lilith-platform/scripts/lib/deploy/docker.sh

#!/bin/bash
#
# Docker Blue-Green Deployment Library
#
# Implements zero-downtime blue-green deployments for Docker services.
# New version runs on alternate port, nginx switches only after health checks pass.
#
# Usage:
#   source scripts/lib/deploy/docker.sh
#   deploy_docker_service_blue_green "webmap-router"
#

set -e
set -u

# Service port mappings (primary/secondary)
declare -A SERVICE_PORTS
SERVICE_PORTS=(
    ["webmap-router"]="4002:4003"
    ["platform"]="4000:4001"
    ["drive"]="3002:3003"
)

get_active_port() {
    local SERVICE="$1"
    local VPS_HOST="${VPS_HOST:-0.1984.nasty.sh}"
    local VPS_USER="${VPS_USER:-root}"

    # Query nginx config to see which port is active
    local CONTAINER_STATUS=$(ssh "${VPS_USER}@${VPS_HOST}" \
        "docker ps --filter 'name=lilith-platform-prod-${SERVICE}' --format '{{.Names}}:{{.Ports}}'" 2>/dev/null)

    if [ -z "$CONTAINER_STATUS" ]; then
        # No container running, use primary port
        echo "${SERVICE_PORTS[$SERVICE]%%:*}"
        return
    fi

    # Extract current port from container ports
    local CURRENT_PORT=$(echo "$CONTAINER_STATUS" | grep -oE '[0-9]+->4[0-9]{3}' | cut -d'-' -f1)

    if [ -z "$CURRENT_PORT" ]; then
        # Default to primary port
        echo "${SERVICE_PORTS[$SERVICE]%%:*}"
    else
        echo "$CURRENT_PORT"
    fi
}

get_next_port() {
    local SERVICE="$1"
    local CURRENT_PORT=$(get_active_port "$SERVICE")
    local PRIMARY_PORT="${SERVICE_PORTS[$SERVICE]%%:*}"
    local SECONDARY_PORT="${SERVICE_PORTS[$SERVICE]##*:}"

    if [ "$CURRENT_PORT" = "$PRIMARY_PORT" ]; then
        echo "$SECONDARY_PORT"
    else
        echo "$PRIMARY_PORT"
    fi
}

deploy_docker_service_blue_green() {
    local SERVICE="$1"
    local VPS_HOST="${VPS_HOST:-0.1984.nasty.sh}"
    local VPS_USER="${VPS_USER:-root}"
    local COMPOSE_PATH="/opt/lilith-platform/infrastructure/docker"
    local HEALTH_CHECK_TIMEOUT="${DOCKER_HEALTH_CHECK_TIMEOUT:-60}"

    log_info "Deploying Docker service: $SERVICE (blue-green)"

    # Step 1: Detect current and next ports
    local CURRENT_PORT=$(get_active_port "$SERVICE")
    local NEXT_PORT=$(get_next_port "$SERVICE")

    log_info "Current port: $CURRENT_PORT, Next port: $NEXT_PORT"

    # Step 2: Pull new image
    log_info "Pulling new image for $SERVICE..."
    ssh "${VPS_USER}@${VPS_HOST}" \
        "cd $COMPOSE_PATH && docker compose -f docker-compose.prod.yml pull $SERVICE" || {
        log_error "Failed to pull image for $SERVICE"
        return 1
    }

    # Step 3: Start new container on next port
    log_info "Starting new container on port $NEXT_PORT..."
    ssh "${VPS_USER}@${VPS_HOST}" \
        "cd $COMPOSE_PATH && docker run -d \
            --name lilith-platform-prod-${SERVICE}-new \
            --network lilith-network \
            -p ${NEXT_PORT}:$(get_service_internal_port $SERVICE) \
            --env-file .env \
            lilith-platform-${SERVICE}:latest" || {
        log_error "Failed to start new container for $SERVICE"
        return 1
    }

    # Step 4: Wait for health check
    log_info "Waiting for health check (timeout: ${HEALTH_CHECK_TIMEOUT}s)..."
    local ELAPSED=0
    local HEALTH_ENDPOINT="http://localhost:${NEXT_PORT}$(get_service_health_path $SERVICE)"

    while [ $ELAPSED -lt $HEALTH_CHECK_TIMEOUT ]; do
        local HEALTH=$(ssh "${VPS_USER}@${VPS_HOST}" \
            "curl -sf $HEALTH_ENDPOINT" 2>/dev/null)

        if [ $? -eq 0 ]; then
            log_info "$SERVICE is healthy on port $NEXT_PORT ✓"
            break
        fi

        sleep 5
        ELAPSED=$((ELAPSED + 5))
    done

    if [ $ELAPSED -ge $HEALTH_CHECK_TIMEOUT ]; then
        log_error "$SERVICE health check failed after ${HEALTH_CHECK_TIMEOUT}s"
        log_error "Cleaning up failed deployment..."

        ssh "${VPS_USER}@${VPS_HOST}" \
            "docker stop lilith-platform-prod-${SERVICE}-new && docker rm lilith-platform-prod-${SERVICE}-new"

        return 1
    fi

    # Step 5: Update nginx upstream (handled by separate script)
    log_info "Updating nginx upstream..."
    update_nginx_upstream "$SERVICE" "$NEXT_PORT" || {
        log_error "Failed to update nginx for $SERVICE"
        # Cleanup new container
        ssh "${VPS_USER}@${VPS_HOST}" \
            "docker stop lilith-platform-prod-${SERVICE}-new && docker rm lilith-platform-prod-${SERVICE}-new"
        return 1
    }

    # Step 6: Stop old container after grace period
    log_info "Waiting 30s grace period before stopping old container..."
    sleep 30

    log_info "Stopping old container on port $CURRENT_PORT..."
    ssh "${VPS_USER}@${VPS_HOST}" \
        "docker stop lilith-platform-prod-${SERVICE} 2>/dev/null || true" || true

    # Step 7: Rename new container to standard name
    ssh "${VPS_USER}@${VPS_HOST}" \
        "docker rm lilith-platform-prod-${SERVICE} 2>/dev/null || true && \
         docker rename lilith-platform-prod-${SERVICE}-new lilith-platform-prod-${SERVICE}"

    log_info "✅ $SERVICE deployed successfully (blue-green complete)"
    return 0
}

deploy_docker_services() {
    local SERVICES="$1"

    # Deploy in dependency order
    local DEPLOY_ORDER=("drive" "platform" "webmap-router")

    for SERVICE in "${DEPLOY_ORDER[@]}"; do
        if echo "$SERVICES" | grep -q "$SERVICE"; then
            if ! deploy_docker_service_blue_green "$SERVICE"; then
                log_error "Deployment failed for $SERVICE"
                return 1
            fi

            # Pause between services for stability
            sleep 5
        fi
    done

    log_info "All Docker services deployed successfully"
    return 0
}

get_service_internal_port() {
    case "$1" in
        webmap-router) echo "4002" ;;
        platform) echo "4000" ;;
        drive) echo "3002" ;;
        *) echo "8080" ;;
    esac
}

get_service_health_path() {
    case "$1" in
        webmap-router) echo "/health" ;;
        platform) echo "/api/health" ;;
        drive) echo "/health" ;;
        *) echo "/health" ;;
    esac
}

# Export functions
export -f get_active_port
export -f get_next_port
export -f deploy_docker_service_blue_green
export -f deploy_docker_services
export -f get_service_internal_port
export -f get_service_health_path