`Feature`: Target Master Only

[wkleinhenz]

As noted in the reddit this service/container should only be run against a single node to prevent constant migration, while this works it would be nice to have the option to run the service on all nodes in case a node needs to be offline for some reason. If possible could you implement a setting similar to the only_on_master from https://github.com/cvk98/Proxmox-load-balancer.

This would allow the service to be run on all cluster nodes but it would only take action if the targeted node was marked as the master

I really like this project so far, let me know if theres any clarification needed

[gyptazy]

Hey @wkleinhenz,

thanks for your feature request (and also thanks for the link if that project, I wasn't aware of that one!)! Proxmox is a multi-master setup and unless we're running Proxmox HA features (see also the related API https://pve.proxmox.com/pve-docs/api-viewer/#/cluster/ha/status/current), it's a little bit tricky to integrate this. I don't want to force people into enabling HA feature (I also run most of my clusters without enabled HA feature), but I see some possibilities, if a potential option has been defined:

• Check if HA services are enabled, then use this as reference
• Configure a node defined as master in the ProxLB config
• Define a node as master by creating a sorted list of all nodes and picking the first (online and reachable) one

So, I could imagine something like the following config options:

Option 1

master_only: True

This enables the feature for auto-detection and tries to get the status from the HA feature. If this isn't available, it falls back to a sorted list of nodes that are online and reachable and picks the first one.

Option 2

master_only: True
master_node: hypervisor01

This simply defines a static node as the master.

Maybe I've overseen some other possibilities. Happy to hear your (or anyones) feedback. At least, this should be easy to integrate.

Cheers, gyptazy

[wkleinhenz]

I think what you proposed in option 1 makes sense to me

[gyptazy]

Ok, cool. I'll try to integrate it in the next (two) weeks. Unfortunately, I do not have that much time for spare projects next week.

[gyptazy]

Hey @wkleinhenz,

I had some time this morning and implemented this in PR #43. Looks like I had it wrong in my mind and the API HA status is always present in a cluster even without activated HA features which makes it a lot easier to implement.

You can give it a try from PR #43 or the feature branch feature/40-option-run-only-on-master-node. You can find my tests in the PR.

Let me know if this fits your needs.

Thanks, gyptazy