On Mon, 2009-02-16 at 11:41 +0300, Nikolay Kondratyev wrote:
> Hi all,
> 
> The main problem I found when testing HA setup was as following:
> If master host fails, all established calls can not be correctly terminated
> and can not be transferred.
> The problem (as I understood it) was that sipx sets Record-Route header,
> pointing to master host IP ADDRESS and not to dns name, which is common for
> both master and distributed/slave host.
> And all UA's will send re-Invite and Bye messages to Master host IP address.
> 
> I asked about it in the list, and was answered that it is not possible to
> use dns name in the Record-Route header because of the problems of some
> widely used phones (I don't remember which phones were mentioned).
> 
> So the question is: does 3.10.3 HA sipx still use ip address in Record-Route
> header? And what about 4.0? 

Yes and yes.

> If not, could it be a configurable option, so sipx will use dns name in the
> Record-Route header but only for UA's that are known to work ok with such
> header?

There is an 'unsupported' configuration parameter that changes this, but
it isn't sensitive to the UA, so it would be safe to change only if all
UAs in your system could cope with it correctly.  By 'unsupported', I
mean:

      * not available through the web ui
      * if you change it and things break, we'll tell you to change it
        back (since we don't test it changed, it's possible that it
        would have unpleasant side effects, and right now those would
        not be considered bugs)

We should probably retest phones some time to be able to characterize
this more precisely.  Even if we change the Record-Route though, many
phones will 'pin' themselves to the proxy to which they sent their
registration; if the phone does that, nothing can help it.

> Regarding the overall impression of HA system: 
> HA system imposes some requirements on the phones.

Only that they support DNS SRV records as the standards specify.

> Media services are still not redundant, auto-attendant, voice mail, ACD
> announcements will not work when master fails. BTW, does 4.0 somehow improve
> that?

No, not yet.  This is something we'll look into post 4.0

> The most critical "points of failure" are disks and network interfaces,
> which can be made redundant using operating system capabilitites...

Yes, and those can be made to work now.

_______________________________________________
sipx-users mailing list
[email protected]
List Archive: http://list.sipfoundry.org/archive/sipx-users
Unsubscribe: http://list.sipfoundry.org/mailman/listinfo/sipx-users

Reply via email to