No support for temperature reading with docker in guest VM?

[jonesofarc]

Hi,

I run docker with a guest VM running on proxmox.
The instructions are clear enough that the sensor proxy has to be installed on proxmox and the socket shared to the container via bind mount.

However, I run docker in a guest VM, which is reasonably common I think and thus cannot pass the socket to the container via bind mount. given that it first has to be exposed to the guest VM.

Are there any work arounds for this?

[rcourtman]

Docker running inside a Proxmox guest cannot mount the host's pulse-sensor-proxy socket, so containerized Pulse in that VM will never see temperatures. That limitation is real.

The right fix is already partially in place. pulse-host-agent—which runs directly on the Proxmox host—has report fields for temperatures, but the code that reads lm-sensors and fills those fields has not shipped yet. Once that collection path lands, you will be able to install pulse-host-agent on the Proxmox host, let it read sensors locally, and have it report temperatures back to your Pulse instance over HTTPS. That works across VM boundaries without exposing the sensor proxy or handing SSH keys into containers.

This issue is being tracked as an enhancement to finish temperature collection inside pulse-host-agent. After the feature is implemented, would you be willing to test it in your environment and confirm the Docker-in-VM flow behaves as expected?

[rcourtman]

The temperature collection feature for pulse-host-agent has been implemented and pushed to main (commit 200d491ae).

You can now test this by installing pulse-host-agent on your Proxmox host. The agent will automatically collect and report temperature data if lm-sensors is installed.

Installation steps are documented in TEMPERATURE_MONITORING.md under the "Docker in VM Setup" section. Let me know if you run into any issues or if the temperatures are reporting correctly in your environment.

[jonesofarc]

Wow amazing response thank you!

Okay as this isn't in a release yet, I added the latest code to my CI/CD and have build a new container. I have followed the docker temperature setup , but no temperature is shown:

[jonesofarc]

BTW in your docs you say "Verify temperatures appear in Pulse UI under the Servers tab" . There is no "Servers" tab, its still just "Hosts", don't know if thats just a typo.

[jonesofarc]

If it helps, here is the sensor report when running sensors, a you can see I have a few that don't report and give i/o errors,
but the coretemps are reported?

iwlwifi_1-virtual-0
Adapter: Virtual device
temp1: N/A

acpitz-acpi-0
Adapter: ACPI interface
temp1: +27.8 C

coretemp-isa-0000
Adapter: ISA adapter
Package id 0: +47.0 C (high = +100.0 C, crit = +100.0 C)
Core 0: +44.0 C (high = +100.0 C, crit = +100.0 C)
Core 1: +45.0 C (high = +100.0 C, crit = +100.0 C)
Core 2: +43.0 C (high = +100.0 C, crit = +100.0 C)
Core 3: +44.0 C (high = +100.0 C, crit = +100.0 C)
Core 4: +43.0 C (high = +100.0 C, crit = +100.0 C)
Core 5: +46.0 C (high = +100.0 C, crit = +100.0 C)
Core 6: +43.0 C (high = +100.0 C, crit = +100.0 C)
Core 7: +42.0 C (high = +100.0 C, crit = +100.0 C)

nvme-pci-0300
Adapter: PCI adapter
Composite: +36.9 C (low = -20.1 C, high = +69.8 C)
(crit = +79.8 C)
ERROR: Can't get value of subfeature temp2_min: I/O error
ERROR: Can't get value of subfeature temp2_max: I/O error
Sensor 1: +51.9 C (low = +0.0 C, high = +0.0 C)

TBH I don't understand Composite and Sensor1 readings for nvme-pci-0300 though

[rcourtman]

Temperatures stay empty because the host agent skips collection unless a.platform equals the literal word linux. normalisePlatform() at internal/hostagent/agent.go line 293 actually returns the distro name, so on Proxmox or Debian the check at line 316 always thinks it is not Linux and bails before running sensors -j. That means every Linux host is currently blocked, not just your setup.

The fix is straightforward. The guard will be updated to check the actual runtime OS (GOOS) or the platform family instead of the distro string. That change will ship in the next release.

Until then it would help to confirm the behavior on your node. Run pulse-host-agent with debug logging enabled or check journalctl -u pulse-host-agent and see whether it prints "Failed to collect sensor data". Sharing that snippet will verify we are hitting the same early-return path.

[jonesofarc]

Hmm, running journalctl -u pulse-host-agent does not show any error or any failed message, simply just

"Started pulse-host-agent.service - Pulse Host Agent."

[rcourtman]

Rebuilding the Docker image already pulled in the fixed host agent binaries, so the missing temperatures now point to the service on the Proxmox host still running the previous build or not running at all.

Check systemctl status pulse-host-agent and confirm the unit is active. Run the exact binary shown in ExecStart with --version to make sure it matches the commit you just built. If the version looks old or the path is unexpected, rerun scripts/install-host-agent.sh --url <pulse-url> --token <token> on the host and restart the service afterward.

The agent only reads JSON output, so run sensors -j directly on the host. If that command returns an error or just {} even though plain sensors shows numbers, lm-sensors might need reinstalling or updating.

If the binary and sensors output look fine but temperatures still stay empty, enable debug logging to see what the agent is doing. Add Environment=PULSE_LOG_LEVEL=debug to /etc/systemd/system/pulse-host-agent.service, run systemctl daemon-reload, restart the service, then watch journalctl -u pulse-host-agent -f for sensor messages or HTTP errors.

The Servers vs Hosts wording in the docs is fixed in commit 082c6c2.

[jonesofarc]

Ah it was because of errors breaking the json. I told it to ignore some sensors and they are appearing :)

I notice that they are not displayed in the main details , but only when the details are expanded, making it kinda hidden.