[Pacemaker] OCF RA dev guide: final heads up

Andrew Beekhof andrew at beekhof.net
Wed Dec 8 09:24:14 UTC 2010


On Mon, Dec 6, 2010 at 4:25 PM, Florian Haas <florian.haas at linbit.com> wrote:
> Hello Bob,
>
> On 2010-12-03 20:12, Bob Schatz wrote:
>> Florian,
>>
>> Thanks for writing this!
>>
>> I already found one or two errors related to return codes in my agent based on
>> your document. :)
>>
>> I have not read the entire document but I do have these comments:
>>
>> 1.Does this document apply to all versions of the agent framework or only
>> certain versions(hopefully all in one place)?  I think the document should have
>> a section which specifies which versions are covered.  Also, if certain areas
>> only apply to a certain version then a "Note" should be mentioned in the
>> section.
>>
>> 2.    In Section 3.8  OCF_NOT_RUNNING, how can a monitor return
>> OCF_FAILED_MASTER?   Is there an environment variable passed to the monitor
>> action which says "I think you are a master - tell me if you or are not"?
>
> No, the very purpose of monitor is to _find out_ the status of the
> resource. If the resource can query its own master/slave status, it
> should do so, and then if it is both a master and failed, it should
> return OCF_FAILED_MASTER.
>
>> 3.    In Section 5.3 monitor action, it would be nice if you show how a
>> OCF_FAILED_MASTER is returned.
>
> Hm. Let me defer that for a little bit.
>
>> 4.    Sections 5.8 migrate_to action and 5.9 migrate_from action, do these apply
>> to master/slave resources also or only to primitive resources?
>
> Good question, and indeed I don't know. It's conceivable that a clone
> set (remember, m/s are just clones with a little extra) has a clone-max
> that is less than the number of nodes in the cluster, and supports
> migration, and therefore a clone instance should be able to live-migrate
> to a different node. I have no clue whether it's indeed implemented that
> way, though.
>
> Andrew, maybe you can shed some extra light on this?

I don't believe there is anything in the PE that prevents it and I
don't have any plans to add such a limitation - being able to migrate
clones seems like a reasonable expectation.
Having said that, I don't know how well its been tested.

>
>> 5.    Section 5.10 notify action, I think you to want to add a note/reference to
>> the Pacemaker Configuration Explained section "10.3.3.9 Proper Interpretation of
>> Notification Environment Variables".  (Section name may be different as I was
>> looking at 1.0 from about a year ago).
>
> Good idea. I'll put that on my to-do list.
>
>> 6.    Section 8.4 Specifying a master preference, starting in at least version
>> of Pacemaker 1.0.9.1 it is possible to specify a negative master score.  I think
>> it would be good to add this to the example as well as a note about which
>> version has this functionality since it was broken in 1.0.6.
>
> Don't you think this would just royally confuse people?
>
> Florian
>
>
> _______________________________________________
> Pacemaker mailing list: Pacemaker at oss.clusterlabs.org
> http://oss.clusterlabs.org/mailman/listinfo/pacemaker
>
> Project Home: http://www.clusterlabs.org
> Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
> Bugs: http://developerbugs.linux-foundation.org/enter_bug.cgi?product=Pacemaker
>
>




More information about the Pacemaker mailing list