This is of course client specific behaviour and has nothing to do with Jabber the protocol,
but most Jabber clients handle this by automatically reconnecting when the socket breaks. So
in practice your connection would break for a few seconds and then be back online with the
next server.
So Jabber failover works (almost) as good as email (because jabber messages are not
transactions in the same way email are, but close enough). But there are many ways a server
can fail and a server that is otherwise dead but accepts logins properly would trap most
(all?) existing Jabber clients. HA is hard in practice.
Posted May 29, 2008 18:22 UTC (Thu) by martinfick (subscriber, #4455)
[Link]
Yes, so jabber client failover will probably work, and as I mentioned, in most cases this
would be fine even if the failover had to be manual (i.e. log out and then back in). I agree
that HA is hard and there could be other weird server behavior as you mention, but I believe
that for email this kind of weird flaky behavior is still likely to at least cause the
alternate MX record to be used. There is no reason that a jabber based authentication
protocol could not be smart enough and even required by the specification to do this.
That still leaves the delegation issue. How do I delegate this to a friend or an ISP (only in
case of failure) so that if my house burns down their server can answer requests to my jabber
id? What naming scheme would allow this?