I have a Python monitoring script using psutil inside a Docker container. On a Mac, I want it to report the laptop's actual physical resources—such as 24 GB of host RAM and host CPU utilization—instead of the metrics from Docker Desktop's Linux virtual machine. Mounting /proc works for host metrics on native Linux, but not on macOS. Is there a way for the container to access macOS host metrics directly without installing the monitoring agent natively on the host? Ideally, the same approach should also work on Windows and be simple enough for other users to set up.
4 Answers
Not directly. On macOS and Windows, Docker Desktop normally runs containers inside a Linux virtual machine, so psutil and /proc can only see the VM's kernel and resource limits. They cannot inspect the host OS just because the container is running on the same computer.
The practical solution is to run a small telemetry exporter on the host—such as a platform-specific Prometheus exporter—and have the container query it over the local network. That keeps the main monitoring application in the container, but something still has to run on macOS or Windows to collect the host's native metrics.
Exactly. The exporter uses the host's native APIs, while the container consumes the exported data. This is much more reliable than trying to mount a Linux-style /proc filesystem.
You could build separate host collectors for each platform, but there is no portable psutil or Docker setting that makes one container automatically see the physical Mac or Windows hardware. For a cross-platform product, use a lightweight native collector or exporter per operating system and let the container retrieve the results through a defined API.
On native Linux, a container can sometimes access host-like metrics by sharing or mounting the relevant kernel interfaces, subject to permissions and namespace settings. That does not translate to Docker Desktop on macOS or Windows because the container is isolated from the actual host kernel. A mounted filesystem would not provide a compatible macOS or Windows version of /proc anyway.
So the container would treat the host exporter like a separate machine and read its metrics over HTTP?