Re: IBM RDAC (Multipath SAN failover) driver released (GPLed) Was: SAN w/ multiple paths on IBM bladecenter

[D Canfield <canfield uindy edu>]

Also, it would appear to me that problem 4) is that there is no upgrade 
path for users who installed multipath using the qla2300 support in 
RHEL3.  I assume that upgrading these boxes to RHEL4 will require 
re-partitioning all of those disks to md or lvm partitions.  Given that 
I get about one email a week from someone trying to setup mutli-path 
booting on a bladecenter, I think this is going to be a bigger issue 
than the Red Hat kernel team believes.

I agree that the "new" solution sounds like a great solution in theory, 
but my reading suggests the same issues as yours.  The killer being no 
way to multi-path /boot, and therefore RHEL4 may be less 
enterprise-ready than RHEL3.

I'm going to try the beta of RHEL4 on a blade this week and see if it's 
even remotely functional on a multi-pathed SAN, but I suspect the 
complete silence we always get on the issue of multipathing /boot hints 
at the answer to the question of whether that will work...  :-)

DC

shane stixrud org wrote:

> On Sat, 11 Sep 2004 shane stixrud org wrote:
>
>  
>
> > On Sat, 11 Sep 2004, Arjan van de Ven wrote:
> >
> >    
> >
> > > On Sat, Sep 11, 2004 at 11:42:42AM -0700, shane stixrud org wrote:
> > >
> > >      
> > >
> > > > It seems that with mdadm I must create software raid partitions and 
> > > > then configure mdadm for multiple paths, which sounds like it would be 
> > > > functional for partitions other than /boot. 
> > > >        
> > > actually that's not the case; you don't need to make special software raid
> > > partitions
> > >      
> > I will have to re-look at mdadm in that case, I take it booting from a 
> > multipathed partition would work as well?  The problems we ran into had to 
> > do wtih anaconda not knowing what to do with multiple paths and the OS in 
> > generally seeing a mirror sda in the form of sdb which caused mount 
> > failures, I presume because of LABEL=/ conflicts.
> >    
>
> > From the little documentation I have found on mdadm it seems I still have 
> couple issues to resolve with a mdadm approach (I will try and confirm this 
> Monday).
>
> 1) I have not found any documentation that suggests Anaconda/kickstart is 
> mdadm/multipath aware and have first hand experience that anaconda will not 
> leave the system in a bootable state if two paths to the same LUN are 
> available during installation.
>
> 2) If I am forced to unplug/disable one of the paths during 
> installation I don't see how mdadm will be able to work its magic during 
> kickstart %post.
>
> 3) If my thinking above is correct I will need to disable the multiple 
> paths, install as though it were a single path server.  Reboot then 
> solve the device LABEL=/ problem I see during filesystem mount 
> time when RHEL3 sees two multiple paths.  Reboot again (or mount -a), 
> configure mdadm manually and perhaps build new ramdisks before rebooting 
> again.
>
> I assume grub will remain pointing at /dev/sda and I hope still boot if 
> the primary path is down.  If my reasoning is accurate, it seems 
> more labor intensive then providing a rdac driver disk to 
> anaconda/kickstart during initaltion and installing our rdac rpm in 
> kickstart %post.
>
> Cheers,
> Shane
>
>

Attachment: smime.p7s
Description: S/MIME Cryptographic Signature