Very Computer

I have two 44P-270's in a rotating HA 4.3.1 classic cluster, with AIX
4.3.3.0.05 and Firewall-1 4.0.7.

I monitor the status of three NICs and fail to the other node when one
fails.  This works great for one adapter.  I can pull the NIC's cable
from the hub and have the node fail fine every time.

But with the other two adapters, when I pull either cable, the same
node always goes down, regardless of whether or not it currently holds
the service addresses.

The systems are running Firewall-1 and maintaining on the order of
60,000 active TCP connections, although they handle this extreme load
just fine, which is spread roughly evenly across the three adapters I
monitor.

I cannot figure why one adapter works perfectly, while the other two
always fail the same node.  Any help is much appreciated.

Thanks,
RR

HACMP does not automatically fallover to another node due to an adapter
failure.  What method did you use to customize this behavior?

Of course, I think the more important question is: Why not let HACMP move the IP
address of the failed adapter to a standby adapter, rather than having the whole
machine fallover?

Gus

--------------------------------------------------

--------------------------------------------------

I attached a post-event to the "network_down" event, where I  shutdown
the node name given me by HA.  I took this hint from the Redbook on
running Firewall-1 on AIX.  The script simply compares the hostname of
the system to the node name passed it by HA.  If they match, come down
now.  If not, continue on as usual.  But even so, only one adapter
fails the correct node; the other two adapters always fail the same
node regardless.

Quote:> Of course, I think the more important question is: Why not let HACMP move the IP
> address of the failed adapter to a standby adapter, rather than having the whole
> machine fallover?

There are not enough slots in the computer for six total Ethernet
adapters.  (I use the built-in adapter as the Firewall-1 management
network and is not under control of HA).

I also realized something else just now, as I was typing.  The one
adapter that does fail properly, is plugged into a 64-bit slot (but it
still the same model card as the other two).  The other two adapters
are both in 32-bit slots.  Not sure if this has any bearing here or
not, but it is certainly a curious fact...

Thanks much,
RR

Hi,
Just a hint, have you add the entries to /usr/sbin/cluster/netmon.cf??
regards,
Victor

I would help tremendously if we could see the script.  Most likely there is an
error in it's logic, but you can blame that on the Redbook if you like.  :-)

Gus

The script is nothing more than a simple comparison of system hostname
to parameter $3.  If they match, come down.  If not, just exit.  There
is no logic having to do with whichever Ethernet adapter is failing
and so this would not explain why one adapter works but the other two
don't.

I believe the problem has to do with my /usr/sbin/cluster/netmon.cf
file, as another poster suggested.  I will be testing this at my next
maintenance window.

Thank you for the help!

RR