omnibase_infra 0.2.6__py3-none-any.whl

omnibase_infra/runtime/service_kernel.py

@@ -0,0 +1,1651 @@
+# SPDX-License-Identifier: MIT
+# Copyright (c) 2025 OmniNode Team
+"""ONEX Kernel - Minimal bootstrap for contract-driven runtime.
+
+This is the kernel entrypoint for the ONEX runtime. It provides a contract-driven
+bootstrap that wires configuration into the existing RuntimeHostProcess.
+
+The kernel is responsible for:
+    1. Loading runtime configuration from contracts or environment
+    2. Creating and starting the event bus (EventBusInmemory or EventBusKafka)
+    3. Building the dependency container (event_bus, config)
+    4. Instantiating RuntimeHostProcess with contract-driven configuration
+    5. Starting the HTTP health server for Docker/K8s probes
+    6. Setting up graceful shutdown signal handlers
+    7. Running the runtime until shutdown is requested
+
+Event Bus Selection:
+    The kernel supports two event bus implementations:
+    - EventBusInmemory: For local development and testing (default)
+    - EventBusKafka: For production use with Kafka/Redpanda
+
+    Selection is determined by:
+    - KAFKA_BOOTSTRAP_SERVERS environment variable (if set, uses Kafka)
+    - config.event_bus.type field in runtime_config.yaml
+
+Usage:
+    # Run with default contracts directory (./contracts)
+    python -m omnibase_infra.runtime.service_kernel
+
+    # Run with custom contracts directory
+    ONEX_CONTRACTS_DIR=/path/to/contracts python -m omnibase_infra.runtime.service_kernel
+
+    # Or via the installed entrypoint
+    onex-runtime
+
+Environment Variables:
+    ONEX_CONTRACTS_DIR: Path to contracts directory (default: ./contracts)
+    ONEX_HTTP_PORT: Port for health check HTTP server (default: 8085)
+    ONEX_LOG_LEVEL: Logging level (default: INFO)
+    ONEX_ENVIRONMENT: Runtime environment name (default: local)
+
+Note:
+    This kernel uses the existing RuntimeHostProcess as the core runtime engine.
+    A future refactor may integrate NodeOrchestrator as the primary execution
+    engine, but for MVP this lean kernel provides contract-driven bootstrap
+    with minimal risk and maximum reuse of tested code.
+"""
+
+from __future__ import annotations
+
+import asyncio
+import logging
+import os
+import signal
+import sys
+import time
+from collections.abc import Awaitable, Callable
+from importlib.metadata import version as get_package_version
+from pathlib import Path
+from typing import cast
+from uuid import UUID
+
+import asyncpg
+import yaml
+from pydantic import ValidationError
+
+from omnibase_core.container import ModelONEXContainer
+from omnibase_infra.enums import EnumConsumerGroupPurpose, EnumInfraTransportType
+from omnibase_infra.errors import (
+    ModelInfraErrorContext,
+    ProtocolConfigurationError,
+    RuntimeHostError,
+    ServiceResolutionError,
+)
+from omnibase_infra.event_bus.event_bus_inmemory import EventBusInmemory
+from omnibase_infra.event_bus.event_bus_kafka import EventBusKafka
+from omnibase_infra.event_bus.models.config import ModelKafkaEventBusConfig
+from omnibase_infra.models import ModelNodeIdentity
+from omnibase_infra.nodes.node_registration_orchestrator.dispatchers import (
+    DispatcherNodeIntrospected,
+)
+from omnibase_infra.nodes.node_registration_orchestrator.introspection_event_router import (
+    IntrospectionEventRouter,
+)
+from omnibase_infra.runtime.handler_registry import RegistryProtocolBinding
+from omnibase_infra.runtime.models import (
+    ModelProjectorPluginLoaderConfig,
+    ModelRuntimeConfig,
+)
+from omnibase_infra.runtime.projector_plugin_loader import (
+    ProjectorPluginLoader,
+    ProjectorShell,
+    ProtocolEventProjector,
+)
+from omnibase_infra.runtime.service_runtime_host_process import RuntimeHostProcess
+from omnibase_infra.runtime.util_container_wiring import (
+    wire_infrastructure_services,
+    wire_registration_handlers,
+)
+
+# Circular Import Note (OMN-529):
+# ---------------------------------
+# ServiceHealth and DEFAULT_HTTP_PORT are imported inside bootstrap() rather than
+# at module level to avoid a circular import. The import chain is:
+#
+#   1. omnibase_infra/runtime/__init__.py imports kernel_bootstrap from kernel.py
+#   2. If kernel.py imported ServiceHealth at module level, it would load service_health.py
+#   3. service_health.py imports ModelHealthCheckResponse from runtime.models
+#   4. This triggers initialization of omnibase_infra.runtime package (step 1)
+#   5. Runtime package tries to import kernel.py which is still initializing -> circular!
+#
+# The lazy import in bootstrap() is acceptable because:
+#   - ServiceHealth is only instantiated at runtime, not at import time
+#   - Type checking uses forward references (no import needed)
+#   - No import-time side effects are bypassed
+#   - The omnibase_infra.services.__init__.py already excludes ServiceHealth exports
+#     to prevent accidental circular imports from other modules
+#
+# See also: omnibase_infra/services/__init__.py "ServiceHealth Import Guide" section
+from omnibase_infra.runtime.util_validation import validate_runtime_config
+from omnibase_infra.utils.correlation import generate_correlation_id
+from omnibase_infra.utils.util_error_sanitization import sanitize_error_message
+
+logger = logging.getLogger(__name__)
+
+# Kernel version - read from installed package metadata to avoid version drift
+# between code and pyproject.toml. Falls back to "unknown" if package is not
+# installed (e.g., during development without editable install).
+try:
+    KERNEL_VERSION = get_package_version("omnibase_infra")
+except Exception:
+    KERNEL_VERSION = "unknown"
+
+# Default configuration
+DEFAULT_CONTRACTS_DIR = "./contracts"
+DEFAULT_RUNTIME_CONFIG = "runtime/runtime_config.yaml"
+
+# Environment variable name for contracts directory
+ENV_CONTRACTS_DIR = "ONEX_CONTRACTS_DIR"
+DEFAULT_INPUT_TOPIC = "requests"
+DEFAULT_OUTPUT_TOPIC = "responses"
+DEFAULT_GROUP_ID = "onex-runtime"
+
+# Port validation constants
+MIN_PORT = 1
+MAX_PORT = 65535
+
+
+def _get_contracts_dir() -> Path:
+    """Get contracts directory from environment.
+
+    Reads the ONEX_CONTRACTS_DIR environment variable. If not set,
+    returns the default contracts directory.
+
+    Returns:
+        Path to the contracts directory.
+    """
+    onex_value = os.environ.get(ENV_CONTRACTS_DIR)
+    if onex_value:
+        return Path(onex_value)
+
+    return Path(DEFAULT_CONTRACTS_DIR)
+
+
+def load_runtime_config(
+    contracts_dir: Path,
+    correlation_id: UUID | None = None,
+) -> ModelRuntimeConfig:
+    """Load runtime configuration from contract file or return defaults.
+
+    Attempts to load runtime_config.yaml from the contracts directory.
+    If the file doesn't exist, returns sensible defaults to allow
+    the runtime to start without requiring a config file.
+
+    Configuration Loading Process:
+        1. Check for runtime_config.yaml in contracts directory
+        2. If found, parse YAML and validate against ModelRuntimeConfig schema
+        3. If not found, construct config from environment variables and defaults
+        4. Return fully validated configuration model
+
+    Configuration Precedence:
+        - File-based config is returned as-is when present (no environment overrides)
+        - Environment variables are only used when no config file exists
+        - Defaults are used when neither file nor environment variables are set
+        - Note: Environment overrides (e.g., ONEX_ENVIRONMENT) are applied by the
+          caller (bootstrap), not by this function
+
+    Args:
+        contracts_dir: Path to the contracts directory containing runtime_config.yaml.
+            Example: Path("./contracts") or Path("/app/contracts")
+        correlation_id: Optional correlation ID for distributed tracing. If not
+            provided, a new one will be generated. Passing a correlation_id from
+            the caller (e.g., bootstrap) ensures consistent tracing across the
+            initialization sequence.
+
+    Returns:
+        ModelRuntimeConfig: Fully validated configuration model with runtime settings.
+            Contains event bus configuration, topic names, consumer group, shutdown
+            behavior, and logging configuration.
+
+    Raises:
+        ProtocolConfigurationError: If config file exists but cannot be parsed,
+            fails validation, or cannot be read due to filesystem errors. Error
+            includes correlation_id for tracing and detailed context for debugging.
+
+    Example:
+        >>> contracts_dir = Path("./contracts")
+        >>> config = load_runtime_config(contracts_dir)
+        >>> print(config.input_topic)
+        requests
+        >>> print(config.event_bus.type)
+        inmemory
+
+    Example Error:
+        >>> # If runtime_config.yaml has invalid YAML syntax
+        >>> load_runtime_config(Path("./invalid"))
+        ProtocolConfigurationError: Failed to parse runtime config YAML at ./invalid/runtime/runtime_config.yaml
+        (correlation_id: 123e4567-e89b-12d3-a456-426614174000)
+    """
+    config_path = contracts_dir / DEFAULT_RUNTIME_CONFIG
+    # Use passed correlation_id for consistent tracing, or generate new one
+    effective_correlation_id = correlation_id or generate_correlation_id()
+    context = ModelInfraErrorContext(
+        transport_type=EnumInfraTransportType.RUNTIME,
+        operation="load_config",
+        target_name=str(config_path),
+        correlation_id=effective_correlation_id,
+    )
+
+    if config_path.exists():
+        logger.info(
+            "Loading runtime config from %s (correlation_id=%s)",
+            config_path,
+            effective_correlation_id,
+        )
+        try:
+            with config_path.open(encoding="utf-8") as f:
+                raw_config = yaml.safe_load(f) or {}
+
+            # Type guard: reject non-mapping YAML payloads
+            # yaml.safe_load() can return list, str, int, etc. for valid YAML
+            # but runtime config must be a dict (mapping) for model validation
+            if not isinstance(raw_config, dict):
+                raise ProtocolConfigurationError(
+                    f"Runtime config at {config_path} must be a YAML mapping (dict), "
+                    f"got {type(raw_config).__name__}",
+                    context=context,
+                    config_path=str(config_path),
+                    error_details=f"Expected dict, got {type(raw_config).__name__}",
+                )
+
+            # Contract validation: validate against schema before Pydantic
+            # This provides early, actionable error messages for pattern/range violations
+            contract_errors = validate_runtime_config(raw_config)
+            if contract_errors:
+                error_count = len(contract_errors)
+                # Create concise summary for log message (first 3 errors)
+                error_summary = "; ".join(contract_errors[:3])
+                if error_count > 3:
+                    error_summary += f" (and {error_count - 3} more...)"
+                raise ProtocolConfigurationError(
+                    f"Contract validation failed at {config_path}: {error_count} error(s). "
+                    f"First errors: {error_summary}",
+                    context=context,
+                    config_path=str(config_path),
+                    # Full error list for structured debugging (not truncated)
+                    validation_errors=contract_errors,
+                    error_count=error_count,
+                )
+            logger.debug(
+                "Contract validation passed (correlation_id=%s)",
+                effective_correlation_id,
+            )
+
+            config = ModelRuntimeConfig.model_validate(raw_config)
+            logger.debug(
+                "Runtime config loaded successfully (correlation_id=%s)",
+                effective_correlation_id,
+                extra={
+                    "input_topic": config.input_topic,
+                    "output_topic": config.output_topic,
+                    "consumer_group": config.consumer_group,
+                    "event_bus_type": config.event_bus.type,
+                },
+            )
+            return config
+        except yaml.YAMLError as e:
+            raise ProtocolConfigurationError(
+                f"Failed to parse runtime config YAML at {config_path}: {e}",
+                context=context,
+                config_path=str(config_path),
+                error_details=str(e),
+            ) from e
+        except ValidationError as e:
+            # Extract validation error details for actionable error messages
+            error_count = e.error_count()
+            # Convert Pydantic errors to list[str] for consistency with contract validation
+            # Both validation_errors fields should have the same type: list[str]
+            pydantic_errors = [
+                f"{'.'.join(str(loc) for loc in err['loc'])}: {err['msg']}"
+                for err in e.errors()
+            ]
+            error_summary = "; ".join(pydantic_errors[:3])
+            raise ProtocolConfigurationError(
+                f"Runtime config validation failed at {config_path}: {error_count} error(s). "
+                f"First errors: {error_summary}",
+                context=context,
+                config_path=str(config_path),
+                validation_errors=pydantic_errors,
+                error_count=error_count,
+            ) from e
+        except UnicodeDecodeError as e:
+            raise ProtocolConfigurationError(
+                f"Runtime config file contains binary or non-UTF-8 content: {config_path}",
+                context=context,
+                config_path=str(config_path),
+                error_details=f"Encoding error at position {e.start}-{e.end}: {e.reason}",
+            ) from e
+        except OSError as e:
+            raise ProtocolConfigurationError(
+                f"Failed to read runtime config at {config_path}: {e}",
+                context=context,
+                config_path=str(config_path),
+                error_details=str(e),
+            ) from e
+
+    # No config file - use environment variables and defaults
+    logger.info(
+        "No runtime config found at %s, using environment/defaults (correlation_id=%s)",
+        config_path,
+        effective_correlation_id,
+    )
+    config = ModelRuntimeConfig(
+        input_topic=os.getenv("ONEX_INPUT_TOPIC", DEFAULT_INPUT_TOPIC),
+        output_topic=os.getenv("ONEX_OUTPUT_TOPIC", DEFAULT_OUTPUT_TOPIC),
+        consumer_group=os.getenv("ONEX_GROUP_ID", DEFAULT_GROUP_ID),
+    )
+    logger.debug(
+        "Runtime config constructed from environment/defaults (correlation_id=%s)",
+        effective_correlation_id,
+        extra={
+            "input_topic": config.input_topic,
+            "output_topic": config.output_topic,
+            "consumer_group": config.consumer_group,
+        },
+    )
+    return config
+
+
+async def bootstrap() -> int:
+    """Bootstrap the ONEX runtime from contracts.
+
+    This is the main async entrypoint that orchestrates the complete runtime
+    initialization and lifecycle management. The bootstrap process follows a
+    structured sequence to ensure proper resource initialization and cleanup.
+
+    Bootstrap Sequence:
+        1. Determine contracts directory from ONEX_CONTRACTS_DIR environment variable
+        2. Load and validate runtime configuration from contracts or environment
+        3. Create and initialize event bus (EventBusInmemory or EventBusKafka based on config)
+        4. Create ModelONEXContainer and wire infrastructure services (async)
+        5. Resolve RegistryProtocolBinding from container (async)
+        6. Instantiate RuntimeHostProcess with validated configuration and pre-resolved registry
+        7. Setup graceful shutdown signal handlers (SIGINT, SIGTERM)
+        8. Start runtime and HTTP health server for Docker/Kubernetes health probes
+        9. Run runtime until shutdown signal received
+        10. Perform graceful shutdown with configurable timeout
+        11. Clean up resources in finally block to prevent resource leaks
+
+    Error Handling:
+        - Configuration errors: Logged with full context and correlation_id
+        - Runtime errors: Caught and logged with detailed error information
+        - Unexpected errors: Logged with exception details for debugging
+        - All errors include correlation_id for distributed tracing
+
+    Shutdown Behavior:
+        - Health server stopped first (fast, non-blocking operation)
+        - Runtime stopped with configurable grace period (default: 30s)
+        - Timeout enforcement prevents indefinite shutdown hangs
+        - Finally block ensures cleanup even on unexpected errors
+
+    Returns:
+        Exit code (0 for success, non-zero for errors).
+            - 0: Clean shutdown after successful operation
+            - 1: Configuration error, runtime error, or unexpected failure
+
+    Environment Variables:
+        ONEX_CONTRACTS_DIR: Path to contracts directory (default: ./contracts)
+        ONEX_HTTP_PORT: Port for health check server (default: 8085)
+        ONEX_LOG_LEVEL: Logging level (default: INFO)
+        ONEX_ENVIRONMENT: Environment name (default: local)
+        ONEX_INPUT_TOPIC: Input topic override (default: requests)
+        ONEX_OUTPUT_TOPIC: Output topic override (default: responses)
+        ONEX_GROUP_ID: Consumer group override (default: onex-runtime)
+
+    Example:
+        >>> # Run bootstrap and handle exit code
+        >>> exit_code = await bootstrap()
+        >>> if exit_code == 0:
+        ...     print("Runtime shutdown successfully")
+        ... else:
+        ...     print("Runtime encountered errors")
+
+    Example Startup Log:
+        ============================================================
+        ONEX Runtime Kernel v0.1.0
+        Environment: production
+        Contracts: /app/contracts
+        Event Bus: inmemory (group: onex-runtime)
+        Topics: requests → responses
+        Health endpoint: http://0.0.0.0:8085/health
+        ============================================================
+    """
+    # Lazy import to break circular dependency chain - see "Circular Import Note"
+    # comment near line 98 for detailed explanation of the import cycle.
+    from omnibase_infra.services.service_health import (
+        DEFAULT_HTTP_PORT,
+        ServiceHealth,
+    )
+
+    # Initialize resources to None for cleanup guard in finally block
+    runtime: RuntimeHostProcess | None = None
+    health_server: ServiceHealth | None = None
+    postgres_pool: asyncpg.Pool | None = None
+    introspection_unsubscribe: Callable[[], Awaitable[None]] | None = None
+    correlation_id = generate_correlation_id()
+    bootstrap_start_time = time.time()
+
+    try:
+        # 1. Determine contracts directory
+        contracts_dir = _get_contracts_dir()
+        logger.info(
+            "ONEX Kernel starting with contracts_dir=%s (correlation_id=%s)",
+            contracts_dir,
+            correlation_id,
+        )
+
+        # 2. Load runtime configuration (may raise ProtocolConfigurationError)
+        # Pass correlation_id for consistent tracing across initialization sequence
+        config_start_time = time.time()
+        config = load_runtime_config(contracts_dir, correlation_id=correlation_id)
+        config_duration = time.time() - config_start_time
+        # Log only safe config fields (no credentials or sensitive data)
+        # Full config.model_dump() could leak passwords, API keys, connection strings
+        logger.debug(
+            "Runtime config loaded in %.3fs (correlation_id=%s)",
+            config_duration,
+            correlation_id,
+            extra={
+                "duration_seconds": config_duration,
+                "input_topic": config.input_topic,
+                "output_topic": config.output_topic,
+                "consumer_group": config.consumer_group,
+                "event_bus_type": config.event_bus.type,
+                "shutdown_grace_period": config.shutdown.grace_period_seconds,
+            },
+        )
+
+        # 3. Create event bus
+        # Dispatch based on configuration or environment variable:
+        # - If KAFKA_BOOTSTRAP_SERVERS env var is set, use EventBusKafka
+        # - If config.event_bus.type == "kafka", use EventBusKafka
+        # - Otherwise, use EventBusInmemory for local development/testing
+        # Environment override takes precedence over config for environment field.
+        environment = os.getenv("ONEX_ENVIRONMENT") or config.event_bus.environment
+        kafka_bootstrap_servers = os.getenv("KAFKA_BOOTSTRAP_SERVERS")
+
+        # Explicit bool evaluation (not truthy string) for kafka usage.
+        # KAFKA_BOOTSTRAP_SERVERS env var takes precedence over config.event_bus.type.
+        # This prevents implicit "kafka but localhost" fallback scenarios.
+        use_kafka: bool = (
+            bool(kafka_bootstrap_servers) or config.event_bus.type == "kafka"
+        )
+
+        # Validate bootstrap_servers is provided when kafka is requested via config
+        # This prevents confusing implicit localhost:9092 fallback
+        if use_kafka and not kafka_bootstrap_servers:
+            context = ModelInfraErrorContext(
+                transport_type=EnumInfraTransportType.KAFKA,
+                operation="configure_event_bus",
+                correlation_id=correlation_id,
+            )
+            raise ProtocolConfigurationError(
+                "Kafka event bus requested (config.event_bus.type='kafka') but "
+                "KAFKA_BOOTSTRAP_SERVERS environment variable is not set. "
+                "Set KAFKA_BOOTSTRAP_SERVERS to the broker address (e.g., 'kafka:9092') "
+                "or use event_bus.type='inmemory' for local development.",
+                context=context,
+                parameter="KAFKA_BOOTSTRAP_SERVERS",
+            )
+
+        event_bus_start_time = time.time()
+        event_bus: EventBusInmemory | EventBusKafka
+        event_bus_type: str
+
+        if use_kafka:
+            # Use EventBusKafka for production/integration testing
+            # NOTE: bootstrap_servers is guaranteed non-empty at this point due to validation
+            # above, but mypy cannot narrow the Optional[str] type through control flow.
+            kafka_config = ModelKafkaEventBusConfig(
+                bootstrap_servers=kafka_bootstrap_servers,  # type: ignore[arg-type]  # NOTE: control flow narrowing limitation
+                environment=environment,
+                circuit_breaker_threshold=config.event_bus.circuit_breaker_threshold,
+            )
+            event_bus = EventBusKafka(config=kafka_config)
+            event_bus_type = "kafka"
+
+            # Start EventBusKafka to connect to Kafka/Redpanda and enable consumers
+            # Without this, the event bus cannot publish or consume messages
+            try:
+                await event_bus.start()
+                logger.debug(
+                    "EventBusKafka started successfully (correlation_id=%s)",
+                    correlation_id,
+                )
+            except Exception as e:
+                context = ModelInfraErrorContext(
+                    transport_type=EnumInfraTransportType.KAFKA,
+                    operation="start_event_bus",
+                    correlation_id=correlation_id,
+                    target_name=kafka_bootstrap_servers,
+                )
+                raise RuntimeHostError(
+                    f"Failed to start EventBusKafka: {sanitize_error_message(e)}",
+                    context=context,
+                ) from e
+
+            logger.info(
+                "Using EventBusKafka (correlation_id=%s)",
+                correlation_id,
+                extra={
+                    "bootstrap_servers": kafka_bootstrap_servers,
+                    "environment": environment,
+                    "consumer_group": config.consumer_group,
+                },
+            )
+        else:
+            # Use EventBusInmemory for local development/testing
+            event_bus = EventBusInmemory(
+                environment=environment,
+                group=config.consumer_group,
+            )
+            event_bus_type = "inmemory"
+
+        event_bus_duration = time.time() - event_bus_start_time
+        logger.debug(
+            "Event bus created in %.3fs (correlation_id=%s)",
+            event_bus_duration,
+            correlation_id,
+            extra={
+                "duration_seconds": event_bus_duration,
+                "event_bus_type": event_bus_type,
+                "environment": environment,
+                "consumer_group": config.consumer_group,
+            },
+        )
+
+        # 4. Create and wire container for dependency injection
+        container_start_time = time.time()
+        container = ModelONEXContainer()
+        if container.service_registry is None:
+            logger.warning(
+                "DEGRADED_MODE: service_registry is None (omnibase_core circular import bug?), "
+                "skipping container wiring (correlation_id=%s)",
+                correlation_id,
+                extra={
+                    "error_type": "NoneType",
+                    "correlation_id": correlation_id,
+                    "degraded_mode": True,
+                    "degraded_reason": "service_registry_unavailable",
+                    "component": "container_wiring",
+                },
+            )
+            wire_summary: dict[str, list[str] | str] = {
+                "services": [],
+                "status": "degraded",
+            }  # Empty summary for degraded mode
+        else:
+            try:
+                wire_summary = await wire_infrastructure_services(container)
+            except ServiceResolutionError as e:
+                # Service resolution failed during wiring - container configuration issue.
+                logger.warning(
+                    "DEGRADED_MODE: Container wiring failed due to service resolution error, "
+                    "continuing in degraded mode (correlation_id=%s): %s",
+                    correlation_id,
+                    e,
+                    extra={
+                        "error_type": type(e).__name__,
+                        "correlation_id": correlation_id,
+                        "degraded_mode": True,
+                        "degraded_reason": "service_resolution_error",
+                        "component": "container_wiring",
+                    },
+                )
+                wire_summary = {"services": [], "status": "degraded"}
+            except (RuntimeError, AttributeError) as e:
+                # Unexpected error during wiring - container internals issue.
+                logger.warning(
+                    "DEGRADED_MODE: Container wiring failed with unexpected error, "
+                    "continuing in degraded mode (correlation_id=%s): %s",
+                    correlation_id,
+                    e,
+                    extra={
+                        "error_type": type(e).__name__,
+                        "correlation_id": correlation_id,
+                        "degraded_mode": True,
+                        "degraded_reason": "wiring_error",
+                        "component": "container_wiring",
+                    },
+                )
+                wire_summary = {"services": [], "status": "degraded"}
+        container_duration = time.time() - container_start_time
+        logger.debug(
+            "Container wired in %.3fs (correlation_id=%s)",
+            container_duration,
+            correlation_id,
+            extra={
+                "duration_seconds": container_duration,
+                "services": wire_summary["services"],
+            },
+        )
+
+        # 4.5. Create PostgreSQL pool for projections
+        # Only create if POSTGRES_HOST is set (indicates registration should be enabled)
+        projector: ProjectorShell | None = None
+        introspection_dispatcher: DispatcherNodeIntrospected | None = None
+        consul_handler = None  # Will be initialized if Consul is configured
+
+        postgres_host = os.getenv("POSTGRES_HOST")
+        if postgres_host:
+            postgres_pool_start_time = time.time()
+            try:
+                postgres_pool = await asyncpg.create_pool(
+                    user=os.getenv("POSTGRES_USER", "postgres"),
+                    password=os.getenv("POSTGRES_PASSWORD", ""),
+                    host=postgres_host,
+                    port=int(os.getenv("POSTGRES_PORT", "5432")),
+                    database=os.getenv("POSTGRES_DATABASE", "omninode_bridge"),
+                    min_size=2,
+                    max_size=10,
+                )
+                postgres_pool_duration = time.time() - postgres_pool_start_time
+                logger.info(
+                    "PostgreSQL pool created in %.3fs (correlation_id=%s)",
+                    postgres_pool_duration,
+                    correlation_id,
+                    extra={
+                        "host": postgres_host,
+                        "port": os.getenv("POSTGRES_PORT", "5432"),
+                        "database": os.getenv("POSTGRES_DATABASE", "omninode_bridge"),
+                    },
+                )
+
+                # 4.6. Load projectors from contracts via ProjectorPluginLoader (OMN-1170/1169)
+                #
+                # This section implements fully contract-driven projector management. The
+                # loader discovers projector contracts from the package's projectors/contracts
+                # directory and creates ProjectorShell instances for runtime use.
+                #
+                # Contract-driven approach:
+                # - Projector behavior defined in YAML contracts (registration_projector.yaml)
+                # - ProjectorPluginLoader discovers and loads projectors from contracts
+                # - ProjectorShell provides generic projection operations (project, partial_update)
+                # - Schema initialization is decoupled - SQL executed directly from schema file
+                #
+                # The registration projector is identified by projector_id="registration-projector"
+                # and is passed to wire_registration_handlers() for handler injection.
+                #
+                projector_contracts_dir = (
+                    Path(__file__).parent.parent / "projectors" / "contracts"
+                )
+
+                # Try to discover projectors from contracts
+                projector_loader = ProjectorPluginLoader(
+                    config=ModelProjectorPluginLoaderConfig(graceful_mode=True),
+                    container=container,
+                    pool=postgres_pool,
+                )
+
+                discovered_projectors: list[ProtocolEventProjector] = []
+                if projector_contracts_dir.exists():
+                    try:
+                        discovered_projectors = (
+                            await projector_loader.load_from_directory(
+                                projector_contracts_dir
+                            )
+                        )
+                        if discovered_projectors:
+                            logger.info(
+                                "Discovered %d projector(s) from contracts (correlation_id=%s)",
+                                len(discovered_projectors),
+                                correlation_id,
+                                extra={
+                                    "discovered_count": len(discovered_projectors),
+                                    "contracts_dir": str(projector_contracts_dir),
+                                    "projector_ids": [
+                                        getattr(p, "projector_id", "unknown")
+                                        for p in discovered_projectors
+                                    ],
+                                },
+                            )
+                        else:
+                            logger.warning(
+                                "No projector contracts found in %s (correlation_id=%s)",
+                                projector_contracts_dir,
+                                correlation_id,
+                                extra={
+                                    "contracts_dir": str(projector_contracts_dir),
+                                },
+                            )
+                    except Exception as discovery_error:
+                        # Log warning but continue - projector discovery is best-effort
+                        # Registration features will be unavailable if discovery fails
+                        logger.warning(
+                            "Projector contract discovery failed: %s (correlation_id=%s)",
+                            sanitize_error_message(discovery_error),
+                            correlation_id,
+                            extra={
+                                "error_type": type(discovery_error).__name__,
+                                "contracts_dir": str(projector_contracts_dir),
+                            },
+                        )
+                else:
+                    logger.debug(
+                        "Projector contracts directory not found, skipping discovery "
+                        "(correlation_id=%s)",
+                        correlation_id,
+                        extra={
+                            "contracts_dir": str(projector_contracts_dir),
+                        },
+                    )
+
+                # Extract registration projector from discovered projectors (OMN-1169)
+                # This replaces the legacy ProjectorRegistration with contract-loaded ProjectorShell
+                registration_projector_id = "registration-projector"
+                for discovered in discovered_projectors:
+                    if (
+                        getattr(discovered, "projector_id", None)
+                        == registration_projector_id
+                    ):
+                        # Cast to ProjectorShell (loader creates ProjectorShell when pool provided)
+                        if isinstance(discovered, ProjectorShell):
+                            projector = discovered
+                            logger.info(
+                                "Using contract-loaded ProjectorShell for registration "
+                                "(correlation_id=%s)",
+                                correlation_id,
+                                extra={
+                                    "projector_id": registration_projector_id,
+                                    "aggregate_type": projector.aggregate_type,
+                                },
+                            )
+                        break
+
+                if projector is None:
+                    # Fallback: No registration projector discovered from contracts
+                    logger.warning(
+                        "Registration projector not found in contracts, "
+                        "registration features will be unavailable (correlation_id=%s)",
+                        correlation_id,
+                        extra={
+                            "expected_projector_id": registration_projector_id,
+                            "discovered_count": len(discovered_projectors),
+                        },
+                    )
+
+                # Initialize schema by executing SQL file directly
+                # Schema initialization is decoupled from the projector - it just ensures
+                # the table and indexes exist. The ProjectorShell uses the schema at runtime.
+                schema_file = (
+                    Path(__file__).parent.parent
+                    / "schemas"
+                    / "schema_registration_projection.sql"
+                )
+                if schema_file.exists():
+                    try:
+                        schema_sql = schema_file.read_text()
+                        async with postgres_pool.acquire() as conn:
+                            await conn.execute(schema_sql)
+                        logger.info(
+                            "Registration projection schema initialized (correlation_id=%s)",
+                            correlation_id,
+                        )
+                    except Exception as schema_error:
+                        # Log warning but continue - schema may already exist
+                        logger.warning(
+                            "Schema initialization encountered error: %s (correlation_id=%s)",
+                            sanitize_error_message(schema_error),
+                            correlation_id,
+                            extra={
+                                "error_type": type(schema_error).__name__,
+                            },
+                        )
+                else:
+                    logger.warning(
+                        "Schema file not found: %s (correlation_id=%s)",
+                        schema_file,
+                        correlation_id,
+                    )
+
+                # 4.6.5. Initialize HandlerConsul if Consul is configured
+                # CONSUL_HOST determines whether to enable Consul registration
+                consul_host = os.getenv("CONSUL_HOST")
+                if consul_host:
+                    # Validate CONSUL_PORT environment variable
+                    consul_port_str = os.getenv("CONSUL_PORT", "8500")
+                    try:
+                        consul_port = int(consul_port_str)
+                        if not MIN_PORT <= consul_port <= MAX_PORT:
+                            logger.warning(
+                                "CONSUL_PORT %d outside valid range %d-%d, using default 8500 (correlation_id=%s)",
+                                consul_port,
+                                MIN_PORT,
+                                MAX_PORT,
+                                correlation_id,
+                            )
+                            consul_port = 8500
+                    except ValueError:
+                        logger.warning(
+                            "Invalid CONSUL_PORT value '%s', using default 8500 (correlation_id=%s)",
+                            consul_port_str,
+                            correlation_id,
+                        )
+                        consul_port = 8500
+
+                    try:
+                        # Deferred import: Only load HandlerConsul when Consul is configured.
+                        # This avoids loading the consul dependency (and its transitive deps)
+                        # when Consul integration is disabled, reducing startup time.
+                        from omnibase_infra.handlers import HandlerConsul
+
+                        consul_handler = HandlerConsul(container)
+                        await consul_handler.initialize(
+                            {
+                                "host": consul_host,
+                                "port": consul_port,
+                            }
+                        )
+                        logger.info(
+                            "HandlerConsul initialized for dual registration (correlation_id=%s)",
+                            correlation_id,
+                            extra={
+                                "consul_host": consul_host,
+                                "consul_port": consul_port,
+                            },
+                        )
+                    except Exception as consul_error:
+                        # Log warning but continue without Consul (PostgreSQL is source of truth)
+                        # Use sanitize_error_message to prevent credential leakage in logs
+                        logger.warning(
+                            "Failed to initialize HandlerConsul, proceeding without Consul: %s (correlation_id=%s)",
+                            sanitize_error_message(consul_error),
+                            correlation_id,
+                            extra={
+                                "error_type": type(consul_error).__name__,
+                            },
+                        )
+                        consul_handler = None
+                else:
+                    logger.debug(
+                        "CONSUL_HOST not set, Consul registration disabled (correlation_id=%s)",
+                        correlation_id,
+                    )
+
+                # 4.7. Wire registration handlers with projector and consul_handler
+                registration_summary = await wire_registration_handlers(
+                    container,
+                    postgres_pool,
+                    projector=projector,
+                    consul_handler=consul_handler,
+                )
+                logger.info(
+                    "Registration handlers wired (correlation_id=%s)",
+                    correlation_id,
+                    extra={
+                        "services": registration_summary["services"],
+                    },
+                )
+
+                # 4.8. Create introspection dispatcher for routing events
+                # Deferred import: HandlerNodeIntrospected depends on PostgreSQL and
+                # registration infrastructure. Only loaded after PostgreSQL pool is
+                # successfully created and registration handlers are wired.
+                from omnibase_infra.nodes.node_registration_orchestrator.handlers import (
+                    HandlerNodeIntrospected,
+                )
+
+                # Check if service_registry is available (may be None in omnibase_core 0.6.x)
+                if container.service_registry is None:
+                    logger.warning(
+                        "DEGRADED_MODE: ServiceRegistry not available, skipping introspection dispatcher creation (correlation_id=%s)",
+                        correlation_id,
+                        extra={
+                            "error_type": "NoneType",
+                            "correlation_id": correlation_id,
+                            "degraded_mode": True,
+                            "degraded_reason": "service_registry_unavailable",
+                            "component": "introspection_dispatcher",
+                        },
+                    )
+                    # Set introspection_dispatcher to None and continue without it
+                    introspection_dispatcher = None
+                else:
+                    logger.debug(
+                        "Resolving HandlerNodeIntrospected from container (correlation_id=%s)",
+                        correlation_id,
+                    )
+                    handler_introspected: HandlerNodeIntrospected = (
+                        await container.service_registry.resolve_service(
+                            HandlerNodeIntrospected
+                        )
+                    )
+                    logger.debug(
+                        "HandlerNodeIntrospected resolved successfully (correlation_id=%s)",
+                        correlation_id,
+                        extra={
+                            "handler_class": handler_introspected.__class__.__name__,
+                        },
+                    )
+
+                    introspection_dispatcher = DispatcherNodeIntrospected(
+                        handler_introspected
+                    )
+                    logger.info(
+                        "Introspection dispatcher created and wired (correlation_id=%s)",
+                        correlation_id,
+                        extra={
+                            "dispatcher_class": introspection_dispatcher.__class__.__name__,
+                            "handler_class": handler_introspected.__class__.__name__,
+                        },
+                    )
+
+            except Exception as pool_error:
+                # Log warning but continue without registration support
+                # Use sanitize_error_message to prevent credential leakage in logs
+                # (PostgreSQL connection errors may include DSN with password)
+                logger.warning(
+                    "Failed to initialize PostgreSQL pool for registration: %s (correlation_id=%s)",
+                    sanitize_error_message(pool_error),
+                    correlation_id,
+                    extra={
+                        "error_type": type(pool_error).__name__,
+                    },
+                )
+                if postgres_pool is not None:
+                    try:
+                        await postgres_pool.close()
+                    except Exception as cleanup_error:
+                        # Sanitize cleanup errors to prevent credential leakage
+                        # NOTE: Do NOT use exc_info=True here - tracebacks may contain
+                        # connection strings with credentials from PostgreSQL errors
+                        logger.warning(
+                            "Cleanup failed for PostgreSQL pool close: %s (correlation_id=%s)",
+                            sanitize_error_message(cleanup_error),
+                            correlation_id,
+                        )
+                    postgres_pool = None
+                projector = None
+                introspection_dispatcher = None
+        else:
+            logger.debug(
+                "POSTGRES_HOST not set, skipping registration handler wiring (correlation_id=%s)",
+                correlation_id,
+            )
+
+        # 5. Resolve RegistryProtocolBinding from container or create new instance
+        # NOTE: Fallback to creating new instance is intentional degraded mode behavior.
+        # The handler registry is optional for basic runtime operation - core event
+        # processing continues even without explicit handler bindings. However,
+        # ProtocolConfigurationError should NOT be masked as it indicates invalid
+        # configuration that would cause undefined behavior.
+        handler_registry: RegistryProtocolBinding | None = None
+
+        # Check if service_registry is available (may be None in omnibase_core 0.6.x)
+        if container.service_registry is not None:
+            try:
+                handler_registry = await container.service_registry.resolve_service(
+                    RegistryProtocolBinding
+                )
+            except ServiceResolutionError as e:
+                # Service not registered - expected in minimal configurations.
+                # Create a new instance directly as fallback.
+                logger.warning(
+                    "DEGRADED_MODE: RegistryProtocolBinding not registered in container, "
+                    "creating new instance (correlation_id=%s): %s",
+                    correlation_id,
+                    e,
+                    extra={
+                        "error_type": type(e).__name__,
+                        "correlation_id": correlation_id,
+                        "degraded_mode": True,
+                        "degraded_reason": "service_not_registered",
+                        "component": "handler_registry",
+                    },
+                )
+                handler_registry = RegistryProtocolBinding()
+            except (RuntimeError, AttributeError) as e:
+                # Unexpected resolution failure - container internals issue.
+                # Log with more diagnostic context but still allow degraded operation.
+                logger.warning(
+                    "DEGRADED_MODE: Unexpected error resolving RegistryProtocolBinding, "
+                    "creating new instance (correlation_id=%s): %s",
+                    correlation_id,
+                    e,
+                    extra={
+                        "error_type": type(e).__name__,
+                        "correlation_id": correlation_id,
+                        "degraded_mode": True,
+                        "degraded_reason": "resolution_error",
+                        "component": "handler_registry",
+                    },
+                )
+                handler_registry = RegistryProtocolBinding()
+            # NOTE: ProtocolConfigurationError is NOT caught here - configuration
+            # errors should propagate and stop startup to prevent undefined behavior.
+        else:
+            # ServiceRegistry not available, create a new RegistryProtocolBinding directly
+            logger.warning(
+                "DEGRADED_MODE: ServiceRegistry not available, creating RegistryProtocolBinding directly (correlation_id=%s)",
+                correlation_id,
+                extra={
+                    "error_type": "NoneType",
+                    "correlation_id": correlation_id,
+                    "degraded_mode": True,
+                    "degraded_reason": "service_registry_unavailable",
+                    "component": "handler_registry",
+                },
+            )
+            handler_registry = RegistryProtocolBinding()
+
+        # 6. Create runtime host process with config and pre-resolved registry
+        # RuntimeHostProcess accepts config as dict; cast model_dump() result to
+        # dict[str, object] to avoid implicit Any typing (Pydantic's model_dump()
+        # returns dict[str, Any] but all our model fields are strongly typed)
+        #
+        # NOTE: RuntimeHostProcess expects 'service_name' and 'node_name' keys,
+        # but ModelRuntimeConfig uses 'name'. Map 'name' -> 'service_name'/'node_name'
+        # for compatibility. (OMN-1602)
+        #
+        # INVARIANT: In the current runtime model, `ModelRuntimeConfig.name` represents
+        # both `service_name` and `node_name` by design; multi-node services require
+        # schema expansion.
+        #
+        # TRIGGER FOR SPLIT: Split when ServiceKernel supports registering multiple
+        # node contracts under one service runtime.
+        #
+        # Why both fields get the same value:
+        # - For services using simplified config with just 'name', there's no semantic
+        #   distinction between service and node - a single service hosts a single node
+        # - RuntimeHostProcess uses these to construct ModelNodeIdentity for Kafka
+        #   consumer group IDs and event routing
+        # - The introspection consumer group format is:
+        #   {env}.{service_name}.{node_name}.{purpose}.{version}
+        #   e.g., "local.my-service.my-service.introspection.v1"
+        # - When service_name == node_name, the format is intentionally redundant but
+        #   maintains consistency with multi-node deployments where they would differ
+        runtime_create_start_time = time.time()
+        runtime_config_dict = cast("dict[str, object]", config.model_dump())
+        if config.name:
+            runtime_config_dict["service_name"] = config.name
+            runtime_config_dict["node_name"] = config.name
+        runtime = RuntimeHostProcess(
+            container=container,
+            event_bus=event_bus,
+            input_topic=config.input_topic,
+            output_topic=config.output_topic,
+            config=runtime_config_dict,
+            handler_registry=handler_registry,
+            # Pass contracts directory for handler discovery (OMN-1317)
+            # This enables contract-based handler registration instead of
+            # falling back to wire_handlers() with an empty registry
+            contract_paths=[str(contracts_dir)],
+        )
+        runtime_create_duration = time.time() - runtime_create_start_time
+        logger.debug(
+            "Runtime host process created in %.3fs (correlation_id=%s)",
+            runtime_create_duration,
+            correlation_id,
+            extra={
+                "duration_seconds": runtime_create_duration,
+                "input_topic": config.input_topic,
+                "output_topic": config.output_topic,
+            },
+        )
+
+        # 7. Setup graceful shutdown
+        shutdown_event = asyncio.Event()
+        loop = asyncio.get_running_loop()
+
+        def handle_shutdown(sig: signal.Signals) -> None:
+            """Handle shutdown signal with correlation tracking."""
+            logger.info(
+                "Received %s, initiating graceful shutdown... (correlation_id=%s)",
+                sig.name,
+                correlation_id,
+            )
+            shutdown_event.set()
+
+        # Register signal handlers for graceful shutdown
+        if sys.platform != "win32":
+            # Unix: Use asyncio's signal handler for proper event loop integration
+            for sig in (signal.SIGINT, signal.SIGTERM):
+                loop.add_signal_handler(sig, handle_shutdown, sig)
+        else:
+            # Windows: asyncio signal handlers not supported, use signal.signal()
+            # for SIGINT (Ctrl+C). Note: SIGTERM not available on Windows.
+            #
+            # Thread-safety: On Windows, signal.signal() handlers execute in a
+            # different thread than the event loop. While asyncio.Event.set() is
+            # documented as thread-safe, we use loop.call_soon_threadsafe() to
+            # schedule the set() call on the event loop thread. This ensures
+            # proper cross-thread communication and avoids potential race
+            # conditions with any event loop state inspection.
+            def windows_handler(signum: int, frame: object) -> None:
+                """Windows-compatible signal handler wrapper.
+
+                Uses call_soon_threadsafe to safely communicate with the event
+                loop from the signal handler thread.
+                """
+                sig = signal.Signals(signum)
+                logger.info(
+                    "Received %s, initiating graceful shutdown... (correlation_id=%s)",
+                    sig.name,
+                    correlation_id,
+                )
+                loop.call_soon_threadsafe(shutdown_event.set)
+
+            signal.signal(signal.SIGINT, windows_handler)
+
+        # 8. Start runtime and health server
+        runtime_start_time = time.time()
+        logger.info(
+            "Starting ONEX runtime... (correlation_id=%s)",
+            correlation_id,
+        )
+        await runtime.start()
+        runtime_start_duration = time.time() - runtime_start_time
+        logger.debug(
+            "Runtime started in %.3fs (correlation_id=%s)",
+            runtime_start_duration,
+            correlation_id,
+            extra={
+                "duration_seconds": runtime_start_duration,
+            },
+        )
+
+        # 9. Start HTTP health server for Docker/K8s probes
+        # Port can be configured via ONEX_HTTP_PORT environment variable
+        http_port_str = os.getenv("ONEX_HTTP_PORT", str(DEFAULT_HTTP_PORT))
+        try:
+            http_port = int(http_port_str)
+            if not MIN_PORT <= http_port <= MAX_PORT:
+                logger.warning(
+                    "ONEX_HTTP_PORT %d outside valid range %d-%d, using default %d (correlation_id=%s)",
+                    http_port,
+                    MIN_PORT,
+                    MAX_PORT,
+                    DEFAULT_HTTP_PORT,
+                    correlation_id,
+                )
+                http_port = DEFAULT_HTTP_PORT
+        except ValueError:
+            logger.warning(
+                "Invalid ONEX_HTTP_PORT value '%s', using default %d (correlation_id=%s)",
+                http_port_str,
+                DEFAULT_HTTP_PORT,
+                correlation_id,
+            )
+            http_port = DEFAULT_HTTP_PORT
+
+        health_server = ServiceHealth(
+            container=container,
+            runtime=runtime,
+            port=http_port,
+            version=KERNEL_VERSION,
+        )
+        health_start_time = time.time()
+        await health_server.start()
+        health_start_duration = time.time() - health_start_time
+        logger.debug(
+            "Health server started in %.3fs (correlation_id=%s)",
+            health_start_duration,
+            correlation_id,
+            extra={
+                "duration_seconds": health_start_duration,
+                "port": http_port,
+            },
+        )
+
+        # 9.5. Start introspection event consumer if dispatcher is available
+        # This consumer subscribes to the input topic and routes introspection
+        # events to the HandlerNodeIntrospected via DispatcherNodeIntrospected.
+        # Unlike RuntimeHostProcess which routes based on handler_type field,
+        # this consumer directly parses introspection events from JSON.
+        #
+        # The message handler is extracted to IntrospectionMessageHandler for
+        # better testability and separation of concerns (PR #101 code quality).
+        #
+        # Duck typing approach per CLAUDE.md architectural guidelines:
+        # Check for subscribe() capability via hasattr/callable instead of isinstance.
+        # This enables any event bus implementing subscribe() to participate in
+        # introspection event consumption, following protocol-based polymorphism.
+        #
+        # Production considerations:
+        # - EventBusKafka: Uses distributed consumer groups for production workloads
+        # - EventBusInmemory: subscribe() works for testing scenarios
+        # - Other implementations: Will work if they implement subscribe()
+        #
+        # The duck typing approach allows new event bus implementations to
+        # participate in introspection without modifying this code.
+        has_subscribe = hasattr(event_bus, "subscribe") and callable(
+            getattr(event_bus, "subscribe", None)
+        )
+        if introspection_dispatcher is not None and has_subscribe:
+            # Create extracted event router with container-based DI pattern
+            # Dependencies are passed explicitly since they are created at runtime
+            # by the kernel and may not be registered in the container yet
+            introspection_event_router = IntrospectionEventRouter(
+                container=container,
+                output_topic=config.output_topic,
+                dispatcher=introspection_dispatcher,
+                event_bus=event_bus,
+            )
+
+            # Create typed node identity for introspection subscription (OMN-1602)
+            # Uses ModelNodeIdentity + EnumConsumerGroupPurpose instead of hardcoded
+            # group_id suffix hack for proper semantic consumer group naming.
+            #
+            # Required fields from config (fail-fast if missing):
+            # - service_name: from config.name (required for node identification)
+            # - node_name: from config.name (required for node identification)
+            # Optional with defaults:
+            # - env: from environment variable or event_bus.environment
+            # - version: from config.contract_version or "v1"
+            if not config.name:
+                context = ModelInfraErrorContext(
+                    transport_type=EnumInfraTransportType.RUNTIME,
+                    operation="create_node_identity",
+                    correlation_id=correlation_id,
+                )
+                raise ProtocolConfigurationError(
+                    "Runtime config requires 'name' field for service identification. "
+                    "Add 'name: your-service-name' to runtime_config.yaml. "
+                    "This is required for typed introspection subscription (OMN-1602).",
+                    context=context,
+                    parameter="name",
+                )
+
+            introspection_node_identity = ModelNodeIdentity(
+                env=environment,
+                service=config.name,
+                node_name=config.name,
+                version=config.contract_version or "v1",
+            )
+
+            # Subscribe with callback - returns unsubscribe function
+            subscribe_start_time = time.time()
+            logger.info(
+                "Subscribing to introspection events on event bus (correlation_id=%s)",
+                correlation_id,
+                extra={
+                    "topic": config.input_topic,
+                    "node_identity": {
+                        "env": introspection_node_identity.env,
+                        "service": introspection_node_identity.service,
+                        "node_name": introspection_node_identity.node_name,
+                        "version": introspection_node_identity.version,
+                    },
+                    "purpose": EnumConsumerGroupPurpose.INTROSPECTION.value,
+                    "event_bus_type": event_bus_type,
+                },
+            )
+
+            introspection_unsubscribe = await event_bus.subscribe(
+                topic=config.input_topic,
+                node_identity=introspection_node_identity,
+                on_message=introspection_event_router.handle_message,
+                purpose=EnumConsumerGroupPurpose.INTROSPECTION,
+            )
+            subscribe_duration = time.time() - subscribe_start_time
+
+            logger.info(
+                "Introspection event consumer started successfully in %.3fs (correlation_id=%s)",
+                subscribe_duration,
+                correlation_id,
+                extra={
+                    "topic": config.input_topic,
+                    "node_identity": {
+                        "env": introspection_node_identity.env,
+                        "service": introspection_node_identity.service,
+                        "node_name": introspection_node_identity.node_name,
+                        "version": introspection_node_identity.version,
+                    },
+                    "purpose": EnumConsumerGroupPurpose.INTROSPECTION.value,
+                    "subscribe_duration_seconds": subscribe_duration,
+                    "event_bus_type": event_bus_type,
+                },
+            )
+
+        # Calculate total bootstrap time
+        bootstrap_duration = time.time() - bootstrap_start_time
+
+        # Display startup banner with key configuration
+        if introspection_dispatcher is not None:
+            if consul_handler is not None:
+                registration_status = "enabled (PostgreSQL + Consul)"
+            else:
+                registration_status = "enabled (PostgreSQL only)"
+        else:
+            registration_status = "disabled"
+        banner_lines = [
+            "=" * 60,
+            f"ONEX Runtime Kernel v{KERNEL_VERSION}",
+            f"Environment: {environment}",
+            f"Contracts: {contracts_dir}",
+            f"Event Bus: {event_bus_type} (group: {config.consumer_group})",
+            f"Topics: {config.input_topic} → {config.output_topic}",
+            f"Registration: {registration_status}",
+            f"Health endpoint: http://0.0.0.0:{http_port}/health",
+            f"Bootstrap time: {bootstrap_duration:.3f}s",
+            f"Correlation ID: {correlation_id}",
+            "=" * 60,
+        ]
+        banner = "\n".join(banner_lines)
+        logger.info("\n%s", banner)
+
+        logger.info(
+            "ONEX runtime started successfully in %.3fs (correlation_id=%s)",
+            bootstrap_duration,
+            correlation_id,
+            extra={
+                "bootstrap_duration_seconds": bootstrap_duration,
+                "config_load_seconds": config_duration,
+                "event_bus_create_seconds": event_bus_duration,
+                "container_wire_seconds": container_duration,
+                "runtime_create_seconds": runtime_create_duration,
+                "runtime_start_seconds": runtime_start_duration,
+                "health_start_seconds": health_start_duration,
+            },
+        )
+
+        # Wait for shutdown signal
+        await shutdown_event.wait()
+
+        grace_period = config.shutdown.grace_period_seconds
+        shutdown_start_time = time.time()
+        logger.info(
+            "Shutdown signal received, stopping runtime (timeout=%ss, correlation_id=%s)",
+            grace_period,
+            correlation_id,
+        )
+
+        # Stop introspection consumer first (fast)
+        if introspection_unsubscribe is not None:
+            try:
+                await introspection_unsubscribe()
+                logger.debug(
+                    "Introspection consumer stopped (correlation_id=%s)",
+                    correlation_id,
+                )
+            except Exception as consumer_stop_error:
+                logger.warning(
+                    "Failed to stop introspection consumer: %s (correlation_id=%s)",
+                    sanitize_error_message(consumer_stop_error),
+                    correlation_id,
+                )
+            introspection_unsubscribe = None
+
+        # Stop health server (fast, non-blocking)
+        if health_server is not None:
+            try:
+                health_stop_start_time = time.time()
+                await health_server.stop()
+                health_stop_duration = time.time() - health_stop_start_time
+                logger.debug(
+                    "Health server stopped in %.3fs (correlation_id=%s)",
+                    health_stop_duration,
+                    correlation_id,
+                    extra={
+                        "duration_seconds": health_stop_duration,
+                    },
+                )
+            except Exception as health_stop_error:
+                logger.warning(
+                    "Failed to stop health server: %s (correlation_id=%s)",
+                    health_stop_error,
+                    correlation_id,
+                    extra={
+                        "error_type": type(health_stop_error).__name__,
+                    },
+                )
+            health_server = None
+
+        # Stop runtime with timeout
+        try:
+            runtime_stop_start_time = time.time()
+            await asyncio.wait_for(runtime.stop(), timeout=grace_period)
+            runtime_stop_duration = time.time() - runtime_stop_start_time
+            logger.debug(
+                "Runtime stopped in %.3fs (correlation_id=%s)",
+                runtime_stop_duration,
+                correlation_id,
+                extra={
+                    "duration_seconds": runtime_stop_duration,
+                },
+            )
+        except TimeoutError:
+            logger.warning(
+                "Graceful shutdown timed out after %s seconds, forcing stop (correlation_id=%s)",
+                grace_period,
+                correlation_id,
+            )
+        runtime = None  # Mark as stopped to prevent double-stop in finally
+
+        # Close PostgreSQL pool
+        if postgres_pool is not None:
+            try:
+                pool_close_start_time = time.time()
+                await postgres_pool.close()
+                pool_close_duration = time.time() - pool_close_start_time
+                logger.debug(
+                    "PostgreSQL pool closed in %.3fs (correlation_id=%s)",
+                    pool_close_duration,
+                    correlation_id,
+                )
+            except Exception as pool_close_error:
+                # Sanitize to prevent credential leakage
+                logger.warning(
+                    "Failed to close PostgreSQL pool: %s (correlation_id=%s)",
+                    sanitize_error_message(pool_close_error),
+                    correlation_id,
+                )
+            postgres_pool = None
+
+        shutdown_duration = time.time() - shutdown_start_time
+        logger.info(
+            "ONEX runtime stopped successfully in %.3fs (correlation_id=%s)",
+            shutdown_duration,
+            correlation_id,
+            extra={
+                "shutdown_duration_seconds": shutdown_duration,
+            },
+        )
+        return 0
+
+    except ProtocolConfigurationError as e:
+        # Configuration errors already have proper context and chaining
+        error_code = getattr(getattr(e, "model", None), "error_code", None)
+        error_code_name = getattr(error_code, "name", None)
+        logger.exception(
+            "ONEX runtime configuration failed (correlation_id=%s)",
+            correlation_id,
+            extra={
+                "error_type": type(e).__name__,
+                "error_code": str(error_code_name)
+                if error_code_name is not None
+                else None,
+            },
+        )
+        return 1
+
+    except RuntimeHostError as e:
+        # Runtime host errors already have proper structure
+        error_code = getattr(getattr(e, "model", None), "error_code", None)
+        error_code_name = getattr(error_code, "name", None)
+        logger.exception(
+            "ONEX runtime host error (correlation_id=%s)",
+            correlation_id,
+            extra={
+                "error_type": type(e).__name__,
+                "error_code": str(error_code_name)
+                if error_code_name is not None
+                else None,
+            },
+        )
+        return 1
+
+    except Exception as e:
+        # Unexpected errors: log with full context and return error code
+        # (consistent with ProtocolConfigurationError and RuntimeHostError handlers)
+        # Sanitize error message to prevent credential leakage
+        logger.exception(
+            "ONEX runtime failed with unexpected error: %s (correlation_id=%s)",
+            sanitize_error_message(e),
+            correlation_id,
+            extra={
+                "error_type": type(e).__name__,
+            },
+        )
+        return 1
+
+    finally:
+        # Guard cleanup - stop all resources if not already stopped
+        # Order: introspection consumer -> health server -> runtime -> pool
+
+        if introspection_unsubscribe is not None:
+            try:
+                await introspection_unsubscribe()
+            except Exception as cleanup_error:
+                logger.warning(
+                    "Failed to stop introspection consumer during cleanup: %s (correlation_id=%s)",
+                    sanitize_error_message(cleanup_error),
+                    correlation_id,
+                )
+
+        if health_server is not None:
+            try:
+                await health_server.stop()
+            except Exception as cleanup_error:
+                logger.warning(
+                    "Failed to stop health server during cleanup: %s (correlation_id=%s)",
+                    sanitize_error_message(cleanup_error),
+                    correlation_id,
+                )
+
+        if runtime is not None:
+            try:
+                await runtime.stop()
+            except Exception as cleanup_error:
+                # Log cleanup failures with context instead of suppressing them
+                # Sanitize to prevent potential credential leakage from runtime errors
+                logger.warning(
+                    "Failed to stop runtime during cleanup: %s (correlation_id=%s)",
+                    sanitize_error_message(cleanup_error),
+                    correlation_id,
+                )
+
+        if postgres_pool is not None:
+            try:
+                await postgres_pool.close()
+            except Exception as cleanup_error:
+                # Sanitize to prevent credential leakage from PostgreSQL errors
+                logger.warning(
+                    "Failed to close PostgreSQL pool during cleanup: %s (correlation_id=%s)",
+                    sanitize_error_message(cleanup_error),
+                    correlation_id,
+                )
+
+
+def configure_logging() -> None:
+    """Configure logging for the kernel with structured format.
+
+    Sets up structured logging with appropriate log level from the
+    ONEX_LOG_LEVEL environment variable (default: INFO). This function
+    must be called early in the bootstrap process to ensure logging
+    is available for all subsequent operations.
+
+    Logging Configuration:
+        - Log Level: Controlled by ONEX_LOG_LEVEL environment variable
+        - Format: Timestamp, level, logger name, message, extras
+        - Date Format: ISO-8601 compatible (YYYY-MM-DD HH:MM:SS)
+        - Structured Extras: Support for correlation_id and custom fields
+
+    Bootstrap Order Rationale:
+        This function is called BEFORE runtime config is loaded because logging
+        must be available during config loading itself (to log errors, warnings,
+        and info about config discovery). Therefore, logging configuration uses
+        environment variables rather than contract-based config values.
+
+        This is a deliberate chicken-and-egg solution:
+        - Environment variables control early bootstrap logging
+        - Contract config controls runtime behavior after bootstrap
+
+    Environment Variables:
+        ONEX_LOG_LEVEL: Logging level (DEBUG, INFO, WARNING, ERROR, CRITICAL)
+            Default: INFO
+
+    Log Format Example:
+        2025-01-15 10:30:45 [INFO] omnibase_infra.runtime.service_kernel: ONEX Kernel v0.1.0
+        2025-01-15 10:30:45 [DEBUG] omnibase_infra.runtime.service_kernel: Runtime config loaded
+            (correlation_id=123e4567-e89b-12d3-a456-426614174000)
+
+    Structured Logging Extras:
+        All log calls support structured extras for observability:
+        - correlation_id: UUID for distributed tracing
+        - duration_seconds: Operation timing metrics
+        - error_type: Exception class name for error analysis
+        - Custom fields: Any JSON-serializable data
+
+    Example:
+        >>> configure_logging()
+        >>> logger.info("Operation completed", extra={"duration_seconds": 1.234})
+    """
+    log_level = os.getenv("ONEX_LOG_LEVEL", "INFO").upper()
+
+    # Validate log level and provide helpful error if invalid
+    valid_levels = {"DEBUG", "INFO", "WARNING", "ERROR", "CRITICAL"}
+    if log_level not in valid_levels:
+        print(
+            f"Warning: Invalid ONEX_LOG_LEVEL '{log_level}', using INFO. "
+            f"Valid levels: {', '.join(sorted(valid_levels))}",
+            file=sys.stderr,
+        )
+        log_level = "INFO"
+
+    logging.basicConfig(
+        level=getattr(logging, log_level, logging.INFO),
+        format="%(asctime)s [%(levelname)s] %(name)s: %(message)s",
+        datefmt="%Y-%m-%d %H:%M:%S",
+    )
+
+
+def main() -> None:
+    """Entry point for the ONEX runtime kernel.
+
+    This is the synchronous entry point for the kernel. It configures
+    logging, initiates the async bootstrap process, and handles the
+    final exit code.
+
+    Execution Flow:
+        1. Configure logging from environment variables
+        2. Log kernel version for startup identification
+        3. Run async bootstrap function in event loop
+        4. Exit with appropriate exit code (0=success, 1=error)
+
+    Exit Codes:
+        0: Successful startup and clean shutdown
+        1: Configuration error, runtime error, or unexpected failure
+
+    This function is the target for:
+        - The installed entrypoint: `onex-runtime`
+        - Direct module execution: `python -m omnibase_infra.runtime.service_kernel`
+        - Docker CMD/ENTRYPOINT in container deployments
+
+    Example:
+        >>> # From command line
+        >>> python -m omnibase_infra.runtime.service_kernel
+        >>> # Or via installed entrypoint
+        >>> onex-runtime
+
+    Docker Usage:
+        CMD ["onex-runtime"]
+        # Container will start runtime and expose health endpoint
+    """
+    configure_logging()
+    logger.info("ONEX Kernel v%s initializing...", KERNEL_VERSION)
+    exit_code = asyncio.run(bootstrap())
+    sys.exit(exit_code)
+
+
+if __name__ == "__main__":
+    main()
+
+
+__all__: list[str] = [
+    "ENV_CONTRACTS_DIR",
+    "bootstrap",
+    "load_runtime_config",
+    "main",
+]