Hello.
I have one XMPP server in production (approx 1.5k simultaneous
connections from 170k users), running the latest jabberd2 from Ubuntu's
repos (this one : http://packages.ubuntu.com/precise/net/jabberd2).
At first, we had to face "ghost connections" (from users not properly
disconnecting). To solve that, I changed c2s.xml to have
<check>
<interval>600</interval>
<idle>0</idle>
<keepalive>600</keepalive>
</check>
So "unplugged" connections will be closed correctly after 10min. It
solved the problem.
But yesterday, in our system logs, I saw a step in established
connections : +500 connections in a few seconds.
In c2s.log I found approx. 500 connections from one IP (probably a bug
in client or malicious attempt), with 500 lines like :
Thu Jan 3 20:24:41 2013 [notice] [700] [XXX.XXX.XXX.XXX, port=2090] connect
Thu Jan 3 20:24:41 2013 [notice] [737] [XXX.XXX.XXX.XXX, port=2091] connect
Thu Jan 3 20:24:41 2013 [notice] [786] [XXX.XXX.XXX.XXX, port=2092] connect
Thu Jan 3 20:24:41 2013 [notice] [794] [XXX.XXX.XXX.XXX, port=2093] connect
Thu Jan 3 20:24:41 2013 [notice] [796] [XXX.XXX.XXX.XXX, port=2094] connect
Thu Jan 3 20:24:41 2013 [notice] [798] [XXX.XXX.XXX.XXX, port=2095] connect
....
Is goes on for 10s, then stops 10s, then they all disconnect :
Thu Jan 3 20:25:03 2013 [notice] [1482] [XXX.XXX.XXX.XXX, port=2740]
disconnect jid=unbound, packets: 0
Thu Jan 3 20:25:03 2013 [notice] [1486] [XXX.XXX.XXX.XXX, port=2744]
disconnect jid=unbound, packets: 0
Thu Jan 3 20:25:03 2013 [notice] [1490] [XXX.XXX.XXX.XXX, port=2748]
disconnect jid=unbound, packets: 0
Thu Jan 3 20:25:03 2013 [notice] [1494] [XXX.XXX.XXX.XXX, port=2752]
disconnect jid=unbound, packets: 0
Thu Jan 3 20:25:03 2013 [notice] [1498] [XXX.XXX.XXX.XXX, port=2756]
disconnect jid=unbound, packets: 0
Thu Jan 3 20:25:03 2013 [notice] [1502] [XXX.XXX.XXX.XXX, port=2760]
disconnect jid=unbound, packets: 0
....
My problem is that I still can see the 500 established connections
(netstat -n | grep 'XXX.XXX.XXX.XXX'). Even if this "connection spam"
occurred yesterday (long enough for <keepalive> to do its job)
To prevent this to happen again, I have changed c2s.xml to
<io>
<limits>
<connects>10</connects>
</limits>
</io>
But if I understand correctly, it should only decrease the number of
created connections, and it would not prevent them from being kept
established even when unnecessary.
I've looked at the code, and I think 0-packet connections are properly
added to c2s->dead_sess (and thus should be cleaned). I couldn't find
any problem, but yet I'm not very familiar with this code.
Is there a configuration item that I missed to prevent that ? Is it a
bug that has been fixed since but not yet corrected in Ubuntu's repos ?
Thanks in advance for any help. Apart from that, the server is doing
great : very low resources consumption. ^_^
--
Sylvain "Gugli" Guglielmi
Gamedev@Nadeo@Ubisoft
