[ClusterLabs] Need advice: deep pacemaker integration, best approach?

[Ken Gaillot]

On Sun, 2024-06-09 at 23:13 +0300, alexey at pavlyuts.ru wrote:
> Hi All,
>  
> We intend to integrate Pacemaker as failover engine into a very
> specific product. The handmade prototype works pretty well. It
> includes a couple of dozens coordinated resources to implement one
> target application instance with its full network configuration. The
> prototype was made with pcs shell, but the process is very complex
> and annoying for mass- rollout by field engineers.
>  
> Our goal is to develop a kind of configuration shell to allow a user
> to setup, monitor and manage app instance as entities, not as a set
> of cluster resources. Means, user deals with app settings and status,
> the shell translates it to resources configuration and status and
> back.
>  
> The shell be made with Python, as it is the best for us for now. The
> question for me: what is the best approach to put Pacemaker under the
> capote. I did not consider to build it over pcs as pcs output quite
> hard to render, so I have to use more machine-friendly interface to
> pacemaker for sure but the question is which ones fits our needs the
> best.

pcs, crm shell, and the Pacemaker command-line tools are the basic
options
 
> It seems like the best way is to use custom resource agents, XML
> structures and cibadmin to manage configuration and get status
> information. However, it is not clean: should cibadmin be used
> exclusively, or there also other API to pacemaker config pull/push?

If you're using the command-line tools, yes, cibadmin is the interface
for CIB XML configuration changes. Other tools can perform certain
configuration changes at a logically higher level (for example,
crm_attribute can set node attributes in the CIB), but cibadmin can
handle any XML changes.

>  
> Also, it is not clear how to manage resource errors and cleanup? Are
> there other ways that call to crm_resource for cleanup and failed
> resource restart? Could it be made via CIB manipulation like force
> lrm history records delete?

That's what crm_resource --cleanup is for

>  
> I understand that the source is the ultimate answer for any question,
> but I will be very grateful for any advice from ones who has all the
> answers on their fingertips.
>  
> Thank you in advance for sharing your thoughts and experience!
>  
> Sincerely,
>  
> Alex
>  
>  
> _______________________________________________
> Manage your subscription:
> https://lists.clusterlabs.org/mailman/listinfo/users
> 
> ClusterLabs home: https://www.clusterlabs.org/
-- 
Ken Gaillot <kgaillot at redhat.com>