fence_recorder: add new fence agent recorder

[ajfloeder]

Summary

Adds fence_recorder, a new Pacemaker fence agent that uses a request/response file pattern to coordinate fencing with external systems.

Use Case

This agent is designed for environments where fencing decisions need to be coordinated with external systems (e.g., storage controllers, cloud providers, or custom orchestration) that perform cleanup operations before fencing completes.

Instead of directly fencing nodes, fence_recorder:

1. Writes a fence request to a configurable directory
2. Waits for an external responder to process the request and write a response
3. Returns success/failure to Pacemaker based on the response

[knet-jenkins]

Can one of the project admins check and authorise this run please: https://ci.kronosnet.org/job/fence-agents/job/fence-agents-pipeline/job/PR-646/1/input

[knet-jenkins]

Can one of the project admins check and authorise this run please: https://ci.kronosnet.org/job/fence-agents/job/fence-agents-pipeline/job/PR-646/2/input

[knet-jenkins]

Can one of the project admins check and authorise this run please: https://ci.kronosnet.org/job/fence-agents/job/fence-agents-pipeline/job/PR-646/3/input

[ofaaland]

We need this fence agent (or something like it) for a composable storage system, we would greatly appreciate review & feedback.

[fabbione]

Hey guys, the maintainer is on Xmas vacation till beginning of Jan.

[oalbrigt]

Do you have a real world use case where this would be needed?

[ajfloeder]

@oalbrigt Our use case has the fence agent and the storage controller on the same node. During the critical moment of fencing (when things are broken), local files are the most reliable way for two different technology stacks (Pacemaker/Python and Kubernetes/Go) to communicate.

[oalbrigt]

What kind of information do you need to send to your storage systems before fencing?

I'm just trying to understand, and also see if there is any other way we can achieve this with what we have, or mabye add it as a feature to an existing agent.

It might also work against the logic of fencing, in that it might end up blocking the fencing of the node, which is also not a good thing as it might lead to data corruption or other issues.

[ajfloeder]

For our scenario, we need to know the name of the node to be fenced. We also need to control when that node is unfenced, so the existence of the files becomes our auditing mechanism.
The dummy fence agent is relatively close, so I will give that some more thought.

[ajfloeder]

The request/response nature of the fence_recorder allows us to have separate software perform the fence action and reliably write the fence response file to indicate the result of the fencing operation. I don't see how fence_dummy could be made to do that.

[oalbrigt]

I'll make a patch, and link it here, so you can test it and give any suggestions for improvements.

[ajfloeder]

Are you open to a recorder mode on the fence_dummy agent? I'm working on that change right now.

[oalbrigt]

Yeah. That sounds good to me.