docker.yml

name: Docker Prod

# PUSH TRIGGER REMOVED 2026-07-23: prod moved to k3s. A push here previously ran don's
# self-hosted runner and `docker stack deploy`, respawning the retired revenue stack
# (zombie incident). Manual-only now; k3s deploys via GHCR image + ArgoCD.
on:
  workflow_dispatch:
concurrency:
  group: ${{ github.ref }}
  cancel-in-progress: true
jobs:
  build:
    runs-on: self-hosted
    env:
      # Per-commit tag: swarm only rolls tasks when the service spec changes,
      # a re-deployed identical :latest spec is a no-op and ships nothing.
      NEWAPI_IMAGE: unorouter-new-api:${{ github.sha }}

    steps:
      - uses: actions/checkout@v4

      - name: Create env file
        run: |
          echo DEBUG_CRYPTO_SECRET=${{ secrets.DEBUG_CRYPTO_SECRET }} > .env
          echo DEBUG_POSTGRES_DB=${{ secrets.DEBUG_POSTGRES_DB }} >> .env
          echo DEBUG_POSTGRES_PASSWORD=${{ secrets.DEBUG_POSTGRES_PASSWORD }} >> .env
          echo DEBUG_POSTGRES_USER=${{ secrets.DEBUG_POSTGRES_USER }} >> .env
          echo DEBUG_SESSION_SECRET=${{ secrets.DEBUG_SESSION_SECRET }} >> .env

          echo CRYPTO_SECRET=${{ secrets.CRYPTO_SECRET }} >> .env
          echo POSTGRES_DB=${{ secrets.POSTGRES_DB }} >> .env
          echo POSTGRES_PASSWORD=${{ secrets.POSTGRES_PASSWORD }} >> .env
          echo POSTGRES_USER=${{ secrets.POSTGRES_USER }} >> .env
          echo SESSION_SECRET=${{ secrets.SESSION_SECRET }} >> .env
          echo SMTP_BCC=${{ secrets.SMTP_BCC }} >> .env
          echo POSTHOG_KEY=${{ secrets.POSTHOG_KEY }} >> .env
          echo POSTHOG_HOST=${{ secrets.POSTHOG_HOST }} >> .env
          echo VAPID_PUBLIC_KEY=${{ secrets.VAPID_PUBLIC_KEY }} >> .env
          echo VAPID_PRIVATE_KEY=${{ secrets.VAPID_PRIVATE_KEY }} >> .env

          echo ERROR_LOG_ENABLED=true >> .env

      # Zero-downtime HA via single-node Swarm: the app runs as a swarm stack
      # (master + slave) with start-first rolling behind Traefik's WRR.

      # grep -qx (exact line): plain `grep active` matches the substring in
      # "inactive" and would skip init on a fresh host.
      - name: Ensure swarm active
        run: docker info --format '{{.Swarm.LocalNodeState}}' | grep -qx active || docker swarm init

      # --attachable so the standalone Traefik + plain-compose datastores can join
      # this swarm overlay too. Ignore "already exists" on redeploys.
      - name: Ensure overlay network
        run: docker network create --driver overlay --attachable newapi-overlay 2>/dev/null || true

      # Cold-start safe: --wait blocks until postgres/redis are healthy so the
      # master never stack-deploys into a not-yet-ready DB.
      - name: Ensure datastores healthy (cold-start safe)
        run: docker compose -f docker-compose-traefik.yml up -d --wait --wait-timeout 120 unorouter-new-api-postgres unorouter-new-api-redis

      - name: Ensure support services
        run: docker compose -f docker-compose-traefik.yml up -d unorouter-new-api-pgbackup unorouter-new-api-logrotate unorouter-mcp

      # Local tag only; single node needs no registry push. A 2nd node later would
      # require pushing to a registry (a remote node can't see a local image).
      - name: Build app image
        run: docker build -t "$NEWAPI_IMAGE" -t unorouter-new-api:latest .

      # stack deploy does not auto-load .env, so source it first (else ${POSTGRES_*}
      # interpolate empty). NEWAPI_HOST_DIR from $HOME keeps the stack host-agnostic.
      - name: Deploy app stack (rolling)
        run: |
          set -a; . ./.env; set +a
          export NEWAPI_HOST_DIR="$HOME/new-api"
          docker stack deploy -c docker-stack-traefik.yml --resolve-image never newapi

      # Swarm tasks are named newapi_<service>.<n>.<id>, so resolve by name filter.
      # Must match on $NEWAPI_IMAGE: during a start-first roll the OLD task is
      # still healthy and would pass a plain health check.
      - name: Wait for master healthy
        run: |
          for i in $(seq 1 120); do
            cid=$(docker ps -q -f name=newapi_newapi-master -f health=healthy | head -1)
            if [ -n "$cid" ] \
              && [ "$(docker inspect "$cid" --format '{{.Config.Image}}')" = "$NEWAPI_IMAGE" ] \
              && docker exec "$cid" wget -q -O /dev/null http://localhost:3000/api/status; then
              echo "master healthy on $NEWAPI_IMAGE"; exit 0
            fi
            echo "waiting for master... ($i)"; sleep 2
          done
          echo "master did not become healthy on $NEWAPI_IMAGE in time"; docker service ps newapi_newapi-master --no-trunc; exit 1

      - name: Wait for slaves healthy
        run: |
          for i in $(seq 1 120); do
            n=$(docker ps -q -f name=newapi_newapi-slave -f health=healthy \
              | xargs -r docker inspect --format '{{.Config.Image}}' \
              | grep -cx "$NEWAPI_IMAGE" || true)
            if [ "$n" -ge 2 ]; then
              echo "$n slaves healthy on $NEWAPI_IMAGE"; exit 0
            fi
            echo "waiting for slaves... ($i, healthy on new image: $n)"; sleep 2
          done
          echo "slaves did not become healthy on $NEWAPI_IMAGE in time"; docker service ps newapi_newapi-slave --no-trunc; exit 1

      # - name: Docker run debug
      #   run: docker compose -f docker-compose-debug.yml up -d --build

      - name: remove unused images, containers, volumes, and networks
        run: docker image prune -a -f --filter "until=24h" && docker container prune -f --filter "until=24h" && docker volume prune -f && docker network prune -f --filter "until=24h"