[Linux-cluster] service startup order

[Digimer]

On 07/28/2011 02:24 AM, Ralph.Grothe at itdz-berlin.de wrote:
> Hi Digimer, hi Lazlo,
> 
> sorry, for intruding your thread but this is something that I am
> also interested in and which I haven't fully fathomed yet.
> 
>> -----Original Message-----
>> From: linux-cluster-bounces at redhat.com 
>> [mailto:linux-cluster-bounces at redhat.com] On Behalf Of Digimer
>> Sent: Thursday, July 28, 2011 2:51 AM
>> To: linux clustering
>> Subject: Re: [Linux-cluster] service startup order
>>
>> Parallel services will be started simultaneously. Services 
>> configured as
>> service trees will start in the order that they are defined 
>> (and stopped
>> in reverse order).
>>
>> This covers the start order well:
>> - https://fedorahosted.org/cluster/wiki/ResourceTrees
>>
> 
> The referred to wiki article only seems to treat
> starting/stopping order and hierarchy (parent-child vs. sibling)
> of resources within one service aka resource group.
> That sounds pretty clear.
> But what about ordering and possible dependencies between
> separate services?
> 
> You mentioned service trees. You didn't actually mean resource
> trees?
> If however your wording was deliberate (what I assume) I wonder
> if one can nest service tag blocks as one can nest resource tags
> within a single service block to express dependencies or
> hierarchy and hence starting order between and of services?
> Because all the sample configuration snippets I have seen so far
> in various docs lack such nesting of services.
> 
> The reason that interests me is because I have such a case where
> a customer requires such a dependency between two distinct
> services that during normal operation (i.e. no node has left the
> cluster) are hosted on different nodes.
> I told them, from what I have perceived of HA clustering and RHCS
> in particular so far, that if they wish to express such an
> interdependency that they would have to put all resources which
> are now split up in two services, in a nested manner that would
> map the intended hierarchy, in a single service.
> 
> Because they insisted on their layout I searched a little and
> discovered the, in the official RHCS Admin doc not mentioned,
> service tag attributes "depend" and "depend_mode".
> 
> However, their usage at first seemed pretty useless because the
> clusterware seemed to completely ignore them and start/stop
> services in sometimes unpredictable ways and even restarted them
> at random.
> Until I, more by accident, discovered that additionally the "rm"
> tag's attribute "central_processing" needed to be defined and
> assigned to "1" or "true" for this feature to work approximately.
> I say apprimately here because we still have issues with this
> cluster that require futher testing.
> I hardly dare mentioning, that unfortunately this system already
> went into production, now of course lacking any HA,
> why we had to defer further testing.
> 
> 
> Regards
> Ralph

I am late in returning to the thread, my apologies. :)

I did not choose my words carefully, and I did mean resources, not
services. When I used services, I was thinking about ordered daemon
starting (that is, a resource group of system services/init.d scripts).

As far as I understand, and as has been mentioned already further down
this thread, rgmanager is limited in it's ability for complex dependency
checking. That is why Pacemaker is so attractive and why it will
eventually replace rgmanager as the primary cluster manager eventually
(EL7?).

-- 
Digimer
E-Mail:              digimer at alteeve.com
Freenode handle:     digimer
Papers and Projects: http://alteeve.com
Node Assassin:       http://nodeassassin.org
"At what point did we forget that the Space Shuttle was, essentially,
a program that strapped human beings to an explosion and tried to stab
through the sky with fire and math?"