Mailing List Archive: drbd +ha + expanding capacity.

Hey,

Let's imagine a situation like this : I have 2 machines configured as HA
cluster, each with one disk synced with drbd. Now if I want to add one disk
to each machine, how does it work ?

1) Switch off one machine and add a disk
2) Resync the "old" disk with the machine still active (full sync I suppose)
3) Switch off the second machine and add the second disk in it
4) Resync the second (full sync again ?)
5) Do nasty tricks to agregate the 2 disks on both machines as one space.

Is this technicaly possible without stopping services ? I'm particulary
wondering about the time (and load) it's going to take to resync the
machines twice...

Did anybody tried this already ?

Nico

--
Administrateur Système/Réseau - GHS 38, rue du Texel 75014 Paris
Tél : 01 43 21 16 66 - nbouthors@example.com - nicolas@bouthors.org

* Nicolas Bouthors <nbouthors@example.com> [010926 16:10]:
> Hey,
>
> Let's imagine a situation like this : I have 2 machines configured as HA
> cluster, each with one disk synced with drbd. Now if I want to add one disk
> to each machine, how does it work ?
>
> 1) Switch off one machine and add a disk
I suppose you will do a gracefull shutdown.
> 2) Resync the "old" disk with the machine still active (full sync I suppose)
Since you did a gracefull shutdown, a quick sync is sufficent.
> 3) Switch off the second machine and add the second disk in it
> 4) Resync the second (full sync again ?)
Quick.
> 5) Do nasty tricks to agregate the 2 disks on both machines as one space.
I would add the new disk as new drbd device.

In case you rearrange the disk to form RAID0 units, you have to do a
full sync off course. (Since the data needs to be restored)

>
> Is this technicaly possible without stopping services ? I'm particulary
> wondering about the time (and load) it's going to take to resync the
> machines twice...

The only purpose of the syncQuick code is to bring the down the
cost (time and load) of maintainance work.

PS:

gracefull shutdown (p|s) ==> syncQuick
hard crash of primary ==> syncAll
hard crash of secondary ==> syncQuick

PS2: The trick with the gracefull shutdown of a node in primary state is,
that it will be switched into secondary state at module unloading
time. -- There is nothing wrong with this!

-Philipp