Suggestion/question

[pbrown-gr]

I was interested when I first learned of PDM. But, I'm not quite sure where it's intended to "live".

Reading through https://pdm.proxmox.com/docs I sort of get mixed messages whether this is intended to be a standalone app you might run on a desktop, or a standalone monitoring host. Or an in-cluster VM. Or an additional node thats PART of the cluster directly. Orrr...... ?

It would be appreciated by new users such as myself, if that was made clearer in the documentation, early on.

 

[Bu66as]

Hi @pbrown-gr, PDM is its own small server and not a desktop app. You install it from its own ISO (or on top of Debian), and it gives you a web UI. It's not a cluster node, just talks to your PVE nodes/clusters via API tokens, so it can manage several separate clusters from one place, migrations between them included.

Where to run it is up to you: a small VM or a separate little box both work fine. A VM inside the cluster you manage is OK too, but keep in mind that PDM is gone exactly when that cluster has problems. I'd put it somewhere outside if you have the option. Doesn't need much in terms of resources.

Agree though, a short "where does this run" paragraph near the start of the docs would help a lot.

 

[J-Rod]

standalone app you might run on a desktop, or a standalone monitoring host. Or an in-cluster VM. Or an additional node thats PART of the cluster directly

All of those options are technically possible and I could invent scenarios where each one would be the best choice.

However, I think the most sensible/common solution is to install it from the ISO as a VM in your PVE host, cluster or HA cluster.

As a side note, Proxmox PVE/PBS/PDM design isn't a trivial topic, so it would be ideal if you provided information on your environment, workload and requirements for reliability and data safety so we can all throw in our opinions on the best configuration to build. Or if it's a home lab, what hardware you already have that you're "forced" to work with.

 

[pbrown-gr]

I have an existing 8 node cluster running 100-ish vms.
No other clusters.

It seems a little overkill for my purposes. Originally, all I was looking for, was a VIP to access the cluster reliably for admin and API purposes.
I'm kinda surprised a simple service for this doesnt already exist. I'll probably write an LXC to handle this, I guess...

 

[Bu66as]

Fair enough, for a single cluster PDM really is overkill. Every node runs pveproxy and serves the full cluster UI and API, and the auth tickets are valid on every node in the cluster. A VIP that just moves between nodes therefore works fine. The simplest version is keepalived with VRRP directly on 2-3 of the nodes. It isn't officially supported, but I've seen it running without problems in a few setups. Just don't put the VIP on the corosync network. If you go the LXC route, I'd make it two containers with HAProxy + keepalived rather than one, otherwise you've just moved the single point of failure into the container Either way, it's the certificate: each node has its own cert, so via the VIP you'll get name mismatches. Either add the VIP's hostname to the node certs via ACME, or terminate TLS on the HAProxy with a proper cert for that name.