Re: [SSI-users] Issues of Booting the Second Potential Root Node First with Separated /boot Partitions

[Brian J. W. <Bri...@hp...>]

Denny Chou [周岱民] wrote:
> It's OK to have the first (original) node boot up and the 2nd node join 
> the cluster. Root failover also works fine.

Great!

> It should be reasonable when 
> first node totally damaged or need to be maintained then the second node 
> maybe need to boot up alone.

Certainly.

> 1. The second node boot up OK but without mounting the local /boot 
> partition.

That's not good. There should always be a /boot mounted, so that you are
able to edit your grub.conf, upgrade your kernel, rebuild your ramdisk
and run ssi-ksync.

There's no trivial solution I can think of for this, so it won't be
fixed right away (at least not in the stable branch). Can you enter a
bug report on our SourceForge page, so that we don't forget about it?

> 2. Boot up the first node to join the cluster will hang when mounting 
> local file system and there are panic messages in the (already booted) 
> second node.

What does the panic message say?

> For the issue 2, I'd fixed it by modifying the /dev/pts entry in the 
> /etc/fstab to
> 
> none                    /dev/pts        devpts  gid=5,mode=620,node=*   
> 0       0
> 
> I don't know whether this is a right way to fix the problem or not but 
> it does work.

John Byrne, who implemented the clusterwide devpts, said this is a good
solution. I've modified the installation scripts to do this by default.

> I'd tried to label the second node /boot partition just like in 
> the first node and modify the fstab to
> 
> LABEL=/boot     /boot                   ext3    chard,defaults,node=1:2 
> 1       2
> 
> This fixed the problem of not mounting /boot when root failover to the 
> second node. After failover, the local /boot partition of second node 
> mounted on /boot directory. 

The chard mounts were not designed for doing something like this, but
perhaps it's a reasonable work-around.

> But after I boot up the second node first 
> and then have the first node boot up to join the cluster, the 'df -k' 
> shows a strange result below.
> 
> Filesystem           1K-blocks      Used Available Use% Mounted on
> /dev/2/sdb1            4127204   1540532   2377020  40% /
> /dev/2/sda1             506833     28909    451756   7% /boot
> /dev/1/sda1             506833     28909    451756   7% /boot

This could be because of bad informaton in /etc/mtab. What does `onall
cat /proc/mounts' tell you?

Thanks for the feedback,

Brian