Registry_auth flag not being set correctly on deploys

[2904nando]

URL

No response

Issue Description

I'm using this action for a personal project and it is awesome, just cuts a whole bunch of config I would need to manually set up.

But sometimes it appears that the registry_auth flag is completely ignored. I don't actually know why, I'm using it with the registry_host: ghcr.io and when I run docker service ps <service_name> on the target server, I get the message "No such image: ghcr.io/<image>…"

It's apparently a problem with the registry auth and the only way I can make it work after each deploy is to go into the server and docker service update <service_name> --with-registry-auth and wait for all the replicas to pull the updated image, which works as planned.

My yaml using this action has the following config (removing some stuff for security reasons ofc):

- name: Docker Stack Deploy
        uses: cssnr/stack-deploy-action@v1
        with:
          name: "REDACTED"
          file: "docker-stack.yaml"
          host: ${{ REDACTED }}
          port: ${{ REDACTED }}
          user: deploy
          ssh_key: ${{ REDACTED }}
          env_file: "./envfile"
          registry_auth: true
          registry_host: ghcr.io
          registry_user: ${{ github.actor }}
          registry_pass: ${{ secrets.GITHUB_TOKEN }}

Log Output

"No such image: ghcr.io/<image>…"

[smashedr]

Go to the Action -> Workflow Run -> Logs -> and inspect the deploy command.

That is the exact command used to deploy. If you notice that it missing the registry auth flag or is malformed that could be the issue.

Let me know if you find any anomalies in the commands vs the set options.

echo -e "::group::Deploying Docker ${DEPLOY_TYPE} Stack: \u001b[36;1m${INPUT_NAME}"
echo -e "\u001b[33;1m${COMMAND[*]}\n"
exec 5>&1
set +e
# shellcheck disable=SC2034
STACK_RESULTS=$( "${COMMAND[@]}" 2>&1 | tee >(cat >&5) ; exit "${PIPESTATUS[0]}" )
EXIT_STATUS="$?"
set -e
echo "::endgroup::"

[2904nando]

Yea it seems like the command is running as it should, with the flag. So maybe it is some other bug on the server itself?

[smashedr]

I am at a loss on what could cause the issue. However, if you say that running this command after the deployment fixes it?

docker service update <service_name> --with-registry-auth 

If so, I can add an option to run service update at the end of the deploy. Would this fix the issue?

[2904nando]

Maybe this could fix it, but I'm not sure if this is only hiding the real issue. At this moment it's just inconsistent. Every time the deployments succeeds, but sometimes the replicas start running normally (no manual input at all), other times the service encounters an issue and manually docker service ps <service-name> shows the "No Such image..." error, which is always resolved by the manual update.

I'll search a little bit more to try to find a specific point of failure.

[smashedr]

Just a heads up, I created a new action that may work better for multiple/complex deployments.

This one just sets up SSH and the context, then lets you run deploy commands in subsequent steps.

See the README.md on GitHub for more details: https://github.com/cssnr/docker-context-action

Feel free to open a Discussion or Issue over there if you have more questions...