[DRBD-user] Bringing up a failed master

Luc de Louw luc at delouw.ch
Mon Nov 29 12:20:07 CET 2004


Quoting Andreas Huck <drbd at huck.it>:

> Hi,
>
> On Monday 29 November 2004 10:58, Gareth Ansell wrote:
>> Hi,
>>
>> I am about to run a contingency test for some of our servers.  This
>> involves shutting down the main server (drbd master), starting the
>> standby server (drbd secondary -> drbd master), and restarting the main
>> server as the secondary server.
>>
>> Is there any special procedures to follow when I restart the (ex)main
>> server.  Will it come up as a drbd secondary server automatically?
>>
>> What I am concerned about is that when it is restarted it will bring
>> itself up into its previous drbd primary state, and cause all sorts of
>> data loss.
>
> your main server (drbd master) should become slave before
> shutting down. Your cluster manager (heartbeat?) could handle
> this. It will come up as secondary again and the primary should sync the
> changed blocks. But beware of any (misconfigured) autofailback feature
> of your cluster manager.

I successfully made such tests multiple times without any problems.

/etc/ha.d/haresources:
[..]
drbddisk::r0 Filesystem::/dev/drbd0::/data::ext3
[..]

/etc/ha.d/ha.cf
[..]
nice_failback on
[..]

/etc/drbd.conf
[..]
   startup {
     degr-wfc-timeout 120;    # 2 minutes.
     wfc-timeout 120;    # 2 minutes.
   }
[..]

For tests I just made a power down, and later power on the primary node.

rgds

Luc







More information about the drbd-user mailing list