[Pacemaker] drbd diskless -> failover to other node

[jimbob palmer]

2010/8/27 Florian Haas <florian.haas at linbit.com>:
> On 2010-08-27 10:31, jimbob palmer wrote:
>> 2010/8/27, Florian Haas <florian.haas at linbit.com>:
>>> On 2010-08-26 16:43, jimbob palmer wrote:
>>>> How can I configure pacemaker to failover when the primary node goes
>>> diskless?
>>>>
>>>> Many thanks.
>>>
>>> man drbd.conf
>>>
>>> Look for the local-io-error handler and the on-io-error option.
>>>
>>> I doubt that it's a good idea to do this though; you're deliberately
>>> foregoing a pretty handy DRBD feature there.
>>
>> What do you mean? If the primary server loses it's disk, I'd like to
>> switch to the server with a working disk.
>
> Which means you're causing a service interruption when you don't need
> to. Instead, your application could continue running on the same node,
> DRBD will ensure that the application transparently writes to and reads
> from the peer when it thinks it's writing locally, and you can coolly
> hot-swap the drive out from under DRBD, and resync.
>
> Usually service interruption is worse that degraded performance. Would
> you rather fail over automatically, perhaps during peak hours and under
> full load (which is what you are headed for), or would you rather have
> your app continue to run where it is, to then switch over at a time of
> your choosing?

Are you saying that if a server loses its disk, it will transparently
write to the secondary server without any need to failover at all?

WOW. I never knew DRBD did this. This is a _fantastic_ feature :)