goodgo-platform/scripts/deploy-production.sh

#!/usr/bin/env bash
# ==============================================================================
# GoodGo Platform — Manual Production Deploy Script
# Backup for CI/CD pipeline. Use when GitHub Actions is unavailable.
#
# Usage (from the server):
#   cd ~/goodgo
#   ./deploy-production.sh [image-tag]
#
# Usage (from local machine):
#   ssh ubuntu@185.225.232.65 'cd ~/goodgo && ./deploy-production.sh abc1234'
# ==============================================================================

set -euo pipefail

# ── Configuration ─────────────────────────────────────────────────────────────
COMPOSE_FILE="docker-compose.prod.yml"
IMAGE_TAG="${1:-latest}"
HEALTH_URL="http://127.0.0.1:3001/health"
HEALTH_RETRIES=15
HEALTH_INTERVAL=5
ROLLBACK_ON_FAIL=true

# ── Colors ────────────────────────────────────────────────────────────────────
RED='\033[0;31m'
GREEN='\033[0;32m'
YELLOW='\033[1;33m'
CYAN='\033[0;36m'
NC='\033[0m'

log()  { echo -e "${GREEN}[DEPLOY]${NC} $(date +%H:%M:%S) $*"; }
warn() { echo -e "${YELLOW}[WARN]${NC}   $(date +%H:%M:%S) $*"; }
err()  { echo -e "${RED}[ERROR]${NC}  $(date +%H:%M:%S) $*" >&2; }
info() { echo -e "${CYAN}[INFO]${NC}   $(date +%H:%M:%S) $*"; }

# ── Pre-flight Checks ────────────────────────────────────────────────────────
if [ ! -f "$COMPOSE_FILE" ]; then
  err "Compose file not found: $COMPOSE_FILE"
  err "Are you in the ~/goodgo directory?"
  exit 1
fi

if [ ! -f ".env" ]; then
  err ".env file not found. Copy from infra/env.production.example"
  exit 1
fi

log "=========================================="
log " GoodGo Platform — Production Deploy"
log " Image tag: ${IMAGE_TAG}"
log " Compose:   ${COMPOSE_FILE}"
log "=========================================="
echo ""

# ── Step 1: Record Current State (for rollback) ──────────────────────────────
log "Step 1/6: Recording current state for rollback..."
PREV_API=$(docker inspect --format='{{.Config.Image}}' goodgo-api 2>/dev/null || echo "none")
PREV_WEB=$(docker inspect --format='{{.Config.Image}}' goodgo-web 2>/dev/null || echo "none")
PREV_AI=$(docker inspect --format='{{.Config.Image}}' goodgo-ai-services 2>/dev/null || echo "none")
info "Previous API: ${PREV_API}"
info "Previous Web: ${PREV_WEB}"
info "Previous AI:  ${PREV_AI}"

# ── Step 2: Pull New Images ──────────────────────────────────────────────────
log "Step 2/6: Pulling new images (tag: ${IMAGE_TAG})..."
export IMAGE_TAG
docker compose -f "$COMPOSE_FILE" pull api web ai-services
log "Images pulled successfully."

# ── Step 3: Rolling Update ───────────────────────────────────────────────────
log "Step 3/6: Rolling update (zero-downtime)..."

info "Updating API..."
docker compose -f "$COMPOSE_FILE" up -d --no-deps --wait api
info "API updated and healthy."

info "Updating Web..."
docker compose -f "$COMPOSE_FILE" up -d --no-deps --wait web
info "Web updated and healthy."

info "Updating AI Services..."
docker compose -f "$COMPOSE_FILE" up -d --no-deps --wait ai-services
info "AI Services updated and healthy."

log "Rolling update complete."

# ── Step 4: Database Migrations ──────────────────────────────────────────────
log "Step 4/6: Running database migrations..."
docker compose -f "$COMPOSE_FILE" exec -T api npx prisma migrate deploy
log "Migrations complete."

# ── Step 5: Health Check Verification ────────────────────────────────────────
log "Step 5/6: Verifying deployment health..."
HEALTHY=false
for i in $(seq 1 "$HEALTH_RETRIES"); do
  if curl -sf "$HEALTH_URL" > /dev/null 2>&1; then
    HEALTHY=true
    break
  fi
  info "Waiting for health check... (${i}/${HEALTH_RETRIES})"
  sleep "$HEALTH_INTERVAL"
done

if $HEALTHY; then
  log "Health check passed!"
else
  err "Health check failed after ${HEALTH_RETRIES} attempts!"

  if $ROLLBACK_ON_FAIL; then
    warn "Initiating rollback..."

    # Rollback: stop current, docker compose will use previously cached images
    docker compose -f "$COMPOSE_FILE" stop api web ai-services
    docker compose -f "$COMPOSE_FILE" up -d --wait api web ai-services

    warn "Rollback complete. Verifying..."
    sleep 5
    if curl -sf "$HEALTH_URL" > /dev/null 2>&1; then
      warn "Services recovered after rollback."
    else
      err "CRITICAL: Services still unhealthy after rollback!"
      err "Manual intervention required."
    fi
  fi

  exit 1
fi

# ── Step 6: Cleanup ──────────────────────────────────────────────────────────
log "Step 6/6: Cleaning up old images..."
docker image prune -f
log "Cleanup complete."

# ── Summary ──────────────────────────────────────────────────────────────────
echo ""
log "=========================================="
log " Deployment successful!"
log "=========================================="
log ""
log " Services:"
info "  API:      $(docker inspect --format='{{.Config.Image}}' goodgo-api)"
info "  Web:      $(docker inspect --format='{{.Config.Image}}' goodgo-web)"
info "  AI:       $(docker inspect --format='{{.Config.Image}}' goodgo-ai-services)"
log ""
log " Endpoints:"
info "  Web:      https://platform.goodgo.vn"
info "  API:      https://api.goodgo.vn"
info "  Grafana:  https://grafana.goodgo.vn"
log ""
log " Run smoke tests:"
info "  ./scripts/smoke-test.sh https://api.goodgo.vn"
log "=========================================="