[Linux-cluster] IP adress missing from interface (sometimes)

[Miroslav Zubcic]

Lon Hohberger <lhh at redhat.com> writes:

> There are three problems which could explain this...

> - an odd problem in bond routing in U4 + U5 kernels.
> - a problem with the e1000 driver with multiple NICs in bonded mode
> which causes "ifconfig" to fail spuriously.

Yes. I have setup another script last night to check output of
ifconfig every 1 second, it fails 7-8 times a day - 4 times last night
until this morning (central european time here).

> - a bug with clumanager versions < 1.2.26.1 re: bonded IPs.

I have: # rpm -qa | grep clum
clumanager-1.2.26.1-1

There is no new version on RHN, just checked with up2date(8).

> I think it's one of the first two, or a combination of both.

I agree. That was my thinking also. We have escalated problem with
e1000 driver to HP (e1000 is part of their official support pack for
RHEL 3 on DL 380).

e1000 driver version is e1000-5.7.6a-1. From
ftp://ftp.compaq.com/pub/products/servers/supportsoftware/linux

> The new (1.2.26.1) version of clumanager is available on RHN.
> Unsupported test kernels which should fix the first two issues are
> available here:
>
> http://people.redhat.com/linville/kernels/rhel3/

I'm afraid that I cannot put test kernels on that machines. They are
in production and my head can get on the chuck if I do this. :-)

> You can also try downgrading the initscripts package to the GA version

Which version of initscripts? I have initscripts-7.31.22.EL-2

> to work around the regression in bond routing.  If neither solve your
> problem, please file a ticket with Red Hat Support:

> http://www.redhat.com/apps/support/

OK. Thank you.


-- 
Miroslav Zubcic, Nimium d.o.o., email: <mvz at nimium.hr>
Tel: +385 01 4852 639, Fax: +385 01 4852 640, Mobile: +385 098 942 8672
Mrazoviceva 12, 10000 Zagreb, Hrvatska