#!/usr/bin/env bash
# Zero-downtime blue-green deploy for MuseHub.
#
# Strategy:
#   Two slots — blue (port 10003) and green (port 10004).
#   The active slot serves traffic via nginx. The inactive slot is stopped.
#   Deploy:
#     1. Pull the new image from ECR (old slot keeps serving).
#     2. Run migrations against the live DB (before swap — forward-compatible).
#     3. Start the inactive slot with the new image.
#     4. Health-check the new slot.
#     5. Flip nginx to the new slot (nginx -s reload — instant, zero downtime).
#     6. Stop the old slot.
#
# Called by deploy/push.sh via SSM — do not run directly in production.
# For manual use on the instance:
#   ECR_IMAGE=992382692655.dkr.ecr.us-east-1.amazonaws.com/musehub/musehub \
#   IMAGE_TAG=<tag> bash deploy/deploy.sh
#
# First-time setup:
#   bash deploy/deploy.sh --init
#   (Initialises .active-slot and /etc/nginx/musehub-active-port if missing)

set -euo pipefail

APP_DIR="/opt/musehub"
SLOT_FILE="$APP_DIR/.active-slot"
NGINX_PORT_FILE="/etc/nginx/musehub-active-port"
ECR_REGISTRY="992382692655.dkr.ecr.us-east-1.amazonaws.com"
ECR_IMAGE="${ECR_IMAGE:-${ECR_REGISTRY}/musehub/musehub}"
IMAGE_TAG="${IMAGE_TAG:-latest}"
FULL_IMAGE="${ECR_IMAGE}:${IMAGE_TAG}"
REGION="us-east-1"
HEALTH_URL_BLUE="http://127.0.0.1:10003/healthz"
HEALTH_URL_GREEN="http://127.0.0.1:10004/healthz"
HEALTH_RETRIES=30              # × 2s = 60s max wait

cd "$APP_DIR"

# ── Helpers ───────────────────────────────────────────────────────────────────

log()  { echo "[deploy] $*"; }
die()  { echo "[deploy] ERROR: $*" >&2; exit 1; }

health_check() {
    local url="$1"
    local slot="$2"
    log "Health-checking $slot at $url ..."
    for i in $(seq 1 "$HEALTH_RETRIES"); do
        if curl -sf --max-time 3 "$url" > /dev/null 2>&1; then
            log "$slot is healthy (attempt $i)"
            return 0
        fi
        sleep 2
    done
    die "$slot failed health check after $((HEALTH_RETRIES * 2))s"
}

nginx_point_to() {
    local port="$1"
    echo "server 127.0.0.1:${port};" | sudo tee "$NGINX_PORT_FILE" > /dev/null
    sudo nginx -s reload
    log "nginx now pointing to port $port"
}

# ── Init mode ─────────────────────────────────────────────────────────────────

if [ "${1:-}" = "--init" ]; then
    log "Init: writing .active-slot=blue and nginx upstream file"
    echo "blue" > "$SLOT_FILE"
    echo "server 127.0.0.1:10003;" | sudo tee "$NGINX_PORT_FILE" > /dev/null
    sudo nginx -s reload
    log "Done. Run 'bash deploy/deploy.sh' (with ECR_IMAGE and IMAGE_TAG set) to deploy."
    exit 0
fi

# ── Validate required env vars ────────────────────────────────────────────────

[ -n "${ECR_IMAGE:-}" ] || die "ECR_IMAGE is not set."
[ -n "${IMAGE_TAG:-}" ] || die "IMAGE_TAG is not set."

# ── Read active slot ──────────────────────────────────────────────────────────

if [ ! -f "$SLOT_FILE" ]; then
    die ".active-slot not found. Run: bash deploy/deploy.sh --init"
fi

ACTIVE_SLOT=$(cat "$SLOT_FILE")
if [ "$ACTIVE_SLOT" = "blue" ]; then
    NEW_SLOT="green"
    NEW_PORT=10004
    OLD_CONTAINER="musehub-blue"
    NEW_CONTAINER="musehub-green"
    HEALTH_URL="$HEALTH_URL_GREEN"
else
    NEW_SLOT="blue"
    NEW_PORT=10003
    OLD_CONTAINER="musehub-green"
    NEW_CONTAINER="musehub-blue"
    HEALTH_URL="$HEALTH_URL_BLUE"
fi

log "Image:       $FULL_IMAGE"
log "Active slot: $ACTIVE_SLOT → deploying to: $NEW_SLOT (port $NEW_PORT)"

# ── Step 1: Login to ECR and pull new image ───────────────────────────────────

log "[1/6] Pulling image from ECR..."
aws ecr get-login-password --region "$REGION" | \
    sudo docker login --username AWS --password-stdin "$ECR_REGISTRY"
sudo docker pull "$FULL_IMAGE"
log "Pull complete."

# ── Step 2: Run migrations against the live DB ────────────────────────────────

log "[2/6] Running migrations..."
DB_PASSWORD=$(grep ^DB_PASSWORD "$APP_DIR/.env" | cut -d= -f2)
sudo docker run --rm \
    --network musehub_musehub-internal \
    --env-file "$APP_DIR/.env" \
    -e "DATABASE_URL=postgresql+asyncpg://musehub:${DB_PASSWORD}@postgres:5432/musehub" \
    -e SKIP_MIGRATIONS=0 \
    "$FULL_IMAGE" alembic upgrade head
log "Migrations complete."

# ── Step 3: Start the new slot ────────────────────────────────────────────────

log "[3/6] Starting $NEW_SLOT on port $NEW_PORT..."

# Remove if a failed previous deploy left it around
sudo docker rm -f "$NEW_CONTAINER" 2>/dev/null || true

sudo docker run -d \
    --name "$NEW_CONTAINER" \
    --network musehub_musehub-internal \
    --env-file "$APP_DIR/.env" \
    -e "DATABASE_URL=postgresql+asyncpg://musehub:${DB_PASSWORD}@postgres:5432/musehub" \
    -e SKIP_MIGRATIONS=1 \
    -v musehub_data:/data \
    -p "127.0.0.1:${NEW_PORT}:10003" \
    --restart unless-stopped \
    "$FULL_IMAGE"

# ── Step 4: Health-check the new slot ────────────────────────────────────────

health_check "$HEALTH_URL" "$NEW_SLOT"

# ── Step 5: Flip nginx to the new slot (instant, zero downtime) ───────────────

log "[5/6] Switching nginx to $NEW_SLOT (port $NEW_PORT)..."
nginx_point_to "$NEW_PORT"
echo "$NEW_SLOT" > "$SLOT_FILE"

# ── Step 6: Stop the old slot ────────────────────────────────────────────────

log "[6/6] Stopping old slot ($ACTIVE_SLOT)..."
sudo docker rm -f "$OLD_CONTAINER" 2>/dev/null || true

log ""
log "Deploy complete. Active slot: $NEW_SLOT (port $NEW_PORT)"
log "Image: $FULL_IMAGE"