[j-nsp] JUNOS

[Richard A Steenbergen]

On Fri, Jan 08, 2010 at 03:13:38AM -0600, Richard A Steenbergen wrote:
> In our experience 9.5R3 has been "mostly ok", i.e. nothing is really
> really broken. The closest thing to bad is a one-time mystery rpd crash
> which seems to duplicate a backtrace we saw in 9.4R3, everything else is 
> in the annoyance category (for example if you hit ctrl-c in the cli 
> anywhere other than at the prompt it locks up the cli process :P). I've 
> certainly seen worse. :)

Ok I should have known better than to jinx it like that... just
discovered a pretty nasty bug in 9.5R3 where when an ae interface (I
THINK one member of the interface is enough to do it, but still
investigating) flaps the rpd coredumps. It looks like the interface 
routes attached to the ae are somehow "going away", and rpd crashes 
after that. After rpd comes back up the interface routes on the ae are 
still missing, the remote side is unreachable and if you try to bounce 
the subints that are affected you get:

%DAEMON-3: KRT ADD for x.x.x.x/32 => { ifl 81 addr x.x.x.x } 
failed, error "ENOENT -- Item not found".

Only bouncing the entire ae (deactivate, commit, reactivate) brings the
routes back. Will keep you guys updated as I find out more, but I'm now
up to 5 routers that have crashed exactly this way with identical rpd
backtraces and symptoms, including two that just blew simultaniously 
when a connected router flapped to them. This happened under 9.4R3 and 
9.5R3, keep an eye out for it. :)

-- 
Richard A Steenbergen <ras at e-gerbil.net>       http://www.e-gerbil.net/ras
GPG Key ID: 0xF8B12CBC (7535 7F59 8204 ED1F CC1C 53AF 4C41 5ECA F8B1 2CBC)