https://github.com/open-telemetry/opentelemetry-collector-contrib/tree/main/receiver/hostmetricsreceiver 

Summary:                                                                                                                     

[hostmetrics receiver] Filesystem scraper produces no metrics.

Component: OpenTelemetry Collector Contrib - hostmetricsreceiver

Affects Version: 0.143.0, 0.144.0

Description{}

The hostmetrics receiver filesystem scraper stopped producing metrics in containerized environments (Kubernetes/OpenShift) after upgrading from collector version 0.142.0 to 0.143.0+.

Possible Root Cause Analysis{}

The regression might be caused by the gopsutil library upgrade from v4.25.11 to v4.25.12 introduced in collector version 0.143.0.

Specifically, https://github.com/shirou/gopsutil/pull/1931 (which fixes https://github.com/shirou/gopsutil/issues/1284)

introduced breaking changes in bind mount detection and filtering logic. This affects how /proc/1/mountinfo is parsed and how bind mounts (like /hostfs) are detected in containerized environments.

  │ Collector Version │ gopsutil Version │ Filesystem Metrics │

  │ 0.142.0           │ v4.25.11         │ Working            │

  │ 0.143.0+          │ v4.25.12         │ Not collected      │

Filesystem Metrics are affected like below:

  - system.filesystem.inodes.usage

  - system.filesystem.usage

  Environment{}

  - Platform: OpenShift/Kubernetes

  - Deployment mode: DaemonSet

  - Configuration: root_path: /hostfs with host filesystem mounted at /hostfs

  Steps to Reproduce:

  1. Clone the distributed-tracing-qe test repository:

  git clone https://github.com/openshift/distributed-tracing-qe.git

  cd distributed-tracing-qe

  2. Ensure you have access to an OpenShift/Kubernetes cluster:

  export KUBECONFIG=/path/to/kubeconfig

  3. Uncomment the metrics in the check_logs.sh script in hostmetrics receiver test. 

   "system.filesystem.inodes.usage"

   "system.filesystem.usage"

  4. Run the hostmetrics receiver test with collector version 0.144.0:

  chainsaw test --skip-delete --test-dir tests/e2e-otel/hostmetricsreceiver/

  5. Observe that system.filesystem.inodes.usage and system.filesystem.usage metrics are not collected.

  6. To verify the issue, check collector logs for filesystem-related metrics:

  kubectl -n chainsaw-hostmetrics logs -l app.kubernetes.io/component=opentelemetry-collector | grep "Name: system.filesystem"

  7. No output is returned (filesystem metrics are missing).

  8. Compare with collector version 0.142.0 (working):

    - Update collector image to 0.142.0 in otel-hostmetricsreceiver.yaml

    - Re-run the test - filesystem metrics are collected successfully

  Expected Behavior{}

The filesystem scraper should collect system.filesystem.inodes.usage and system.filesystem.usage metrics when the host filesystem is bind-mounted to /hostfs with root_path: /hostfs configuration.

Actual Behavior{}

The filesystem scraper produces no metrics. The filesystemscraper InstrumentationScope is completely absent from the collector output, while all other scrapers (cpu, disk, memory, network, paging, processes, process) work correctly.

References:

  - gopsutil PR introducing the change: https://github.com/shirou/gopsutil/pull/1931

  - gopsutil issue fixed by the PR: https://github.com/shirou/gopsutil/issues/1284

  - Related OTEL collector issue (previously fixed, different regression):

  https://github.com/open-telemetry/opentelemetry-collector-contrib/issues/35990