> Reinstalling the route doesn't unwedge it. Neither does replumbing the
> interface.
Probably because both the interface routes and the default route point to the
same something internally (whether it's an ipif, an ire, or an ill is another
question).
> However, when I do route flush, unplumb, plumb, route add default <gateway>
> it comes comes back to life.
Yeah. I'm wondering if there's some sort of livelock or entry-protocol
mismatch that's going on?
> The interesting thing to me is that after I do that, I can no longer break
> the routing. I even ran it all the way up to about 35k connections for 2
> hours and it didn't falter once. Does that make any sense?
It might if it's some sort of race condition or some sort of lock-entry
mismatch that happens in certain cases.
I'm assuming, BTW, you're on a multi-CPU box (where a single multi-core CPU
counts as multi-CPU).
> As I said, my test boxes come up with initial routing information from
> /etc/defaultrouter.
You're also not running any routing daemons by accident on TOP of that too,
are you? routeadm(1m) will show this, especially in 58.
> I'm about to test on the last developer edition, which is build 55 i think,
> and I'm also going to test on the last Solaris 10 release.
55 contains Surya, but doesn't contain some of the most-recent routeadm(1m)
changes, IIRC. S10u3 (latest one out there, aka. 11/06) won't contain Surya
or the recent routeadm(1m) changes.
Assuming you don't have confidential information, one thing helpful you can
do is when it wedges utter this command:
reboot -d
and share the {unix,vmcore}.* files. Maybe something in the coredump can
show what's getting jammed up.
Dan
_______________________________________________
networking-discuss mailing list
[email protected]
