[ClusterLabs] [EXT] Re: Clarification on resource groups

Eugen Block eblock at nde.ag
Fri Jan 31 14:15:31 UTC 2025 

Hi,

thanks again for taking the time to look into this, I appreciate it!
Currently, I don't have any resource groups. I wanted to understand if  
resource groups can achieve what I currently do with a script. But I  
have the impression that they cannot. I can still try to clarify more,  
I'll only a few services to keep it brief.

controller01:
- virtual-ip
- galera
- rabbit
- clone-cinder (systemd)
- clone-nova (systemd)
- clone-neutron (systemd)

controller02:
- galera
- rabbit
- clone-cinder (systemd)
- clone-nova (systemd)
- clone-neutron (systemd)

The systemd services are multi-active. Now I would like to put only  
the cloned resources into maintenance mode, not the entire node. I  
thought I could define a resource group containing only the systemd  
services, so I could put only them into maintenance mode. That way,  
when controller01 would lose the virtual-ip, it could be moved to  
controller02. But the cloned resources would still be unmanaged so I  
could upgrade them safely. I don't even want to move the group, just  
put it into maintenance mode since the remaining node still has an  
active group of systemd resources.
But from my current understanding, I would have had to create those  
groups during cluster bootstrap, maybe I misunderstood though. I'm  
fine with my script solution for the time being, I was just curious if  
this scenario was already covered by pacemaker.

Thanks,
Eugen

Zitat von "Windl, Ulrich" <u.windl at ukr.de>:

> Hi!
>
> Maybe you should show the resources and their dependencies. If the  
> VIP is in a group, what else is in that group?
> In my first reading I thought you want to manage one resource in a  
> group, but still want the group to move. Is that right?
>
> Kind regards,
> Ulrich Windl
>
>> -----Original Message-----
>> From: Users <users-bounces at clusterlabs.org> On Behalf Of Eugen Block
>> Sent: Monday, January 27, 2025 3:41 PM
>> To: Cluster Labs - All topics related to open-source clustering welcomed
>> <users at clusterlabs.org>
>> Subject: [EXT] Re: [ClusterLabs] Clarification on resource groups
>>
>> Thanks for your response.
>>
>> Maybe I didn't explain myself well enough, I can go into more detail
>> if necessary. I wanted to focus on resource groups first. But let me
>> give an example from a recent upgrade that didn't go as smooth as
>> planned.
>>
>> We have a virtual IP colocated with haproxy, which redirects
>> (OpenStack) API calls to the backend servers. Since there are multiple
>> instances of those API services running, the services are still
>> responsive and functioning properly.
>> So I put one node in maintenance mode to be able to stop all the
>> systemd units wrt OpenStack, but not Galera and RabbitMQ. This node
>> didn't have the VIP at the time. When I was done with the first node,
>> everything was still good. But putting the node with the VIP into
>> maintenance mode was a mistake:
>>
>> During the system update there were also updates for network related
>> packages (I hadn't noticed that on the first node), leading to a
>> restart of the network service, causing the VIP to vanish. But since
>> pacemaker couldn't move the resource, we had an API outage for
>> OpenStack. To avoid that in the future, I will only put some of the
>> resources into maintenance mode, not all of them. That way pacemaker
>> will be able to move the VIP and HAProxy to the already upgraded node,
>> so there will only be a tiny disruption, but most clients won't
>> notice. I successfully tested that behavior on a test cloud, hence I'm
>> confident that this is the right approach in my case.
>>
>> If I put the entire cluster into maintenance mode, the VIP wouldn't be
>> moved away when the network gets interrupted, that would cause a
>> client disruption.
>>
>> Thanks!
>> Eugen
>>
>>
>> Zitat von "Windl, Ulrich" <u.windl at ukr.de>:
>>
>> > Hi!
>> >
>> > Assuming you know what "location" refers to. What you are doing
>> > makes very little sense IMHO:
>> > When working on a service that needs some IP, it makes little sense
>> > to put the service in maintenance mode, but not the IP:
>> > Imagine something bad happens to the network and the cluster wants
>> > to move the IP while you are working on the service. Would you want
>> > that to happen? Also (as I see it), to move the group, the cluster
>> > would stop the service first, but when it's in maintenance mode, it
>> > can't. So it can't move the group (and thus the IP neither).
>> > Pertsonally I put the cluster in maintenance mode as a whole,
>> > preferably for a rather short time. Unless services fail very
>> > frequently, this seems to be a safe mode of operation to me.
>> >
>> > Kind regards,
>> > Ulrich Windl
>> >
>> >> -----Original Message-----
>> >> From: Users <users-bounces at clusterlabs.org> On Behalf Of Eugen Block
>> >> Sent: Monday, January 13, 2025 1:17 PM
>> >> To: users at clusterlabs.org
>> >> Subject: [EXT] [ClusterLabs] Clarification on resource groups
>> >>
>> >> Hi,
>> >>
>> >> I'm hoping to get some clarification on my understanding of resource
>> >> groups [0]. It states:
>> >>
>> >> > One of the most common elements of a cluster is a set of resources
>> >> > that need to be located together, start sequentially, and stop in
>> >> > the reverse order.
>> >>
>> >> Especially the "located together" attribute confuses me.
>> >>
>> >> I'll try to provide some context:
>> >> I have a couple of systemd services as clones and some multi-state
>> >> resources such as galera and rabbitmq, running on two pacemaker nodes.
>> >> In case of an upgrade or any kind of maintenance, I want to use the
>> >> maintenance mode for some resources, but not all of them. For example,
>> >> I want the virtual IP, galera and rabbitmq to be still managed while
>> >> the rest is in maintenance mode. So currently, I would run a for loop
>> >> on the systemd services only, putting them into maintenance. This way,
>> >> if the network stack is updated or something, the virtual IP would be
>> >> moved to the other node. IIUC, this is not covered by the resource
>> >> groups, is it?
>> >>
>> >> Or should I have used it when building the cluster from scratch,
>> >> creating groups containing my systemd services as primitives? And then
>> >> clone a group?
>> >>
>> >> Is there another way of achieving that? I'd appreciate any comments!
>> >>
>> >> Thanks!
>> >> Eugen
>> >>
>> >> [0]
>> >>
>> https://clusterlabs.org/projects/pacemaker/doc/2.1/Pacemaker_Explained/
>> >> singlehtml/index.html#group-resources
>> >>
>> >> _______________________________________________
>> >> Manage your subscription:
>> >> https://lists.clusterlabs.org/mailman/listinfo/users
>> >>
>> >> ClusterLabs home: https://www.clusterlabs.org/
>> > _______________________________________________
>> > Manage your subscription:
>> > https://lists.clusterlabs.org/mailman/listinfo/users
>> >
>> > ClusterLabs home: https://www.clusterlabs.org/
>>
>>
>>
>> _______________________________________________
>> Manage your subscription:
>> https://lists.clusterlabs.org/mailman/listinfo/users
>>
>> ClusterLabs home: https://www.clusterlabs.org/
> _______________________________________________
> Manage your subscription:
> https://lists.clusterlabs.org/mailman/listinfo/users
>
> ClusterLabs home: https://www.clusterlabs.org/

More information about the Users mailing list