🚀 go-pugleaf

Yesterday I searched for the reason our WAN connections didn't come
up for over three hours for the following reasons.

The configuration is a AGS running 9.0(1) / 9.0(2.6) / 9.1(1.3) 
(happens in either cases). The WAN connections run over a X.25
interface with multiple SVCs. We use IGRP between the ciscos.

The router was reloaded about six weeks ago.

After that reload I added dynamically a SVC to this interface pointing
to a site outside our company. For some reason the other site's
administrator requested I had give our X.25 interface a secondary
address with an IP number the others could deal with. So the
config looked like

	interface serial 3
	encapsulation x25
	ip address 192.125.1.9 255.255.255.252 secondary
	ip address 192.125.253.2 255.255.255.0

Yesterday I had to power-on-reset the router after a hardware upgrade.
After that reload all WAN connections via X.25 didn't work. The
reason I found was that all IGRP pakets send via broadcast on
this interface had the secondary address as sender. (Before the
reload but with the same configuration the primary address was
was used to send the updates). But the remote routers were not prepared
to accept these IGRP updates because they are on net 192.125.253.0
only (they don't react on broadcasts with an other network address),
so our dynamic routing on this interface failed completely.

My question: Bug or feature? Or, more generally: If an interface
has one primary and more than one secondary address is it predictable
which address is used in the IGRP updates as sender?

Thanks in advance.

cu,
Peter