix in remoting which will allow us to configure timeouts and validation interval separately'. (stated in JBMESSAGING-1268) 

However, according to our test, this issue was not fixed properly. We suspect the Remoting 2.2.3 still wrongly depend on the socket timeout for failure connection detection. So, we tried to change the timeout of bisocket transport from 0 (infinite) to a lower value e.g. 10 seconds, then we repeat the test steps again and observe the jboss web console, fortunately, we found the failure connection will be closed after 10 seconds. In JBREM-947, a daemon thread and validatorPingTimeout parameter are mentioned, we think daemon thread should use parameter validatorPingPeriod and validatorPingTimeout to close the failure connection instead of socket timeout.

> Bisocket connection won't be closed if pulling out the ethernet cable between client and server. The failure detection code won't close the failure connection, as a result, the subsequent requests will hang after connection account exceeds the threshold
> -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
>
>                 Key: JBMESSAGING-1763
>                 URL: https://jira.jboss.org/jira/browse/JBMESSAGING-1763
>             Project: JBoss Messaging
>          Issue Type: Bug
>          Components: JMS Remoting
>    Affects Versions: 1.4.5.GA, 1.4.6.GA
>         Environment: OS: Windows Server 2003. JBoss App Server 4.2.3.GA, JBoss Messaging 1.4.5 GA, JBoss Remoting 2.2.3 SP1 
>            Reporter: mingjun jiang
>            Assignee: Howard Gao
>             Fix For: 1.4.6.GA.SP1, 1.4.7.GA
>
>
> We are using JBoss App Server 4.2.3.GA, JBoss Messaging 1.4.5 GA and JBoss Remoting 2.2.3 SP1. In our application, there are a lot of Message listeners running on the client side, these message listeners will receive messages from queue/topic deployed in JBoss Messaging 
> Configuration: 
> We created our own JMS Connection factory which uses the default remoting connector. As you know, the default remoting connector is configured to use the bisocket transport. We didn't change the default value of the remoting connector 
> During we run our application, we open the JBoss web console to monitor the value of currentClientPoolSize under "Jboss.remoting" JMX MBean. 
>   
> How to reproduce this issue: 
> 1. Run 5 message listeners in the client side to receive messages from JBoss Messaging, then we observe the value of currentClientPoolSize is 10 
> 2. After processing several messages, we manually pull out the ethernet cable. The value of currentClientPoolSize is still 10. 
> 3. We run another 5 message listeners in client side, then the value of currentClientPoolSize will become 20 
> 4. After we do the same operations above several times, the value of currentClientPoolSize will increase continuously. Once the value of currentClientPoolSize is equal to the MaxPoolSize, then the subsequent incoming client requests will hang, and we will encounter the following exception in server side 
>        2009-10-20 18:08:09,655 ERROR [org.jboss.remoting.transport.socket.ServerThread] Worker thread initialization failure 
> java.net.SocketException: Connection reset 
>       at java.net.SocketInputStream.read(SocketInputStream.java:168) 
>       at java.io.BufferedInputStream.fill(BufferedInputStream.java:218) 
>       at java.io.BufferedInputStream.read(BufferedInputStream.java:235) 
>       at java.io.FilterInputStream.read(FilterInputStream.java:66) 
>       at org.jboss.remoting.transport.socket.ServerThread.readVersion(ServerThread.java:859) 
>       at org.jboss.remoting.transport.socket.ServerThread.processInvocation(ServerThread.java:545) 
>       at org.jboss.remoting.transport.socket.ServerThread.dorun(ServerThread.java:406) 
>       at org.jboss.remoting.transport.socket.ServerThread.run(ServerThread.java:173) 
> Conclusion: JBoss Messaging won't close the failure connections if they are caused by manually pulling out ethernet cable. As a result, the value of currentClientPoolSize will increase continuously and finally the new client requests will hang 
> Note: If we killed the process of message listener in client side, then the value of currentClientPoolSize will decrease to 0 immediately, it seems that the server could detect the failure connection and perform the corresponding resource releasing. 

-- 
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators: https://jira.jboss.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see: http://www.atlassian.com/software/jira