[Linux-HA] Re: failback (Andrew Beekhof)

[Andrew Beekhof]

On 10/4/07, Sripathi, Roopa (Roopa) <rsripathi at alcatel-lucent.com> wrote:
> Hi,
>
> Attached is the input.xml generated from running command :
>
> ptest -L -VVVVVV  --save-input input.xml
>
> The only way failback is happening is by running :
> crm_resource -C  -r IPaddr_cluster -H roopa1
> crm_resource -C  -r RES_X  -H roopa1
>
>
> I need an autofailback to happen once the resource RES_X is fixed on the
> first node(Node A) and the resource RES_X fails on Node B, where it had
> failed over to.
>
> That is, I need a failback to happen without having to run the
> crm_resource -C command.
>
> Is it possible ?
>
> I changed the default resource stickiness to 0, still the same
> behaviour.
>  I am attaching the logs & cib.xml, zipped

the problem is that your resource is moving for the wrong reason.

it is not moving in response to the monitor failure (because
default-resource-failure-stickiness is set to 0) but in response to
the restart failing.

ptest[9794]: 2007/10/05_10:57:16 WARN: unpack_rsc_op: Processing
failed op (qip-named_2_start_0) on roopa2
ptest[9794]: 2007/10/05_10:57:16 WARN: unpack_rsc_op: Processing
failed op (qip-named_2_start_0) on roopa1

start failures are permanent and prevent the resource running there
again until -C is used.

the solution here is to set default-resource-failure-stickiness to a
negative value and read http://linux-ha.org/v2/faq/forced_failover