esports/scripts/deploy-remote.sh

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

sha=${1:?commit SHA required}
directory=${2:?deployment directory required}
origin=${3:?public origin required}
case "$sha" in
  *[!0-9a-f]*|'') echo 'Invalid image tag' >&2; exit 2 ;;
esac
case "$directory:$origin" in
  /opt/esports:https://esports.vash.software)
    environment=/etc/esports/prod.env
    migration_environment=/etc/esports/prod-migrations.env
    host_port=3100 ;;
  /opt/esports-staging:https://esports-staging.vash.software)
    environment=/etc/esports-staging/staging.env
    migration_environment=/etc/esports/staging-migrations.env
    host_port=3101 ;;
  *) echo 'Unrecognized deployment target' >&2; exit 2 ;;
esac

image="git.stanrunge.dev/stan/esports:$sha"
compose="$directory/compose.prod.yml"
if [ "$directory" = /opt/esports-staging ]; then
  environment_name=staging
else
  environment_name=production
fi

# The deploy user only has passwordless sudo for tee and docker.
if [ ! -d "$directory" ]; then
  echo "$directory does not exist; create it once on the host" >&2
  exit 1
fi
previous_tag=$(sed -n 's/^IMAGE_TAG=//p' "$directory/release.env" 2>/dev/null | head -n 1 || true)
write_release() {
  printf 'IMAGE_TAG=%s\nHOST_PORT=%s\n' "$1" "$host_port" | sudo tee "$directory/release.env" >/dev/null
}

sudo tee "$compose" >/dev/null <<EOF
services:
  app:
    image: git.stanrunge.dev/stan/esports:\${IMAGE_TAG}
    restart: unless-stopped
    stop_grace_period: 30s
    env_file:
      - path: $environment
        format: raw
    environment:
      PORT: 3000
      HOST: 0.0.0.0
      BODY_SIZE_LIMIT: 6M
      PROTOCOL_HEADER: x-forwarded-proto
      HOST_HEADER: x-forwarded-host
      ADDRESS_HEADER: CF-Connecting-IP
    ports:
      - 127.0.0.1:\${HOST_PORT}:3000
    networks:
      - backend

networks:
  backend:
    external: true
EOF

switched=0
rollback() {
  if [ "$switched" -eq 1 ] && [ -n "$previous_tag" ]; then
    echo "Health check failed; restoring prior image tag $previous_tag" >&2
    write_release "$previous_tag"
    sudo docker compose --env-file "$directory/release.env" -f "$compose" up -d
  fi
}
trap rollback ERR

port=$host_port
health=$(mktemp)
if [ "${FORCE_DEPLOY:-}" != 1 ] && curl --fail --silent --show-error "http://127.0.0.1:${port}/api/health" > "$health"; then
  HEALTH_FILE=$health python3 - <<'PY'
import json, os
data = json.load(open(os.environ["HEALTH_FILE"]))
active = int(data.get("activeMatches") or data.get("irc", {}).get("activeLobbies") or 0)
if active:
    raise SystemExit(f"{active} matches are still live. Drain them from the admin page, or set FORCE_DEPLOY=1.")
PY
fi

# The old release reports lobbies rather than active matches, and its health
# endpoint may be unavailable. The database check covers both releases.
if [ "${FORCE_DEPLOY:-}" != 1 ]; then
  active_count=$(sudo docker run --rm --network backend --env-file "$environment" "$image" \
    bun -e 'const sql = new Bun.SQL(process.env.DB_URL || process.env.DATABASE_URL, { max: 1 }); const states = ["LOBBY", "ROLLING", "CHOOSING", "BANNING", "PICKING", "TB_SELECT", "PLAYING", "CLOSING"]; const [row] = await sql`select count(*)::int as count from match where state in ${sql(states)}`; console.log(row.count); await sql.end();')
  if [ "$active_count" -gt 0 ]; then
    echo "$active_count matches are live in the database. Drain them before deploying." >&2
    exit 1
  fi
fi

db_version=$(sudo docker run --rm --network backend --env-file "$environment" "$image" \
  bun -e 'const sql = new Bun.SQL(process.env.DB_URL || process.env.DATABASE_URL, { max: 1 }); const [row] = await sql`show server_version_num`; console.log(row.server_version_num); await sql.end();')
case "$db_version" in
  *[!0-9]*|'') echo 'Could not determine the PostgreSQL server version' >&2; exit 1 ;;
esac
if [ "$db_version" -ge 100000 ]; then
  db_major=$((db_version / 10000))
else
  db_major="$((db_version / 10000)).$(((db_version / 100) % 100))"
fi
backup_image="postgres:${db_major}-alpine"
backup_dir=/var/backups/esports
backup_stamp=$(date -u +%Y%m%dT%H%M%SZ)
backup_name="${environment_name}-${backup_stamp}-${sha}.dump"
sudo docker pull "$backup_image"
# The container runs as root, so it creates the directory and sets the modes.
sudo docker run --rm --network backend --env-file "$environment" \
  --mount "type=bind,source=$(dirname "$backup_dir"),target=/backups" \
  --env "BACKUP_FILE=$(basename "$backup_dir")/$backup_name" "$backup_image" \
  sh -eu -c 'mkdir -p -m 0700 "/backups/${BACKUP_FILE%/*}"
    pg_dump --format=custom --no-owner --no-acl --file="/backups/$BACKUP_FILE" "${DB_URL:-$DATABASE_URL}"
    chmod 0600 "/backups/$BACKUP_FILE"
    pg_restore --list "/backups/$BACKUP_FILE" >/dev/null'
echo "Verified $environment_name database backup at $backup_dir/$backup_name"

# The app's database role cannot create tables; the migrations file overrides
# DATABASE_URL with the owner role and keeps the rest (ROOT_ADMIN_USER_ID).
sudo docker run --rm --network backend --env-file "$environment" \
  --env-file "$migration_environment" "$image" bun scripts/migrate.ts

write_release "$sha"
switched=1
sudo docker compose --env-file "$directory/release.env" -f "$compose" up -d
curl --retry 12 --retry-delay 2 --retry-all-errors --fail --silent --show-error \
  "http://127.0.0.1:${port}/api/health" >/dev/null
trap - ERR
echo "Deployed $sha to $origin"