How to achieve automatic zero downtime deployments for Docker Stack?

[avsaase]

Hi, I'm trying to set up automatic deployment from GitHub Actions with zero downtime.

Following the Going Production guide, I build my docker image in a GitHub Action, push it to a registry and then trigger a deployment by calling the api at /api/compose.deploy (updated since I'm using Compose instead of Application).

I have the automatic deployment working with a Compose service but from this comment I understand that zero downtime is not possible with Compose so I'm now trying to switch to Swarm/Stack. The problem is that while I can trigger a redeployment from a GitHub Action, it doesn't pull the latest image so it ends up redeploying the previous version.

My compose file looks like this:

services:
  hello-world-server:
    image: ghcr.io/avsaase/hello-world-server:latest
    deploy:
      replicas: 2
      update_config:
        parallelism: 1
        delay: 5s
        order: start-first
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:3000/health"]
      interval: 30s
      timeout: 10s
      start_period: 5s
      retries: 3
    environment:
      - OTEL_SERVICE_NAME=hello-world-server
      - OTEL_EXPORTER_OTLP_TRACES_ENDPOINT=http://openobserve-kml9i4-openobserve-1:5081
      - OTEL_EXPORTER_OTLP_LOGS_ENDPOINT=http://openobserve-kml9i4-openobserve-1:5081
      - RUST_LOG=info
    networks:
      - dokploy-network

networks:
  dokploy-network:
    external: true

For Compose I was able to force pulling the latest image every time by adding pull_policy: always to the service definition but this is not supported for Docker Stack.

Is there something I'm missing here or is this workflow not supported?

[djsisson]

can you run docker service update --image xxx/xxx:latest --with-registry-auth <SERVICE>

afaik it will always query for the digest even when using the same tag,