Add auto-detection of container information to resource if cgroupv2 is used

[svrnm]

Issue with OpenTelemetry.Extensions.Docker

Is this a feature request or a bug?

Feature Request

What is the expected behavior?

When using a docker container with cgroupv2 turned on I would like to see a container.id being automatically detected.

What is the actual behavior?

When I use a docker container with cgroupv2 turned on I don't get the container.id detected automatically

Additional Context

This is a known "issue", it has been fixed with the JavaScript implementation already (ContainerDetector.ts):

Detection of the container.id in ContainerResource.java currently works by reading from /proc/self/cgroup. That does not work when cgroupv2 is turned on. However, based on these two sources there's another place to read from:

• How to get docker container ID from within the container with cgroup v2
• Python NullResource error when running torizoncore-builder build

In short, the container.id could be read from /proc/self/mountinfo as well. This also works for containerd and not only docker.

Eventually it would be good to have a standard for that, but unfortunately this is a long-standing non-moving OCI issue, until then a combination of both approaches seems to be reliable enough.

Related

We wanted to use the container.id as an example for resource detection at the OTel docs: open-telemetry/opentelemetry.io#1835
Same ticket for Java Instrumentation: open-telemetry/opentelemetry-java-instrumentation#6694

cc @cartermp, @swetharavichandrancisco

Containerd

The Node.JS implementation also works with other container runtime implementations, namely contaienrd, probably others as well.

[iskiselev]

Fixed with #839

[Kielek]

Closing: #693 (comment)