[ http://jira.jboss.com/jira/browse/JGRP-647?page=comments#action_12396167 ] Bela Ban commented on JGRP-647: ------------------------------- The real question here is why is this logged repeatedly ? When a channel is closed/disconnected, the sender thread is interrupted, but only ONCE. Did this get logged continuously or not ? -- This message is automatically generated by JIRA. - If you think it was sent incorrectly contact one of the administrators: http://jira.jboss.com/jira/secure/Administrators.jspa - For more information on JIRA, see: http://www.atlassian.com/software/jira

[ http://jira.jboss.com/jira/browse/JGRP-647?page=comments#action_12398430 ] Galder Zamarreno commented on JGRP-647: --------------------------------------- I had a look at some TRACE logs from a customer (I've asked the customer to see whether it'd be Ok to post them here) and I think I understand what's happening: I had a closer look to this and seems like PingSender thread is started and stopped whenever new members need to be found, see PingWaiter.findInitialMembers(): public Vector findInitialMembers() { ... ping_sender.start(); start_time=System.currentTimeMillis(); time_to_wait=timeout; try { while(rsps.size() < num_rsps && time_to_wait > 0 && thread != null && Thread.currentThread().equals(thread)) { if(log.isTraceEnabled()) // +++ remove log.trace(new StringBuffer("waiting for initial members: time_to_wait=").append(time_to_wait) .append(", got ").append(rsps.size()).append(" rsps")); try { rsps.wait(time_to_wait); } catch(InterruptedException intex) { } catch(Exception e) { log.error("got an exception waiting for responses", e); } time_to_wait=timeout - (System.currentTimeMillis() - start_time); } if(log.isTraceEnabled()) log.trace(new StringBuffer("initial mbrs are ").append(rsps)); return new Vector(rsps); } finally { if(ping_sender != null) ping_sender.stop(); } ... or "initial mbrs are ", so I suspect that the ping waiter might be interrupted while waiting even before the sender sent the message. This leads to ping sender being interrupted in ping_sender.stop() while it's sending the message. The Exception itself probably can be ignored as the ping sender will try to start the thread again when it wants to find new members again. Looking at the log, there's a continuous attempt to send GET_MBRS_REQ which comes from MERGE2.run(), which is periodically trying to find initial members. My concern is, why do we start/interrupt the ping sender thread whenever the node tries to find new members? PING phase only happens once, but finding initial members via MERGE2 happens regularly. Wouldn't it be better to have a thread running all the time and react based on notifications? Interrupting and starting a thread whenever we need to find new members seems to be rather costly. Thoughts? -- This message is automatically generated by JIRA. - If you think it was sent incorrectly contact one of the administrators: http://jira.jboss.com/jira/secure/Administrators.jspa - For more information on JIRA, see: http://www.atlassian.com/software/jira

[ http://jira.jboss.com/jira/browse/JGRP-647?page=comments#action_12398751 ] Bela Ban commented on JGRP-647: ------------------------------- - The exception is now caught - didn't cause any damage though - I checked 2.6 and 2.7: the code was changed there, and we don't interrupt any sender anymore, but instead cancel a task. This will not cause error messages such as the one above to appear though -- This message is automatically generated by JIRA. - If you think it was sent incorrectly contact one of the administrators: http://jira.jboss.com/jira/secure/Administrators.jspa - For more information on JIRA, see: http://www.atlassian.com/software/jira