Summary

When using the containerd runtime, we are seeing an issue when Concourse is managed. The problem is that after the worker gets reselected (web restart - drain etc.) the worker starts reporting/emitting build logs rather randomly and some logs are lost. This impacts logs of jobs that are running, in parallel with the concourse update. I think this could potentially block users from actually wanting to migrate to containerd, if we are not able to find a solution.

Steps to reproduce

Run concourse deploy and have other long running jobs that emit build logs at the moment

Expected results

All build logs are emitted correctly and visualised in the UI as before the update

Actual results

Check the welcome message, before the recreate and after the recreate. Notice how after we have some logs that are not present in the UI, although the container still resumes it's execution just fine.

Additional context

I have tested this in a bosh environment, haven't been able to test it in docker/kubernetes

Triaging info

• Concourse version: v7.13.1, v7.11.2, v7.11.0
• Browser (if applicable): Chrome
• Did this used to work? Worked/works fine with guardian