Unable to get a remotely deployed Pulse to communicate with Proxmox VE cluster

[yajith]

I have been trying to get Pulse to communicate with a Proxmox VE cluster I have, but it seems that communication falls apart due to my cluster ring0 / nodename that's on a private network, is not reachable by my Pulse container.

In my setup, or rather the setup I'm trying to build, Pulse is deployed on a docker host that's separate from my Proxmox VE cluster. I have made sure that the network communication is allowed through and when I'm specifying the Host URL, it is this URL I provide (https://public-ip:port for now). Once I run the quick setup script, I can see Pulse is making a connection initially and the setup dialog closes and the cluster (3 node at the moment) show up under the settings and seem to be trying to collect information.

Observations at that point are:

• Node name specified is not taken, instead it is using the auto-detected actual node name in the cluster which also translates to ring0 interface in my cluster
• Even when I try to change this behavior by adding a custom hostname to the reachable IP, same behavior (I have tried various combinations here including adding a host file based dns name inside of the pulse container with no luck)
• Pulse container log, I can see repeated errors for {"level":"info","component":"pulse-config","cluster":"node01","unhealthyEndpoints"

It seems as if Pulse is grabbing the node names from proxmox and using the IPs to which they resolve to and due to the way my setup is, this isn't working. I was wondering if there is a workaround for such situations.

Thank you in advance!

[rcourtman]

The issue is that Pulse auto-discovers cluster nodes from Proxmox's /api2/json/cluster/status endpoint which returns the ring0 (corosync) addresses. When those IPs aren't reachable from Pulse, connections fail.

Currently there's no UI/API to override the auto-discovered cluster endpoint URLs after initial detection.

Workarounds:

1. DNS/hosts override (recommended): In your Pulse container, add /etc/hosts entries mapping the Proxmox node hostnames to their public IPs:

   <public-ip-1>  node01
   <public-ip-2>  node02
   <public-ip-3>  node03
   

   Pulse prefers hostnames when VerifySSL is enabled, so this should redirect connections to the reachable IPs.

2. Add nodes individually: Instead of cluster discovery, you could try adding each node as a separate Proxmox instance using their public URLs. This bypasses cluster auto-discovery but means managing nodes separately.

3. Network routing: If feasible, set up routing from the Pulse host to the cluster's internal network.

The ability to manually override cluster endpoint URLs would help this use case. I'll keep this in mind for a future enhancement.

[yajith]

1. DNS/hosts override (recommended): In your Pulse container, add /etc/hosts entries mapping the Proxmox node hostnames to their public IPs:

I have tried this approach number of times but seemingly ending up with the same results. Basically, nothing much happens after this point.

On the pulse container logs I can see repeated messages similar to this:
{"level":"error","component":"pulse-config","error":"poll_nodes failed on prx-node01: no healthy nodes available in cluster prx-node01 (all 3 endpoints unreachable: [https://192.168.1.23:8006 https://192.168.1.22:8006 https://192.168.1.21:8006])","instance":"prx-node01","time":"2025-11-30T06:32:53Z","message":"Failed to get nodes"}

2. Add nodes individually: Instead of cluster discovery, you could try adding each node as a separate Proxmox instance using their public URLs. This bypasses cluster auto-discovery but means managing nodes separately.

To be honest, I'm not sure how to do this. If you mean, simply keep adding rest of the nodes with their own urls regardless of them already appearing under the first node addition itself (as shown on the screenshot) I have already done that and just ends up with three entries similar to the one in the screenshot, each with the node name specified, but all non-functional. I have looked in the documentation but haven't found any other option (yet) 🙏

[rcourtman]

The error shows Pulse is still trying to reach the internal cluster IPs (192.168.1.21-23) despite your hosts file entries. A few things to verify:

1. Check the hosts file is actually being read by the container:

docker exec <pulse-container-name> cat /etc/hosts

Make sure it shows your custom entries like:

<public-ip>  prx-node01
<public-ip>  prx-node02
<public-ip>  prx-node03

2. The issue is the cluster returns IP addresses, not hostnames:

Looking at your log: https://192.168.1.23:8006 https://192.168.1.22:8006 https://192.168.1.21:8006

The cluster API is returning raw IPs from the corosync ring, not hostnames. The /etc/hosts trick only works when Pulse uses hostnames. Since it is getting IPs directly from the cluster API, it uses those IPs.

Current workaround:

The cleanest solution for your topology is to expose just one node through your firewall/port forward and add it as a single-node instance (not cluster). When you add a single Proxmox node, Pulse uses only that URL and does not try to discover cluster topology.

To do this:

1. Remove the current cluster connection from Pulse
2. Add a new Proxmox VE instance using just https://<public-ip>:8006 (your forwarded port)
3. Pulse will still discover the cluster membership via API, but will route all requests through that single node

Future enhancement:

The ability to manually override cluster endpoint URLs would solve this properly. I have noted this as a potential enhancement - allowing users to specify reachable URLs for each cluster node rather than relying on auto-discovery.

[yajith]

1. Check the hosts file is actually being read by the container:

So, I can confirm that this is in place. I can even do ping prx-node01 from within the pulse container and can see it is successfully reaching the public IP of the correct host etc.

2. The issue is the cluster returns IP addresses, not hostnames:

I'm researching if I can change this so that the PVE is actually returning something different, as far as I could find so far, it will always return the node IP, not hostnames.

The cleanest solution for your topology is to expose just one node through your firewall/port forward and add it as a single-node instance (not cluster). When you add a single Proxmox node, Pulse uses only that URL and does not try to discover cluster topology.

I believe this is what I've attempted already. I have removed and re-added one node from the cluster many times over, each time it seems to detect other node entries (as you can see from the screenshot) but the communication falls apart from there. One thing to note is that these nodes actually have public IPs on the interfaces themselves and there is no port forward involved to complicate things to begin with.

Thank you very much for trying to help so far, I really appreciate it.