Re: [OSR-devel] IPs in cluster.conf and manual startup

[<go...@bo...>]

On Wed, 7 Nov 2007, Marc Grimme wrote:

>>>> I'm having a problem with the way the ifcfg-eth* files are being handled
>>>> for the initrd. My ifcfg-eth1 file doesn't get transferred across
>>>> verbatim. This is a problem because I have to explicitly specify the
>>>> HWADDR in my ifcfg files (therwise they come up on wrong subnets).
>>>
>>> The initrd creates new ifcfg-ethX files for the defined ethernet
>>> interface for its internal use. I.e. for all <eth /> entries for the node
>>> in the cluster.conf. These file are only valid inside the initrd.
>>
>> Sure, but the iSCSI interface has to be the same in root and initrd,
>> otherwise it won't work.
>>
>>>> It looks as if mkinitrd removes some lines from the file, and this
>>>> causes eth1 to be physical eth0, which puts it on the wrong subnet, and
>>>> that means it can't see the iSCSI shares.  :-(
>>>>
>>>> I checked the /etc/sysconfig/network-scripts/ifcfg-eth1 in the initrd,
>>>> and it is indeed lacking the HWADDR line (and a few others).
>>>>
>>>> Am I missing something here? Does mkinitrd mangle the ifcfg-eth* files?
>>>
>>> The mkinitd does not modify the ifcfg-ethX files.
>>> If you want to define the HWADDR parameter in the ifcfg-ethX files, you
>>> have to make them hostdependent.
>>
>> I did, but network-lib.sh checks if it's in the initrd, and if so saves
>> a backup and creates a new one from the parameters that come from
>> elsewhere (I'm assuimg from cluster.conf).
>>
>> The new one doesn't have the HWADDR parameter in it, because it isn't one
>> of the things passed to network-lib.sh. So, any interface that are bound
>> by MAC address rather than the default ordering won't work properly in the
>> initrd. I have interface name, ip and mac specified in cluster.conf, but
>> it never gets included in the built ifcfg file. :-(
>
> That's a known "Bug" I dare say ;-) . But it should not be a big thing to
> change, should it?!
> Try the rpm attached.
> What do you think?

I'll try it later, when my cluster stabilises a bit. I spent all of 
yesterday getting a 3-node cluster to boot, and it kept failing when 
starting the cman service. Eventually, it booted up fine, and I still 
don't know what I changed in the initrd that would make any difference at 
all - so I'm a bit reluctant to try changing things again. For now I think 
I'll have to live with interface inconsistency on a few machines. Maybe in 
a few days. :-)

On a separate note - is there an option to include a delay after the 
interface comes up? I'm using managed switches and they are quite crap - 
they can take up to 30-60 seconds to  figure out that something has been 
plugged in, so when the interface comes up, the SAN isn't yet accessible 
by the time iSCSI loads. For now I've bodget it by adding a sleep 30 in 
iscsi-lib.sh, but it'd be nicer if it just used the standard ifcfg options 
for this, as it's more generic.

Gordan