[Linux-HA] Dependencies between resources

[Andrew Beekhof]

On 8/4/06, Thurston Nabe <thurston at angelpoints.com> wrote:
> Hello All,
>
> I'm trying to solve the following configuration scenario:
> using crm (Heartbeat 2.0.5)
> Resource A (attempt to always keep running on a single node, has a
> monitor action)
> Resource B (attempt to always keep running on a single node, has a
> monitor action)
>
> IP address (should migrate when applicable)
>
> "applicable" is:
> 1.  IP address monitor fails
> 2.  And this is the tricky part, if IP address is "running" on the same
> node that Resource A is running on, AND Resource A's monitor action
> fails
> 3.  And this is the tricky part, if IP address is "running" on the same
> node that Resource B is running on, AND Resource B's monitor action
> fails
>
> Now this is very much like "group" behavior, but the anomaly is I don't
> want the Resource (A or B) to "migrate", just the IP address
> Essentially, I'd like to define a constraint like this:
> <rsc_location rsc="IP-address">
>    <rule score="-INFINITY" boolean_op="and">
>        <expression attribute="#uname"  operation="eq" value="Resource
> A's machine"/>
>        <expression "evaluate somehow that Resource A's monitor action
> failed"/>
>    </rule>
>    <rule score="-INFINITY" boolean_op="and">
>        <expression attribute="#uname"  operation="eq" value="Resource
> B's machine"/>
>        <expression "evaluate somehow that Resource B's monitor action
> failed"/>
>    </rule>
> </rsc_location>
> My English interpretation of the preceding rules is:
> never run IP-address on machine A (the uname expression) if Resource A
> "just failed"
> and same for the second rule (transposing B for A)
> Assuming that would work, I don't know how to express <expression
> "evaluate somehow that Resource X's monitor action failed"/>
>
> in a way, as I mentioned before, this is what defining a group does,
> i.e. it will attempt to migrate one resource if another resource's
> monitor action failed.  That's what I want, but I don't want all of a
> group's resources to migrate, in this case just the IP-address to
> migrate.
> I've looked at rsc_order, rsc_colocation, etc. and none of them quite
> seem to do what I want.
>
> Maybe it's the responsibility of the Resource A (B) monitor actions to
> set the value of an instance attribute once they detect failure?  How is
> that done in a shell script? Execute cibadmin, crm_attribute, changing
> the value of the OCF_RESKEY_XXX environment variable? And then
> referencing that attribute in a rule's expression?
>
> Any help would be appreciated

This probably isnt possible until after the PE rewrite happens.  No
ETA on that yet though.