[Linux-HA] How to stop monitoring operations on specific hosts

[Andrew Beekhof]

On 12/11/06, Alex Strachan <astrachan at inter-systems.com.au> wrote:
>
>
> > -----Original Message-----
> > From: linux-ha-bounces at lists.linux-ha.org [mailto:linux-ha-
> > bounces at lists.linux-ha.org] On Behalf Of Andrew Beekhof
> > Sent: Monday, 11 December 2006 7:47 PM
> > To: General Linux-HA mailing list
> > Subject: Re: Re: [Linux-HA] How to stop monitoring operations on specific
> > hosts
> >
> > On 12/9/06, Alexander Strachan <astrachan at inter-systems.com.au> wrote:
> > > Quoting Andrew Beekhof <beekhof at gmail.com>:
> > >
> > > > anything with monitor_0 is us checking the resource is active on a
> > > > newly started node.
> > > > this is not optional.
> > >
> > > Hopefully I can pass a parameter to the monitor operation to help
> > distinguish
> > > which monitor is being run; either
> > >   op id="drbddisk_itsapaims1_drbd_mon"
> > > or
> > >   op id="drbddisk_itsapaims2_drbd_mon"
> >
> > why do you need this?
>
> It is a cluster of 4 nodes.
>   2 connected via drbd, then oracle, ...
>   2 connected via drbd, then oracle, ...  (meant to be identical)
>                                           (the same filesystem, Oracle, and
> apps)
>   Resources which can run on any of the four nodes, and resource which only
> run on one of the primary drbd nodes)
>
> I can see that on startup the monitor_0 checks are being done. This will be
> true even if the drbddisk check is for the other nodes in there own drbddisk
> setup; I can't see how the script knows which drbddisk it has just checked.

i dont know anything about that script but presumably it uses one of
the parameters that are passed in.

try putting a call to "env" in the script to see what variables are set for you

> I thought of changing the names of the RA scripts

the names of the scripts (the RAs) or the resources?  i think you mean
the later but wrote the former.

> to have
> itapaims1_drbddisk, itapaims1_oracle allowing me to manage the monitoring;

the id of the resource is passed in, just not the id of the operation.

> from other post I need to install a later CRM-Stable to get advantage of
> some work you have just completed on asymmetric clusters.

[snip]

> > > e.g. we have ORacle running on two of the nodes (4 node cluster), they
> > have
> > > identical SID's so the monitor for either resource will return true.
> > Hmmmm
> >
> > perhaps using a clone for this resource is a good idea
>
> I didn't think clones were supported in a group definition.

you dont need groups
no-one needs groups

they are a syntactic shortcut and don't achieve anything that cant be
done with some colocation and ordering constraints

>  The two
> Oracle's although they are defined the same, they contain different data;

[apart from the fact that this seems like a recipe for disaster to me]

this is therefor how you tell them apart - write a status operation
that looks at some subset of the data to figure out which one its
talking to

> not sure about clones (need to read more about them).  There is also custom
> application, apache and samba which would need to be cloned.