Re: more bgpd weirdness

[Claudio Jeker <cjeker@diehard.n-r-g.com>]

On Fri, Aug 13, 2004 at 01:59:46PM +0200, Arvid Grøtting wrote:
> Henning Brauer <lists-openbsd@bsws.de> writes:
> 
> >> Bad build, huh?  Well, it was from a snapshot.  Perhaps I should go to
> >> -CURRENT from source and see what happens.
> >
> > yes, try that please.
> 
> OK, I'll see if I can try that, but I can't promise anything.
> 

An other test you could do is swap machines (if possible). As I already
said, imsg errors are not common. imsg header errors only happen when the
data on the pipe gets mixed up. I see three possibilities for this:
a) bad compile
b) bad machine e.g. bad RAM
c) a strange bug that can only be triggered in some uncommon moment

> > a high load average is nothing to worry about, however it seems you got 
> > very high CPU use too, which does not match at all with our tests and 
> > our priduction uses. so something is strange with your box/setup/... 
> > and we have to find out what - and as we have seen quite some bad 
> > builds....
> >
> >> I have the same snapshot running on another router with no incident
> >> (apart from the nexthop bug, but that's not a problem as long as I can
> >> keep my internal routing stable.  Which I can.)
> >
> > what is different there?
> 
> One difference that I can think of: The network interface cards.
> 
> The box that works nicely has a D-Link STE-570TX (which uses the dc
> driver), while the one that doesn't has a D-Link STE-580TX (using the
> ste driver, which works fine without my crude private multicast
> reception patch now, by the way).
> 
> The 580 is an ugly beast, I know.  One of the developers wouldn't
> happen to have one, by any chance?  ;-)
> 

Should not matter unless other applications behave strangly to.

> But there's another difference that is perhaps more important: The box
> that works has 3 peers (2 ibgp, 1 ebgp), while the one that doesn't
> has more like 35 peers (2 ibgp, 33 ebgp).  For an error involving the
> SE, this could be significant.
> 

For me it is currently not possible to have so many peers. I'm planning to
get such a test environment up and running somewhen next week.
Normaly 35 sessions should not hurt the SE as all complex tasks are done
in the RDE.

-- 
:wq Claudio