Maissel, Joe wrote:
Since deploying XMPP federation with gmail.com back in November, we
experience intermittent service problems. As far as we know, and
according to Google, the problem is experienced only by our domains -
credit-suisse.com and credit-suisse.tv. Other domains running the same
software and version we use have not experienced the issue.
The problem always manifests itself the following way. Outbound traffic
to gmail.com from credit-suisse.com, presence and messages, always
flows. Inbound traffic from gmail.com stops working for not apparent
reason. Then, after 4 to 48 hours, it all starts working normally. I
was able to capture the traffic while it was in the non-working state.
According to our vendor Cisco/Jabber, what's happening is that gmail.com
is trying to piggyback two streams on a single connection. The
Cisco/Jabber server does not support this and cuts off the connection.
What's puzzling to me is that Cisco/Jabber does not see this issue with
their own federation with gmail.com and neither do any of their
customers.
It seems that you need contacts on at least two google domains -
gmail.com, google.com, googlemail.com - to reproduce this (or you
can simply send urn:xmpp:ping stanzas to these domains).
Here's what the XMPP looks like when it fails (taken from a snoop
session). Any insight would be greatly appreciated.
Afaics your server rejects the subsequent db:result attempts. It should
not do that (search for 'piggybacking' in rfc 3920).
philipp
