Unix socket is not removed if athens crashes about athens HOT 4 CLOSED

If a container crashes, the init container is not executed, so that won't work unfortunately.
Not sure if it makes sense to support this in the helm charts. Running athens using a socket requires some kind of proxy in front of it to be usable. And the helm charts doesn't support multiple containers in the deployment.

I have found a way to work around this issue by using abstract sockets. These sockets does not use the file system.
We build our custom athens docker image to install some required tooling, including curl which makes it easier to implement the live and readiness probes:

FROM gomods/athens:v0.13.0
RUN wget -c "https://github.com/Avinode/git-credential-github-apps/releases/download/v1.2.0/git-credential-github-apps_1.2.0_${TARGETOS}_${TARGETARCH}.tar.gz" -O - \
      | tar -xz -C /usr/local/bin/

RUN apk --no-cache add curl
USER athens

Then we run a deployment with two containers, where one of them is a reverse proxy.

containers:
  - name: proxy
    image: proxy-image
    ports:
      - name: http
        containerPort: 8080
    env:
      - name: PORT
        value: "8080"
      - name: PROXY_URL
        value: "unix://athens/@/var/run/athens/athens.sock"        
  - name: athens
    image: custom-athens-docker-image
    env:
      - name: ATHENS_UNIX_SOCKET
        value: "@/var/run/athens/athens.sock"
    readinessProbe:
      exec:
        command:
          - sh
          - -c
          - curl --fail --abstract-unix-socket $(echo "$ATHENS_UNIX_SOCKET" | cut -d @ -f 2) http://athens/healthz
      initialDelaySeconds: 10
      periodSeconds: 3
      failureThreshold: 3
    livenessProbe:
      exec:
        command:
          - sh
          - -c
          - curl --fail --abstract-unix-socket $(echo "$ATHENS_UNIX_SOCKET" | cut -d @ -f 2) http://athens/healthz
      initialDelaySeconds: 10
      periodSeconds: 3
      failureThreshold: 3

Thanks for the feedback @DrPsychick , appreciate it 👍

from athens.

Would this not better be suited to be solved in the k8s deployment (helm chart or so), because silently overwriting/deleting an existing socket - I don't know if that is the right approach. I would favour "fail early" and expose that the socket is already there, as it is.

from athens.

It's possible to implement some kind of workaround in k8s, but it's not straight forward.
We ran into this issue when athens was OOMKilled, and was unable to restart as the socket file was left behind.

I see your point of not silently deleting an existing socket file, my use case is probably a bit k8s specific.
Would it be an option to make it configurable (ATHENS_UNIX_SOCKET_DELETE_IF_EXISTS=true|false)?

from athens.

Would an init container, "preparing the volume" not be sufficient for a k8s deployment? If you're using the official helm chart, I'd rather support custom init containers for that to solve this k8s specific issue.

https://github.com/gomods/athens-charts

from athens.