Gajim fails to send keepalives with high CPU usage

[danguy]

I am experiencing random disconnects with an Openfire server that has an idle timeout of 6 minutes. Gajim is set to send keepalives every 55 seconds (ie. default, it actually sends them every 56 seconds apparently).

One cause I have identified is that high CPU usage prevents the keepalives from being sent on time.

Here is a log while running a compile job with a nice of 1:

1199657219.65: Connected to <server>, storing initial clock offset
+       73.32: Sending keepalive to <server>
+      100.21: Sending keepalive to <server>
+       56.07: Sending keepalive to <server>
+       76.23: Sending keepalive to <server>
+       68.21: Sending keepalive to <server>
+      406.09: Disconnected from <server>
+       11.63: Connected to <server>
+      135.40: Sending keepalive to <server>
+      361.26: Disconnected from <server>

[asterix]

Are you sure nothing is sent / receive ? this reset timeout counter to 0.

I think it's strange that a server disconnects on inactivity ... this means you do have to chat to stay connected ??

[danguy]

Oh, sorry for the confusion. This log was produced by some print statements added to _connect_success, send_keepalive, and disconnect in connection.py. I postprocessed the timestamps to produce deltas to aid in my understanding of the problem.

The actual code in each case looks like this:

print "%s: Disconnected from %s" % (str(time.time()), self.name)

The server stays connected so long as keepalives or chats are sent. The problem is that under high CPU load, keepalives are not always sent.

[asterix]

could you give me your full patch so I can test here? my server doesn't disconnect after some idletime, so I can't really know if keepalives are sent ...

[danguy]

--- src/common/connection.py	(revision 9231)
+++ src/common/connection.py	(working copy)
@@ -190,6 +190,7 @@
 	def disconnect(self, on_purpose=False):
 		#FIXME: set the Tune to None before disconnection per account
 		#gajim.interface.roster._music_track_changed(None, None)
+		print "%s: Disconnected from %s" % (str(time.time()), self.name)
 		self.on_purpose = on_purpose
 		self.connected = 0
 		self.time_to_reconnect = None
@@ -574,6 +575,8 @@
 		con.RegisterDisconnectHandler(self._disconnectedReconnCB)
 		log.debug(_('Connected to server %s:%s with %s') % (self._current_host['host'],
 			self._current_host['port'], con_type))
+		
+		print "%s: Connected to %s" % (str(time.time()), self.name)
 
 		name = gajim.config.get_per('accounts', self.name, 'name')
 		hostname = gajim.config.get_per('accounts', self.name, 'hostname')
@@ -1550,6 +1553,7 @@
 	def send_keepalive(self):
 		# nothing received for the last foo seconds (60 secs by default)
 		if self.connection:
+			print "%s: Sending keepalive to %s" % (str(time.time()), self.name)
 			self.connection.send(' ')
 
 	def _reconnect_alarm(self):

[alston@…]

I seem to be running into the same problem with my OpenFire? server. Gajim is a great IM client and I would like to recommend it to our users, but I won't be able to until this is resolved.

[asterix]

things have completly changed now. We use XMPP Ping to ping server instead of sending ' '. Do you still have this issue with svn or 0.12-alpha1 version?

[asterix]

reopen if you still have problem with 0.12 branch