[CUWiN-Dev] multiple gateways?

[Bill Comisky]

On Wed, 14 Dec 2005, David Young wrote:

> On Tue, Dec 13, 2005 at 12:00:22PM -0600, Bill Comisky wrote:
>> Upgrading the 2nd gateway did indeed fix the problem of the weird default
>> route.  We're still seeing a lot of hslsd core dumps though (running CUWiN
>> r3715 and the 8-Dec netbsd snapshot).  From the timestamps (and adjusting
>> for the watchdog timeout), they happen anywhere from immediately after
>> hslsd is restarted to 90 minutes later, with the mean at about 22 minutes.
>
> I see a lot of coredumps, too.  It has been a serious operational
> problem.
>
> The segfaults only happen in a couple of places.  I haven't been able
> to track down the root cause, but Router LSAs are implicated, always.
> The Router LSA implementation was overly complicated.  Today, I simplified
> a lot.  I have run the simplified code for more than an hour on four
> routers without a single coredump.  Let's see what happens overnight.
>
> I have attached a patch containing the simplifications, which shorten
> the code by more than 200 lines.  Give them a try?
>
>> I put the latest core files with the output from 'mkstaboot -S -S' here,
>> if they can tell you anything:
>> http://flatiron.cntwireless.org/cuwin/12-Dec-2005/

I've been running CUWiN 3371 patched with the simplify-rtrlsa patch on our 
lawndale gateways for the last 12 hours or so, with some standalone nodes 
upgraded as well and others still running older versions.  We're still 
seeing hslsd core dumps, about 10 on one of the gateways and about 5 on 
the other.  So there seem to be fewer core dumps (before the patches we 
were averaging closer to every 20+ minutes), but they're still happening. 
One of the nodes rebooted after about 12 hours (watchdog apparently) the 
other is still up.  Latest 'mkstaboot -S -S' output and core files are 
here: http://flatiron.cntwireless.org/cuwin/15-Dec-2005/

The last issue I have to report is that we've seen that occasionally a 
gateway will lose (or maybe reboot without) it's default route.  The 
gateway node will have an IP address, be running dhclient, but not have a 
route to the LAN gateway.  Restarting dhcpselect and/or hslsd did not have 
an effect (but did not try restarting dhclient manually).

Bill

--
Bill Comisky
bcomisky at pobox.com