I have only two nodes in my enviroment. kill one node and then restart, the node failed to join groups occasionally.
always print following message:
[org.jgroups.protocols.pbcast.GMS] join(PEK-WKST1DBTB-11909) sent to pek-wkstdkktk-14320 timed out (after 30000 ms), retrying

after debug, it seems the other alive node didn't send out join response, because the alive node holds a ParticipantGmsImpl. (suppose it should be the coordinate)

I uploaded my debug result in
http://ww4.sinaimg.cn/mw600/649a9869jw1dsd6mtpz9qj.jpg
http://ww2.sinaimg.cn/mw600/649a9869jw1dsd6ty4te7j.jpg
you can see "pek-wkstdkktk-14320" is the coordinator but the gms impl is ParticipantGmsImpl