[DRBD-user] How to customize the size of a DRBD partition ?

[Fabrice Durand]

On 9/6/05, Martin Bene <martin.bene at icomedias.com> wrote:
- Show quoted text -
> Hi Fabrice,
>
> > > If you're using metadata internal you _MUST_ create the
> > > filesystem on the drbd device, not on the
> > > /dev/hdxxx device.
>
> > We have created the filesystem with the command
> > mkfs -t ext3 /dev/hdaxx
>
> Please carefully reread the two previous statements and see if they tell
> you something :-)
>
> * Backup any data you don't want to lose that's currently stored on
> drbd.
>
> * Recreate the filesystem using
> mkfs -t ext3 /dev/drbd0
>
> You'll now have a filesystem with aprox. 128MB free space and you won't
> have any more errors "access beyond end of device".
>
> > disk {
> >   on-io-error pass_on;
>
> I don't know your application but for all setups I did pass_on would be
> the worst choice. You either want to fail over to the other node (panic
> on I/O error, let the cluster manager handle the failover) or you want
> to continue running on the current node (use detach, get good data from
> the other node).
>
> Bye, Martin

> You like to live dangerous?
>
> Just do it the simple way like
>
> - prepare space for low level device (e.g. partition or lv)
> - edit /etc/drbd.conf
> /etc/init.d/drbd start
> drbdsetup /dev/drbd1 primary --do-what-I-say
> mke2fs -j /dev/drbd1
> mkdir /mnt/vs03
> /etc/ha.d/resource.d/Filesystem /dev/drbd1 /mnt/vs03 ext3 start
>
> Helmut Wollmersdorfer

Thanks for replying. In fact I omitted to write the previous statement and
I did create the filesystem using
> mkfs -t ext3 /dev/drbd0
Anyway we recreated using the method described by Helmut, and now it's better.
But now, the primary node appears to be Inconsistant. What does this
mean and how to make it Consistent ?

As for the parameter on-io-error :
> > disk {
> >   on-io-error pass_on;
In my application (2 node Heartbeat cluster), I want to fail over to
the other node:
so according to you I shoud use 'panic' option, but I have read this
make a kernel panic. Isn't there a more convenient option, preventing
a kernel panic ?
Thanks,
Fabrice