codexpi/scripts/redeploy.sh

#!/usr/bin/env bash
set -euo pipefail

BASE="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
cd "$BASE"

# The Codex container can mount its default Docker config read-only. Keep build
# metadata in a disposable location so a deployment never fails for that reason.
export DOCKER_CONFIG="${CODEXPI_DOCKER_CONFIG:-/tmp/codexpi-docker-config}"
mkdir -p "$DOCKER_CONFIG"

command -v docker >/dev/null || { echo "Docker est requis." >&2; exit 1; }
if ! docker info >/dev/null 2>&1; then
  socket_gid="$(stat -c '%g' /var/run/docker.sock 2>/dev/null || true)"
  if [[ -n "$socket_gid" && " $(id -G) " != *" $socket_gid "* ]]; then
    echo "Docker n’est pas accessible : le processus courant n’appartient pas au groupe $socket_gid du socket." >&2
    echo "Depuis l’hôte, installez le service de boot puis relancez ce script pour recréer l’API avec le bon groupe :" >&2
    echo "  ./scripts/install-boot-service.sh" >&2
    echo "  ./scripts/redeploy.sh" >&2
  else
    echo "Docker n’est pas accessible." >&2
  fi
  exit 1
fi
docker_gid="$(stat -c '%g' /var/run/docker.sock)"

if docker ps --format '{{.Names}}' | grep -q '^codexpi-safe-redeploy-'; then
  echo "Un déploiement CodexPi est déjà en cours ; aucune seconde bascule ne sera lancée."
  exit 0
fi
if docker inspect codexpi-api-next >/dev/null 2>&1 || docker inspect codexpi-django-next >/dev/null 2>&1; then
  echo "Une version de relève est encore active. Vérifiez-la avant de relancer un déploiement." >&2
  exit 1
fi

stamp="$(date +%s)"
image="codexpi-app:deploy-$stamp"
echo "Construction isolée de la nouvelle image CodexPi ($image)…"
docker build --target production -t "$image" "$BASE"

helper="codexpi-safe-redeploy-$stamp"
docker run -d --rm --name "$helper" \
  -e CODEXPI_DEPLOY_IMAGE="$image" \
  -e DOCKER_GID="$docker_gid" \
  -v /var/run/docker.sock:/var/run/docker.sock \
  -v "$BASE:$BASE" -w "$BASE" \
  docker:cli sh -c '
    set -eu
    api_next=codexpi-api-next
    django_next=codexpi-django-next
    switched=0
    cleanup() {
      if [ "$switched" = 0 ]; then
        docker rm -f "$api_next" "$django_next" >/dev/null 2>&1 || true
      fi
    }
    trap cleanup EXIT
    wait_healthy() {
      container="$1"
      attempts=0
      while [ "$attempts" -lt 60 ]; do
        health=$(docker inspect -f "{{if .State.Health}}{{.State.Health.Status}}{{else}}{{.State.Status}}{{end}}" "$container" 2>/dev/null || true)
        [ "$health" = healthy ] && return 0
        attempts=$((attempts + 1))
        sleep 2
      done
      echo "Le conteneur $container n’est pas devenu sain." >&2
      return 1
    }
    wait_drained() {
      container="$1"
      quiet=0
      echo "Drainage de $container : les discussions en cours restent actives…"
      # Five consecutive quiet checks are enough to confirm that the old API no
      # longer owns a live session, without delaying every deployment by five
      # minutes once the last worker has exited.
      while [ "$quiet" -lt 5 ]; do
        if docker top "$container" -eo pid,args 2>/dev/null | grep -Eq "codexpi\.session_worker|/codex .* exec"; then
          quiet=0
        else
          quiet=$((quiet + 1))
        fi
        sleep 2
      done
    }
    echo "Démarrage de la nouvelle version en parallèle…"
    docker rm -f "$api_next" "$django_next" >/dev/null 2>&1 || true
    CODEXPI_IMAGE="$CODEXPI_DEPLOY_IMAGE" docker compose --project-directory "'"$BASE"'" -f "'"$BASE"'/docker-compose.yml" run -d --no-deps --name "$api_next" -e CODEXPI_SKIP_WORKER_RECOVERY=true api >/dev/null
    CODEXPI_IMAGE="$CODEXPI_DEPLOY_IMAGE" docker compose --project-directory "'"$BASE"'" -f "'"$BASE"'/docker-compose.yml" run -d --no-deps --name "$django_next" django-web >/dev/null
    wait_healthy "$api_next"
    wait_healthy "$django_next"

    echo "Bascule Nginx vers la nouvelle version saine…"
    printf "upstream django_ui { server %s:8001; }\\nupstream fastapi_api { server %s:8000; }\\n" "$django_next" "$api_next" | docker exec -i codexpi-nginx-1 sh -c "cat > /tmp/codexpi-upstreams.conf"
    docker exec codexpi-nginx-1 nginx -t
    docker exec codexpi-nginx-1 nginx -s reload
    switched=1

    wait_drained codexpi-api-1
    echo "Remplacement de l’ancienne version désormais inactive…"
    # Promote only the image which already passed both health checks. The
    # stable tag is what systemd/docker compose will reuse after a host reboot.
    docker image tag "$CODEXPI_DEPLOY_IMAGE" codexpi-app:local
    CODEXPI_IMAGE=codexpi-app:local docker compose --project-directory "'"$BASE"'" -f "'"$BASE"'/docker-compose.yml" up -d --no-deps --force-recreate api django-web
    wait_healthy codexpi-api-1
    wait_healthy codexpi-django-web-1

    echo "Retour vers les services principaux et nettoyage…"
    printf "upstream django_ui { server django-web:8001; }\\nupstream fastapi_api { server api:8000; }\\n" | docker exec -i codexpi-nginx-1 sh -c "cat > /tmp/codexpi-upstreams.conf"
    docker exec codexpi-nginx-1 nginx -t
    docker exec codexpi-nginx-1 nginx -s reload
    wait_drained "$api_next"
    docker rm -f "$api_next" "$django_next" >/dev/null
    switched=0
    trap - EXIT
    echo "Redéploiement terminé sans coupure."
  ' >/dev/null

echo "Image construite. Déploiement blue/green programmé ($helper)."
echo "Les anciennes connexions resteront actives jusqu’à la fin de leur discussion."