Re: [javagroups-users] GossipRouter and 2.4.1-sp1
Brought to you by:
belaban
From: Tom N. D. <tom...@ed...> - 2007-04-19 14:05:21
|
>>In which way ? We test GossipRouter as part of our cruisecontrol and=20 >>unit tests. Do you have a use case that reproduces the issue ? If so,=20 >>please open a JIRA issue as jira.jboss.com. The use case is that we have 10 clients, 2 gossiprouters and under 2.4.1 = the gossiprouters hang when they get incoming messages. The problem I ha= ve with logging a bug is that I can't imagine it's broken for everyone el= se. If it was I'm guessing that the mailing list would be full of people= reporting problems. Although perhaps we're the first who use TCPGOSSIP = to upgrade. I'm happy to open it as an issue if you think it's best. One relevant thing here though is that (I'm guessing) your unit tests all= work on the same machine, a use case that works for us as well. >> Interestingly, this only happens for remote clients. Clients running=20 >> on the same machine as the Router are able to register and get the=20 >> member list. >That's strange, because the code is the same for local and remote=20 >clients, except local clients use TCP's loopback feature. >> Looking at the tcpdump shows me that the packets sent in each case are= =20 >> different. >How do you look at local packets ? This should *not* be possible as TCP=20 >uses loopback for local addresses, and therefore shouldn't even send=20 >packets on the network interface. You can use "tcpdump -i lo" ("tcpdump -i lo port 5065" in our case) to li= sten to the local loopback interface instead of eth0. But perhaps I'm do= ing something wrong here and it's a red herring. >> I've been looking at the sourcecode and it's apparent that the gossip=20 >> side of things (including the way in which the GossipData class is=20 >> read/written to the DataStream) has been re-written. As all we did was= =20 >> drop and replace the jar file, >For both GossipRouter *and* clients right ? Yup. we upgraded all of out clients and routers, restarted our routers t= hen restarted our clients. Since rolling back, I've reproduced the probl= em in a 2 machine development environment. Tom. =0AUnless otherwise agreed expressly in writing by a senior manager of =0A= Eduserv, this communication is to be treated as confidential and the =0Ai= nformation in it may not be used or disclosed except for the purpose=0Afo= r which it has been sent.=0AIf you have reason to believe that you are no= t the intended recipient=0Aof this communication, please contact the send= er immediately.=0ANo employee or agent is authorised to enter into any bi= nding agreement=0Aor contract on behalf of Eduserv or Eduserv Technologie= s Ltd., unless=0Athat agreement is subsequently confirmed by the conclusi= on of a written=0Acontract or the issue of a purchase order.=0AEduserv (L= imited by Guarantee) =E2=80=93 company number 3763109 - and =0AEduserv Te= chnologies Ltd =E2=80=93 company number =E2=80=93 4256630 - are both =0Ac= ompanies incorporated in England and Wales and have their registered =0Ao= ffices at Queen Anne House, 11 Charlotte Street, Bath, BA1 2NE.=0A |