I launch successively (nearly simultaneously) 5 nodes A B C D E using the same protocol stack and one channel to communicate between themselves.

UDP(mcast_addr=231.8.8.8;mcast_port=45578):PING(num_initial_members=4):MERGE2:FD:VERIFY_SUSPECT:pbcast.NAKACK:pbcast.STABLE:FRAG2:pbcast.GMS(shun=true):pbcast.FLUSH

Often as not, it depends on the speed/rythm between each node launch, I get 2 views, ie

{D}

and

{A B C E}

.

Merge occurs later but when it does it's a bit late for my application and I don't think I should have to handle one save in case of a real electric/network failure.

I noticed that on D I was timing out (3000ms) on during the discovery process despite having received the 4 GET_MBRS_RSP of the other nodes. Then D would decide there was no coordinator outside and become coordinator itself.

What seems to happen is D sends two GET_MBRS_REQ and A replies to both, but at the time of the first reply, A is not yet coordinator and when D receives the second response, A became coordinator but D ignores the response and doesn"t add it to its list of Responses.

I have written a workaround in Discovery.Responses method addResponse, it seems to work for my case but I am afraid it would break something else I am not aware of.

public void addResponse(PingRsp rsp) {
if(rsp == null)
return;
promise.getLock().lock();
try {
//Workaround 29/03/2010
int index = ping_rsps.indexOf(rsp);

// equivalent to does not contain.
if (index == -1)

{ ping_rsps.add(rsp); promise.getCond().signalAll(); }

else if (rsp.isCoord()) {
PingRsp pr = ping_rsps.get(index);

//Check if the already existing element is not server
if (!pr.isCoord())

{ ping_rsps.set(index, rsp); promise.getCond().signalAll(); }

}

/*if(!ping_rsps.contains(rsp))

{ ping_rsps.add(rsp); promise.getCond().signalAll(); }

*/ // Old JGroups code
}
finally

{ promise.getLock().unlock(); }

}

Regards
Renaud