Dear list,
We are in the process of setting up a TruCluster 5.1 consisting of a
DS20E (2x667) --nice machine-- and an older 4100 (2x466).
Both of them have a GBit netwark interface (DEPGA).
We set up a few services (NFS,NIS,named) and everything seemed to work
as expected at fisrt. After testing cluster failover with hard and
soft reboots -- everything worked as expected --, our next test
was to unplug one members network connection.
The member whose network connection was unplugged, was running named.
An external client was configured with the clusters default alias IP
address as the primary name server.
pinging the cluster default alias IP address from that client while
unplugging the one member showed that after a short timeout the other
member had taken over as the proxy arp server. An arp -a before and
after the unplugging showed that the MAC address of the default alias
had changed.
This all seems pretty normal and the way we expected things to work.
BUT: ping was essentially the only thing that worked. Especcailly we
could no longer contatct the name server from the client. Eqiuvalent
test with NFS mounts from the cluster on the client showed the same
behaviour. NoGo after unplugging one member.
Interestingly, on the cluster the /etc/gated.conf.member2 got
rewritten by aliasd as soon as the lost connection was discovered
(member2 was the one that was unplugged). A netstat -rn on member2
showed that after the unplugging it had lost its default route and its
route to the cluster alias IP through 127.0.0.1.
As I understood it from the docs, in the case of a network
connectivity loss routing should go through the big expensive black
cable (aka MChan).
Whats wrong here?
A related question: As is stated in some man page a NetRAIN cannot be
configured from 1 GBit and a FastEthernet interface. Is this correct?
Hmm, bummer, it would be nice if we could use the FastEthernet
adapters as a fall back. What else could we do with them? Tru64Unix
supports two adapters in the same subnet. Does this work also in a
cluster? Can we assign two more adresses to the ee0 and tu0 on the
two members and they would be part of the default alias?
Thanks in advance,
Wolfram Klaus
--
Wolfram Klaus (Wolfram.Klaus_at_physik.fu-berlin.de)
Free University Berlin
Physics Department
Received on Tue Apr 03 2001 - 13:00:12 NZST