@tdk-landscape/tdk-cli-core 1.3.15 → 1.3.26

package/engine/entrypoint.star

@@ -0,0 +1,97 @@
+# =============================================================================
+# 🎯 TILT ENTRYPOINT MODULE (Phase 1 Topologies Migration)
+# =============================================================================
+
+# Core
+load("./topologies/tilt/common/utils.star", _Utils = "Utils")
+load("./topologies/tilt/discovery/loading.star", _Manifest = "Manifest")
+load("./topologies/tilt/config/global.star", _Config = "Config")
+
+# Discovery Daemon (auto-discovery of new services)
+load("./topologies/tilt/discovery/discovery_daemon.star", "run_discovery_daemon")
+
+# Registry (migrated to topologies adapter with autogenerated product outputs)
+load(
+    "./topologies/tilt/discovery/resource_registry.star",
+    _GLOBAL_CONFIG_EXPORT = "GLOBAL_CONFIG_EXPORT",
+    _APP_RESOURCES = "APP_RESOURCES",
+    _INFRA_RESOURCES_EXPORT = "INFRA_RESOURCES_EXPORT",
+    _DDD_LIBS_EXPORT = "DDD_LIBS_EXPORT",
+    _DEFAULTS_EXPORT = "DEFAULTS_EXPORT",
+    _RESOURCE_ALIASES = "RESOURCE_ALIASES",
+    _RESOURCE_PATH_MAP = "RESOURCE_PATH_MAP",
+    _get_platform_libs_export = "get_platform_libs_export",
+    _get_product_libs_export = "get_product_libs_export",
+)
+
+# Providers
+load("./topologies/tilt/generators/vite_config.star", _Vite = "Vite")
+load("./topologies/tilt/generators/tsconfig.star", _TSConfig = "TSConfig")
+load("./topologies/platform/docker/index.star", _Docker = "Docker")
+load("./topologies/tilt/generators/npmrc.star", _PackageConfig = "PackageConfig")
+
+# Lifecycle
+load("./topologies/tilt/resources/deps.star", _Libs = "Libs")
+load("./topologies/tilt/resources/conditions.star", _Database = "Database")
+load("./topologies/tilt/resources/declaration.star", _Orchestrator = "Orchestrator")
+load("./topologies/tilt/resources/ordering.star", _Infra = "Infra")
+load("./topologies/tilt/resources/triggers.star", _Watchers = "Watchers")
+
+Utils = _Utils
+Manifest = _Manifest
+Config = _Config
+
+GLOBAL_CONFIG_EXPORT = _GLOBAL_CONFIG_EXPORT
+APP_RESOURCES = _APP_RESOURCES
+INFRA_RESOURCES_EXPORT = _INFRA_RESOURCES_EXPORT
+DDD_LIBS_EXPORT = _DDD_LIBS_EXPORT
+DEFAULTS_EXPORT = _DEFAULTS_EXPORT
+RESOURCE_ALIASES = _RESOURCE_ALIASES
+RESOURCE_PATH_MAP = _RESOURCE_PATH_MAP
+
+def get_platform_libs_export():
+    return _get_platform_libs_export()
+
+def get_product_libs_export():
+    return _get_product_libs_export()
+
+Vite = _Vite
+TSConfig = _TSConfig
+Docker = _Docker
+PackageConfig = _PackageConfig
+
+Libs = _Libs
+Database = _Database
+Orchestrator = _Orchestrator
+Infra = _Infra
+Watchers = _Watchers
+
+# =============================================================================
+# 🔮 AUTO-DISCOVERY DAEMON
+# =============================================================================
+# Start the service discovery daemon if auto-discovery is enabled
+# This monitors for new service.json files and auto-registers them
+# =============================================================================
+
+_AUTO_DISCOVER_ENABLED = os.environ.get("AUTO_DISCOVER", "true").lower() == "true"
+_DISCOVERY_SCAN_INTERVAL = int(os.environ.get("DISCOVERY_SCAN_INTERVAL", "5"))
+
+if _AUTO_DISCOVER_ENABLED:
+    print("🔮 Auto-discovery enabled (scan interval: {}s)".format(_DISCOVERY_SCAN_INTERVAL))
+    # Run daemon in background via local_resource
+    # Use unique name to avoid conflicts on reload
+    local_resource(
+        name="service-discovery-daemon-v2",
+        serve_cmd="cd {} && python3 discovery/resource_snapshot.py watch".format(os.getcwd()),
+        deps=["services/product"],
+        auto_init=True,
+        labels=["infra", "discovery"],
+    )
+    print("👁️  Service discovery daemon running")
+    print("   └─ New services will be auto-detected and registered")
+
+
+# Inlined constant
+DEFAULTS = {}
+
+GLOBAL_CONFIG = {}

package/engine/lifecycle/orchestrator.star

@@ -0,0 +1,260 @@
+# =============================================================================
+# 🔄 TILT LIFECYCLE - SERVICE STARTUP ORCHESTRATOR
+# =============================================================================
+# Path: .tilt/lifecycle/orchestrator.star
+# Purpose: Service startup sequencing and grace period management
+# =============================================================================
+#
+# This module implements Story 4 - Fix API Gateway 504 Errors:
+# - Service startup sequencing
+# - 30-second grace period configuration
+# - Proper health check orchestration
+# - Zero 504 errors on startup
+#
+# =============================================================================
+
+# === INLINED CONSTANTS for pure extension loading ===
+HEALTH_CHECK_PATH = "/health"
+# === END INLINED CONSTANTS ===
+
+
+load('../topologies/platform/docker/networking/traefik_constants.star',
+    'TRAEFIK_STARTUP_GRACE_PERIOD',
+    'TRAEFIK_RESOURCE_STARTUP_DELAY',
+    'TRAEFIK_HEALTHY_THRESHOLD',
+    'TRAEFIK_HEALTHCHECK_INTERVAL',
+    'TRAEFIK_HEALTHCHECK_TIMEOUT',
+    'TRAEFIK_HEALTHCHECK_RETRIES',
+)
+
+
+# =============================================================================
+# 📋 STARTUP SEQUENCING CONFIGURATION
+# =============================================================================
+
+# Service startup phases - ensure infrastructure is ready before apps
+STARTUP_PHASES = {
+    'infra': [
+        'TDK_postgres',
+        'TDK_nats',
+        'TDK_infisical',
+        'verdaccio',
+    ],
+    'proxy': [
+        'TDK_traefik',
+        'TDK_dashy',
+    ],
+    'migrators': [],  # Populated dynamically based on enabled services
+    'apps': [],       # Populated dynamically based on enabled services
+}
+
+# Grace period configuration for each service type
+GRACE_PERIOD_CONFIG = {
+    'backend': {
+        'start_period': '60s',      # Time before health checks begin
+        'interval': '10s',          # Health check interval
+        'timeout': '5s',            # Health check timeout
+        'retries': 6,               # Retries before marking unhealthy
+        'startup_delay': '10s',       # Delay before service receives traffic
+    },
+    'frontend': {
+        'start_period': '30s',
+        'interval': '10s',
+        'timeout': '5s',
+        'retries': 3,
+        'startup_delay': '5s',
+    },
+    'infra': {
+        'start_period': '30s',
+        'interval': '5s',
+        'timeout': '5s',
+        'retries': 10,
+        'startup_delay': '5s',
+    },
+}
+
+# =============================================================================
+# 🚀 STARTUP SEQUENCING FUNCTIONS
+# =============================================================================
+
+def get_resource_phase(resource_name, resource_config):
+    """Determine which startup phase a service belongs to."""
+    # Check if it's an infrastructure service
+    if resource_name in STARTUP_PHASES['infra']:
+        return 'infra'
+    
+    # Check if it's a proxy service
+    if resource_name in STARTUP_PHASES['proxy']:
+        return 'proxy'
+    
+    # Check if service has migrator
+    resources = resource_config.get('resources', [])
+    for res in resources:
+        if 'migrator' in res.get('name', ''):
+            return 'migrators'
+    
+    # Default to apps
+    return 'apps'
+
+
+def get_grace_period_config(resource_type):
+    """Get grace period configuration for a service type."""
+    return GRACE_PERIOD_CONFIG.get(resource_type, GRACE_PERIOD_CONFIG['backend'])
+
+
+def calculate_total_startup_time(resource_type):
+    """
+    Calculate total time for a service to be considered healthy.
+    Formula: start_period + (retries * interval) + buffer
+    """
+    config = get_grace_period_config(resource_type)
+    start_period_seconds = _parse_duration(config['start_period'])
+    interval_seconds = _parse_duration(config['interval'])
+    retries = config['retries']
+    
+    # Total time = start_period + (retries * interval) + 5s buffer
+    total_seconds = start_period_seconds + (retries * interval_seconds) + 5
+    return total_seconds
+
+
+def _parse_duration(duration_str):
+    """Parse duration string (e.g., '60s', '5m') to seconds."""
+    if not duration_str:
+        return 0
+    
+    duration_str = str(duration_str).strip().lower()
+    
+    if duration_str.endswith('s'):
+        return int(duration_str[:-1])
+    elif duration_str.endswith('m'):
+        return int(duration_str[:-1]) * 60
+    elif duration_str.endswith('h'):
+        return int(duration_str[:-1]) * 3600
+    else:
+        # Assume seconds if no unit
+        return int(duration_str)
+
+
+# =============================================================================
+# 🔍 HEALTH CHECK ORCHESTRATION
+# =============================================================================
+
+def get_health_check_config(resource_name, resource_type='backend', port=3000):
+    """
+    Generate health check configuration for a service.
+    Ensures proper sequencing to prevent 504 errors.
+    """
+    config = get_grace_period_config(resource_type)
+    
+    return {
+        'test': ['CMD', 'curl', '-f', '--max-time', '5', 
+                 'http://localhost:' + str(port) + '/health'],
+        'interval': config['interval'],
+        'timeout': config['timeout'],
+        'retries': config['retries'],
+        'start_period': config['start_period'],
+        'disable': False,
+    }
+
+
+def get_traefik_health_check_labels(resource_name, health_path=HEALTH_CHECK_PATH):
+    """
+    Generate Traefik health check labels for load balancer.
+    These ensure Traefik only routes to healthy services.
+    """
+    return {
+        'traefik.http.services.{}.loadbalancer.healthcheck.path'.format(resource_name): health_path,
+        'traefik.http.services.{}.loadbalancer.healthcheck.interval'.format(resource_name): TRAEFIK_HEALTHCHECK_INTERVAL,
+        'traefik.http.services.{}.loadbalancer.healthcheck.timeout'.format(resource_name): TRAEFIK_HEALTHCHECK_TIMEOUT,
+        # Disable health check initially during startup grace period
+        'traefik.http.services.{}.loadbalancer.healthcheck.disable'.format(resource_name): 'false',
+    }
+
+
+# =============================================================================
+# ⏱️ STARTUP DELAY AND SYNCHRONIZATION
+# =============================================================================
+
+def get_startup_delay(resource_type='backend'):
+    """Get the startup delay for a service type before it receives traffic."""
+    config = get_grace_period_config(resource_type)
+    return config['startup_delay']
+
+
+def should_wait_for_dependencies(resource_config):
+    """
+    Determine if a service should wait for dependencies before starting.
+    Returns list of dependencies to wait for.
+    """
+    dependencies = []
+    
+    # Check for database dependency
+    resources = resource_config.get('resources', [])
+    for res in resources:
+        manifest = res.get('_manifest', {})
+        if manifest.get('databaseName') or manifest.get('features', []).count('prisma') > 0:
+            dependencies.append('database-management')
+            break
+    
+    # Check for NATS dependency
+    for res in resources:
+        manifest = res.get('_manifest', {})
+        if 'nats' in manifest.get('features', []):
+            dependencies.append('messaging')
+            break
+    
+    return dependencies
+
+
+# =============================================================================
+# 🎯 SERVICE READINESS CHECKS
+# =============================================================================
+
+def is_resource_ready(resource_name, resource_config):
+    """
+    Check if a service is ready to receive traffic.
+    Used by Tilt resource_deps to sequence service startup.
+    """
+    phase = get_resource_phase(resource_name, resource_config)
+    
+    # Infrastructure services must be healthy first
+    if phase == 'infra':
+        return True  # These are handled by docker-compose health checks
+    
+    # Proxy services depend on infra
+    if phase == 'proxy':
+        return 'infra'  # Wait for infra phase
+    
+    # Migrators depend on database
+    if phase == 'migrators':
+        return 'database-management'
+    
+    # Apps depend on their migrators and infrastructure
+    return 'migrators'  # Wait for migrators
+
+
+# =============================================================================
+# 📊 ORCHESTRATOR PUBLIC API
+# =============================================================================
+
+Orchestrator = struct(
+    # Phase detection
+    get_resource_phase=get_resource_phase,
+    
+    # Grace period configuration
+    get_grace_period_config=get_grace_period_config,
+    calculate_total_startup_time=calculate_total_startup_time,
+    
+    # Health check configuration
+    get_health_check_config=get_health_check_config,
+    get_traefik_health_check_labels=get_traefik_health_check_labels,
+    
+    # Startup synchronization
+    get_startup_delay=get_startup_delay,
+    should_wait_for_dependencies=should_wait_for_dependencies,
+    is_resource_ready=is_resource_ready,
+    
+    # Constants
+    STARTUP_PHASES=STARTUP_PHASES,
+    GRACE_PERIOD_CONFIG=GRACE_PERIOD_CONFIG,
+)