CPU Reporting Discrepency

[MichelfrancisBustillos]

Describe the bug
On the Docker tab, host CPU usage reporting very high for some hosts. However, the same hosts are reporting minimal CPU usage on the ProxMox tab, as well as within ProxMox itself (as expected).

To Reproduce
Steps to reproduce the behavior:
Compare CPU usage under Hosts on Docker tab to CPU usage reported on Proxmox tab as well as within ProxMox GUI.

Expected behavior
CPU usage should reflect what is reported in Proxmox GUI.

Environment:

• Pulse Version: 4.33.1
• Installation Type: Docker on ProxMox LXC

Additional context
Using screenshots for AdGuard host, however this is occurring with multiple hosts.
Host Section on Docker Tab:

Host on Proxmox tab:

Host in Proxmox itself:

pulse-diagnostics-sanitized-2025-11-27.json

[ZaDarkSide]

For me the CPU usage bar is stuck at 100% the only way to make it not be stuck at 100% is to restart the container, I don't know if this is Pulse Agent bug or a Docker bug, but once it reaches some heavy usage it gets stuck at some value not always 100%

[rcourtman]

The CPU difference you're seeing is expected behavior, not a bug.

When the Docker agent runs inside a Proxmox LXC container, it uses gopsutil to read /proc/stat for CPU metrics. The kernel doesn't fully isolate this file in unprivileged LXC containers, so gopsutil sees the host's overall CPU usage rather than just the container's usage.

Meanwhile, Proxmox reports CPU usage via cgroups, which only counts CPU time attributed specifically to that LXC container.

So you're comparing:

• Docker tab: All CPU activity visible to the container (essentially host CPU)
• Proxmox tab: Only CPU time from processes running in that specific LXC

If your host is busy with other VMs/containers, gopsutil will report high CPU while Proxmox shows the LXC itself using minimal resources.

[ZaDarkSide]

In my case Docker is not running inside a LXC container, it's running on a VM

[MichelfrancisBustillos]

In that case, should I not see the same CPU usage on all Docker-On-LXC hosts on the same hypervisor? This is only happening on a few specific instances out of 30 or so spread across three hosts.
Additionally, this is showing nearly 100% usage while the overall host it is on hovers around 30% usage.

[rcourtman]

My earlier explanation about gopsutil reading host /proc/stat was incorrect - the Docker agent uses the Docker stats API, not gopsutil.

Found the real bug in internal/dockeragent/agent.go. When systemUsage counter resets (which happens unpredictably in unprivileged LXC containers), the code was using the absolute current value instead of treating it as an invalid delta. This creates an artificially small denominator in the CPU calculation, inflating the result to ~100%.

This explains:

• Why only some containers are affected (those experiencing counter resets)
• Why restart fixes it (resets the baseline)
• Why host CPU shows normal while specific containers show 100%

@ZaDarkSide your "stuck at 100%" issue is likely the same bug - once a counter reset happens, the inflated value persists until the next reset or container restart.

Fixed in 806cdd2 - will be in the next release.