omnibase_infra 0.2.5__py3-none-any.whl → 0.2.7__py3-none-any.whl

omnibase_infra/runtime/emit_daemon/daemon.py

@@ -0,0 +1,812 @@
+# SPDX-License-Identifier: MIT
+# Copyright (c) 2025 OmniNode Team
+"""Hook Event Daemon - Unix socket server for persistent Kafka event emission.
+
+This module provides the EmitDaemon class that implements a Unix socket server
+for receiving events from Claude Code hooks and publishing them to Kafka with
+fire-and-forget semantics from the caller's perspective.
+
+Architecture:
+    ```
+    +-----------------+     Unix Socket     +-------------+     Kafka     +-------+
+    | Claude Code     | -----------------> | EmitDaemon  | ------------> | Kafka |
+    | Hooks           |   JSON messages    | (this file) |   Events     | Topics|
+    +-----------------+                    +-------------+               +-------+
+                                                 |
+                                                 v
+                                           +------------+
+                                           | Disk Spool |
+                                           | (overflow) |
+                                           +------------+
+    ```
+
+Features:
+    - Unix domain socket server for low-latency local IPC
+    - Bounded in-memory queue with disk spool overflow
+    - Persistent Kafka connection with retry logic
+    - Fire-and-forget semantics for callers
+    - Graceful shutdown with queue drain
+    - PID file management for process tracking
+    - Health check endpoint for monitoring
+
+Protocol:
+    Request format: {"event_type": "prompt.submitted", "payload": {...}}\\n
+    Response format: {"status": "queued"}\\n or {"status": "error", "reason": "..."}\\n
+
+    Special commands:
+    - {"command": "ping"}\\n -> {"status": "ok", "queue_size": N, "spool_size": M}\\n
+
+Related Tickets:
+    - OMN-1610: Hook Event Daemon MVP
+
+.. versionadded:: 0.2.6
+"""
+
+from __future__ import annotations
+
+import asyncio
+import json
+import logging
+import os
+import signal
+from datetime import UTC, datetime
+from pathlib import Path
+from typing import cast
+from uuid import UUID, uuid4
+
+from pydantic import ValidationError
+
+from omnibase_core.errors import OnexError
+from omnibase_infra.event_bus.event_bus_kafka import EventBusKafka
+from omnibase_infra.event_bus.models import ModelEventHeaders
+from omnibase_infra.event_bus.models.config import ModelKafkaEventBusConfig
+from omnibase_infra.protocols import ProtocolEventBusLike
+from omnibase_infra.runtime.emit_daemon.config import ModelEmitDaemonConfig
+from omnibase_infra.runtime.emit_daemon.event_registry import EventRegistry
+from omnibase_infra.runtime.emit_daemon.model_daemon_request import (
+    ModelDaemonEmitRequest,
+    ModelDaemonPingRequest,
+    parse_daemon_request,
+)
+from omnibase_infra.runtime.emit_daemon.model_daemon_response import (
+    ModelDaemonErrorResponse,
+    ModelDaemonPingResponse,
+    ModelDaemonQueuedResponse,
+)
+from omnibase_infra.runtime.emit_daemon.queue import BoundedEventQueue, ModelQueuedEvent
+
+logger = logging.getLogger(__name__)
+
+# Poll interval for publisher loop when queue is empty (seconds)
+PUBLISHER_POLL_INTERVAL_SECONDS: float = 0.1
+
+
+class EmitDaemon:
+    """Unix socket daemon for persistent Kafka event emission.
+
+    Accepts events via Unix socket, queues them, and publishes to Kafka
+    with fire-and-forget semantics from the caller's perspective.
+
+    The daemon operates as follows:
+        1. Listens on a Unix domain socket for incoming events
+        2. Validates event payloads (type, size, required fields)
+        3. Queues events in a bounded in-memory queue
+        4. Background publisher loop dequeues and publishes to Kafka
+        5. On publish failure, events are re-queued with exponential backoff
+        6. On graceful shutdown, queue is drained to disk spool
+
+    Attributes:
+        config: Daemon configuration model
+        queue: Bounded event queue with disk spool
+
+    Example:
+        ```python
+        from omnibase_infra.runtime.emit_daemon import EmitDaemon, ModelEmitDaemonConfig
+
+        config = ModelEmitDaemonConfig(
+            kafka_bootstrap_servers="kafka:9092",
+            socket_path=Path("/tmp/emit.sock"),
+        )
+
+        daemon = EmitDaemon(config)
+        await daemon.start()
+
+        # Daemon runs until SIGTERM or SIGINT
+        # Or call daemon.stop() programmatically
+        ```
+    """
+
+    def __init__(
+        self,
+        config: ModelEmitDaemonConfig,
+        event_bus: ProtocolEventBusLike | None = None,
+    ) -> None:
+        """Initialize daemon with config.
+
+        If event_bus is None, creates EventBusKafka from config.
+
+        Args:
+            config: Daemon configuration model containing socket path,
+                Kafka settings, queue limits, and timeout values.
+            event_bus: Optional event bus for testing. If not provided,
+                creates EventBusKafka from config.
+
+        Example:
+            ```python
+            # Production usage
+            config = ModelEmitDaemonConfig(kafka_bootstrap_servers="kafka:9092")
+            daemon = EmitDaemon(config)
+
+            # Testing with mock event bus
+            mock_bus = MockEventBus()
+            daemon = EmitDaemon(config, event_bus=mock_bus)
+            ```
+        """
+        self._config = config
+        self._event_bus: ProtocolEventBusLike | None = event_bus
+
+        # Event registry for topic resolution and payload enrichment
+        self._registry = EventRegistry(environment=config.environment)
+
+        # Bounded event queue with disk spool overflow
+        self._queue = BoundedEventQueue(
+            max_memory_queue=config.max_memory_queue,
+            max_spool_messages=config.max_spool_messages,
+            max_spool_bytes=config.max_spool_bytes,
+            spool_dir=config.spool_dir,
+        )
+
+        # Server state
+        self._server: asyncio.Server | None = None
+        self._publisher_task: asyncio.Task[None] | None = None
+        self._running = False
+        self._shutdown_event = asyncio.Event()
+
+        # Lock for shared state access
+        self._lock = asyncio.Lock()
+
+        logger.debug(
+            "EmitDaemon initialized",
+            extra={
+                "socket_path": str(config.socket_path),
+                "kafka_servers": config.kafka_bootstrap_servers,
+                "max_memory_queue": config.max_memory_queue,
+            },
+        )
+
+    @property
+    def config(self) -> ModelEmitDaemonConfig:
+        """Get the daemon configuration.
+
+        Returns:
+            The daemon configuration model.
+        """
+        return self._config
+
+    @property
+    def queue(self) -> BoundedEventQueue:
+        """Get the event queue.
+
+        Returns:
+            The bounded event queue with disk spool.
+        """
+        return self._queue
+
+    async def start(self) -> None:
+        """Start the daemon.
+
+        Performs the following startup sequence:
+            1. Check for stale socket/PID and clean up
+            2. Create PID file
+            3. Load any spooled events from disk
+            4. Initialize Kafka event bus
+            5. Start Unix socket server
+            6. Start publisher loop (background task)
+            7. Setup signal handlers for graceful shutdown
+
+        Raises:
+            OSError: If socket creation fails
+            RuntimeError: If another daemon is already running
+        """
+        async with self._lock:
+            if self._running:
+                logger.debug("EmitDaemon already running")
+                return
+
+            # Check and clean up stale socket/PID
+            if self._check_stale_socket():
+                self._cleanup_stale()
+            elif self._config.pid_path.exists():
+                # Another daemon is running
+                pid = self._config.pid_path.read_text().strip()
+                raise OnexError(
+                    f"Another emit daemon is already running with PID {pid}"
+                )
+
+            # Create PID file
+            self._write_pid_file()
+
+            # Load any spooled events from previous runs
+            spool_count = await self._queue.load_spool()
+            if spool_count > 0:
+                logger.info(f"Loaded {spool_count} events from spool")
+
+            # Initialize Kafka event bus if not provided
+            if self._event_bus is None:
+                kafka_config = ModelKafkaEventBusConfig(
+                    bootstrap_servers=self._config.kafka_bootstrap_servers,
+                    environment=self._config.environment,
+                    timeout_seconds=int(self._config.kafka_timeout_seconds),
+                )
+                self._event_bus = EventBusKafka(config=kafka_config)
+
+            # Start the event bus (connects to Kafka)
+            # NOTE: hasattr check required because event_bus can be a mock for testing
+            # that may not implement start(). EventBusKafka always has start(), but
+            # test doubles may omit it if they don't need explicit initialization.
+            if hasattr(self._event_bus, "start"):
+                await self._event_bus.start()  # type: ignore[union-attr]
+
+            # Ensure parent directory exists for socket
+            self._config.socket_path.parent.mkdir(parents=True, exist_ok=True)
+
+            # Remove existing socket file if present
+            if self._config.socket_path.exists():
+                self._config.socket_path.unlink()
+
+            # Start Unix socket server
+            self._server = await asyncio.start_unix_server(
+                self._handle_client,
+                path=str(self._config.socket_path),
+            )
+
+            # Set socket permissions (configurable, defaults to owner and group read/write)
+            self._config.socket_path.chmod(self._config.socket_permissions)
+
+            # Start publisher loop as background task
+            self._publisher_task = asyncio.create_task(self._publisher_loop())
+
+            # Setup signal handlers for graceful shutdown
+            loop = asyncio.get_running_loop()
+            for sig in (signal.SIGTERM, signal.SIGINT):
+                loop.add_signal_handler(sig, self._signal_handler)
+
+            self._running = True
+            self._shutdown_event.clear()
+
+            logger.info(
+                "EmitDaemon started",
+                extra={
+                    "socket_path": str(self._config.socket_path),
+                    "pid": os.getpid(),
+                },
+            )
+
+    async def stop(self) -> None:
+        """Stop the daemon gracefully.
+
+        Performs the following shutdown sequence:
+            1. Stop accepting new connections
+            2. Cancel publisher task
+            3. Drain queue to spool (up to shutdown_drain_seconds)
+            4. Close Kafka connection
+            5. Remove socket and PID file
+
+        This method is safe to call multiple times.
+        """
+        async with self._lock:
+            if not self._running:
+                logger.debug("EmitDaemon not running")
+                return
+
+            self._running = False
+            self._shutdown_event.set()
+
+            logger.info("EmitDaemon stopping...")
+
+            # Remove signal handlers
+            loop = asyncio.get_running_loop()
+            for sig in (signal.SIGTERM, signal.SIGINT):
+                loop.remove_signal_handler(sig)
+
+            # Stop accepting new connections
+            if self._server is not None:
+                self._server.close()
+                await self._server.wait_closed()
+                self._server = None
+
+            # Cancel publisher task
+            if self._publisher_task is not None:
+                self._publisher_task.cancel()
+                try:
+                    await self._publisher_task
+                except asyncio.CancelledError:
+                    pass
+                self._publisher_task = None
+
+            # Drain queue to spool with timeout
+            if self._config.shutdown_drain_seconds > 0:
+                try:
+                    async with asyncio.timeout(self._config.shutdown_drain_seconds):
+                        drained = await self._queue.drain_to_spool()
+                        if drained > 0:
+                            logger.info(f"Drained {drained} events to spool")
+                except TimeoutError:
+                    logger.warning(
+                        "Shutdown drain timeout exceeded, some events may be lost"
+                    )
+
+            # Close Kafka connection
+            # NOTE: hasattr check required because event_bus can be a mock for testing
+            # that may not implement close(). EventBusKafka always has close(), but
+            # test doubles may omit it if they don't need explicit cleanup.
+            if self._event_bus is not None and hasattr(self._event_bus, "close"):
+                await self._event_bus.close()  # type: ignore[union-attr]
+
+            # Remove socket file
+            if self._config.socket_path.exists():
+                try:
+                    self._config.socket_path.unlink()
+                except OSError as e:
+                    logger.warning(f"Failed to remove socket file: {e}")
+
+            # Remove PID file
+            self._remove_pid_file()
+
+            logger.info("EmitDaemon stopped")
+
+    async def run_until_shutdown(self) -> None:
+        """Run the daemon until shutdown signal is received.
+
+        Blocks until SIGTERM/SIGINT is received or stop() is called.
+        Useful for running the daemon as a standalone process.
+
+        Example:
+            ```python
+            daemon = EmitDaemon(config)
+            await daemon.start()
+            await daemon.run_until_shutdown()
+            ```
+        """
+        await self._shutdown_event.wait()
+        await self.stop()
+
+    def _signal_handler(self) -> None:
+        """Handle SIGTERM/SIGINT signals.
+
+        Sets the shutdown event to trigger graceful shutdown.
+        """
+        logger.info("Received shutdown signal")
+        self._shutdown_event.set()
+
+    async def _handle_client(
+        self,
+        reader: asyncio.StreamReader,
+        writer: asyncio.StreamWriter,
+    ) -> None:
+        """Handle a single client connection.
+
+        Protocol: newline-delimited JSON
+        Request: {"event_type": "...", "payload": {...}}
+        Response: {"status": "queued"} or {"status": "error", "reason": "..."}
+
+        Special commands:
+        - {"command": "ping"} -> {"status": "ok", "queue_size": N, "spool_size": M}
+
+        Args:
+            reader: Async stream reader for the client connection
+            writer: Async stream writer for the client connection
+        """
+        peer = "unix_client"
+        logger.debug(f"Client connected: {peer}")
+
+        try:
+            while not self._shutdown_event.is_set():
+                try:
+                    # Read line with timeout
+                    line = await asyncio.wait_for(
+                        reader.readline(),
+                        timeout=self._config.socket_timeout_seconds,
+                    )
+                except TimeoutError:
+                    # Client timeout - close connection
+                    logger.debug(f"Client timeout: {peer}")
+                    break
+
+                if not line:
+                    # Client disconnected
+                    break
+
+                # Process the request
+                response = await self._process_request(line)
+
+                # Send response
+                writer.write(response.encode("utf-8") + b"\n")
+                await writer.drain()
+
+        except ConnectionResetError:
+            logger.debug(f"Client connection reset: {peer}")
+        except Exception as e:
+            logger.exception(f"Error handling client {peer}: {e}")
+        finally:
+            try:
+                writer.close()
+                await writer.wait_closed()
+            except Exception:
+                pass
+            logger.debug(f"Client disconnected: {peer}")
+
+    async def _process_request(self, line: bytes) -> str:
+        """Process a single request line.
+
+        Uses typed request models (ModelDaemonPingRequest, ModelDaemonEmitRequest)
+        for compile-time type safety instead of dict[str, object] with isinstance checks.
+
+        Args:
+            line: Raw request line (JSON bytes with optional newline)
+
+        Returns:
+            JSON response string
+        """
+        try:
+            # Parse JSON request
+            raw_request = json.loads(line.decode("utf-8").strip())
+        except (json.JSONDecodeError, UnicodeDecodeError) as e:
+            return ModelDaemonErrorResponse(
+                reason=f"Invalid JSON: {e}"
+            ).model_dump_json()
+
+        if not isinstance(raw_request, dict):
+            return ModelDaemonErrorResponse(
+                reason="Request must be a JSON object"
+            ).model_dump_json()
+
+        # Parse into typed request model
+        try:
+            request = parse_daemon_request(raw_request)
+        except (ValueError, ValidationError) as e:
+            return ModelDaemonErrorResponse(reason=str(e)).model_dump_json()
+
+        # Dispatch based on request type
+        if isinstance(request, ModelDaemonPingRequest):
+            return await self._handle_ping(request)
+        elif isinstance(request, ModelDaemonEmitRequest):
+            return await self._handle_emit(request)
+        else:
+            # Should be unreachable due to exhaustive type check above
+            return ModelDaemonErrorResponse(
+                reason="Unknown request type"
+            ).model_dump_json()
+
+    async def _handle_ping(self, request: ModelDaemonPingRequest) -> str:
+        """Handle ping command request.
+
+        Args:
+            request: Typed ping request model
+
+        Returns:
+            JSON response string with queue status
+        """
+        return ModelDaemonPingResponse(
+            queue_size=self._queue.memory_size(),
+            spool_size=self._queue.spool_size(),
+        ).model_dump_json()
+
+    async def _handle_emit(self, request: ModelDaemonEmitRequest) -> str:
+        """Handle event emission request.
+
+        Args:
+            request: Typed emit request model with event_type and payload
+
+        Returns:
+            JSON response string (queued or error)
+        """
+        event_type = request.event_type
+
+        # Normalize payload to dict (JsonType could be various types)
+        raw_payload = request.payload
+        if raw_payload is None:
+            raw_payload = {}
+        if not isinstance(raw_payload, dict):
+            return ModelDaemonErrorResponse(
+                reason="'payload' must be a JSON object"
+            ).model_dump_json()
+
+        # Cast to dict[str, object] after isinstance check for type safety
+        payload: dict[str, object] = cast("dict[str, object]", raw_payload)
+
+        # Check payload size
+        payload_json = json.dumps(payload)
+        if len(payload_json.encode("utf-8")) > self._config.max_payload_bytes:
+            return ModelDaemonErrorResponse(
+                reason=f"Payload exceeds maximum size of {self._config.max_payload_bytes} bytes"
+            ).model_dump_json()
+
+        # Validate event type is registered
+        try:
+            topic = self._registry.resolve_topic(event_type)
+        except OnexError as e:
+            return ModelDaemonErrorResponse(reason=str(e)).model_dump_json()
+
+        # Validate payload has required fields
+        try:
+            self._registry.validate_payload(event_type, payload)
+        except OnexError as e:
+            return ModelDaemonErrorResponse(reason=str(e)).model_dump_json()
+
+        # Extract correlation_id from payload if present
+        correlation_id = payload.get("correlation_id")
+        if not isinstance(correlation_id, str):
+            correlation_id = None
+
+        # Inject metadata into payload
+        enriched_payload = self._registry.inject_metadata(
+            event_type,
+            payload,
+            correlation_id=correlation_id,
+        )
+
+        # Get partition key
+        partition_key = self._registry.get_partition_key(event_type, enriched_payload)
+
+        # Create queued event
+        event_id = str(uuid4())
+        queued_event = ModelQueuedEvent(
+            event_id=event_id,
+            event_type=event_type,
+            topic=topic,
+            payload=enriched_payload,
+            partition_key=partition_key,
+            queued_at=datetime.now(UTC),
+        )
+
+        # Enqueue the event
+        success = await self._queue.enqueue(queued_event)
+        if success:
+            logger.debug(
+                f"Event queued: {event_id}",
+                extra={
+                    "event_type": event_type,
+                    "topic": topic,
+                },
+            )
+            return ModelDaemonQueuedResponse(event_id=event_id).model_dump_json()
+        else:
+            return ModelDaemonErrorResponse(
+                reason="Failed to queue event (queue may be full)"
+            ).model_dump_json()
+
+    async def _publisher_loop(self) -> None:
+        """Background task that dequeues and publishes events to Kafka.
+
+        Runs continuously until stopped. On publish failure:
+        - Increment retry_count
+        - Re-queue with exponential backoff
+        - After max_retry_attempts (from config), log error and drop event
+        """
+        logger.info("Publisher loop started")
+
+        # NOTE: Using non-locking total_size() is intentional here.
+        # While this creates a theoretical race condition during shutdown
+        # (a concurrent enqueue could complete after the size check but before
+        # the loop re-evaluates), it avoids lock contention in this hot loop.
+        # For fire-and-forget semantics, this trade-off is acceptable - events
+        # queued during the final shutdown window may be lost, which is
+        # documented behavior (see shutdown_drain_seconds config and the
+        # drain_to_spool() call in stop()). The queue's total_size_locked()
+        # method exists for cases requiring accurate counts.
+        while self._running or self._queue.total_size() > 0:
+            try:
+                # Dequeue next event
+                event = await self._queue.dequeue()
+
+                if event is None:
+                    # Queue empty, wait briefly and check again
+                    await asyncio.sleep(PUBLISHER_POLL_INTERVAL_SECONDS)
+                    continue
+
+                # Attempt to publish
+                success = await self._publish_event(event)
+
+                if not success:
+                    # Increment retry count
+                    event.retry_count += 1
+
+                    if event.retry_count >= self._config.max_retry_attempts:
+                        # Max retries exceeded - drop event
+                        logger.error(
+                            f"Dropping event {event.event_id} after {event.retry_count} retries",
+                            extra={
+                                "event_type": event.event_type,
+                                "topic": event.topic,
+                            },
+                        )
+                    else:
+                        # Re-queue with backoff (capped to prevent excessive delays)
+                        uncapped_backoff = self._config.backoff_base_seconds * (
+                            2 ** (event.retry_count - 1)
+                        )
+                        backoff = min(
+                            uncapped_backoff, self._config.max_backoff_seconds
+                        )
+                        logger.warning(
+                            f"Publish failed for {event.event_id}, retry {event.retry_count}/{self._config.max_retry_attempts} in {backoff}s",
+                            extra={
+                                "event_type": event.event_type,
+                                "topic": event.topic,
+                            },
+                        )
+
+                        # Wait for backoff period
+                        await asyncio.sleep(backoff)
+
+                        # Re-enqueue with error handling
+                        requeue_success = await self._queue.enqueue(event)
+                        if not requeue_success:
+                            logger.error(
+                                f"Failed to re-enqueue event {event.event_id} after backoff, event lost",
+                                extra={
+                                    "event_type": event.event_type,
+                                    "topic": event.topic,
+                                    "retry_count": event.retry_count,
+                                },
+                            )
+
+            except asyncio.CancelledError:
+                logger.info("Publisher loop cancelled")
+                break
+            except Exception as e:
+                logger.exception(f"Unexpected error in publisher loop: {e}")
+                await asyncio.sleep(1.0)  # Brief pause before continuing
+
+        logger.info("Publisher loop stopped")
+
+    async def _publish_event(self, event: ModelQueuedEvent) -> bool:
+        """Publish a single event to Kafka.
+
+        Args:
+            event: The queued event to publish
+
+        Returns:
+            True if publish succeeded, False otherwise
+        """
+        if self._event_bus is None:
+            logger.error("Event bus not initialized")
+            return False
+
+        try:
+            # Prepare message key and value
+            key = event.partition_key.encode("utf-8") if event.partition_key else None
+            value = json.dumps(event.payload).encode("utf-8")
+
+            # Extract correlation_id from enriched payload (injected by registry)
+            # Type guard: payload is always a dict in practice (created in _handle_event)
+            payload_correlation_id = (
+                event.payload.get("correlation_id")
+                if isinstance(event.payload, dict)
+                else None
+            )
+            if isinstance(payload_correlation_id, str):
+                try:
+                    correlation_id = UUID(payload_correlation_id)
+                except ValueError:
+                    correlation_id = uuid4()
+            else:
+                correlation_id = uuid4()
+
+            # Create event headers
+            headers = ModelEventHeaders(
+                source="emit-daemon",
+                event_type=event.event_type,
+                timestamp=event.queued_at,
+                correlation_id=correlation_id,
+            )
+
+            # Publish to Kafka
+            # NOTE: headers parameter is Kafka-specific, not in minimal protocol.
+            # When _event_bus is None, we create EventBusKafka which supports headers.
+            # For testing mocks, they can accept **kwargs or ignore extra params.
+            await self._event_bus.publish(  # type: ignore[call-arg]
+                topic=event.topic,
+                key=key,
+                value=value,
+                headers=headers,
+            )
+
+            logger.debug(
+                f"Published event {event.event_id}",
+                extra={
+                    "event_type": event.event_type,
+                    "topic": event.topic,
+                },
+            )
+            return True
+
+        except Exception as e:
+            logger.warning(
+                f"Failed to publish event {event.event_id}: {e}",
+                extra={
+                    "event_type": event.event_type,
+                    "topic": event.topic,
+                    "error": str(e),
+                },
+            )
+            return False
+
+    def _write_pid_file(self) -> None:
+        """Write current PID to pid_path.
+
+        Creates parent directories if needed.
+        """
+        try:
+            self._config.pid_path.parent.mkdir(parents=True, exist_ok=True)
+            self._config.pid_path.write_text(str(os.getpid()))
+            logger.debug(f"PID file created: {self._config.pid_path}")
+        except OSError as e:
+            logger.warning(f"Failed to write PID file: {e}")
+
+    def _remove_pid_file(self) -> None:
+        """Remove PID file if it exists."""
+        try:
+            if self._config.pid_path.exists():
+                self._config.pid_path.unlink()
+                logger.debug(f"PID file removed: {self._config.pid_path}")
+        except OSError as e:
+            logger.warning(f"Failed to remove PID file: {e}")
+
+    def _check_stale_socket(self) -> bool:
+        """Check if socket/PID are stale (process not running).
+
+        A socket/PID is considered stale if:
+        - PID file exists but the process is not running
+        - Socket file exists but no PID file exists
+
+        Returns:
+            True if stale (safe to clean up), False if daemon is running.
+        """
+        # Check if PID file exists
+        if not self._config.pid_path.exists():
+            # No PID file - socket is stale if it exists
+            return self._config.socket_path.exists()
+
+        # Read PID from file
+        try:
+            pid_str = self._config.pid_path.read_text().strip()
+            pid = int(pid_str)
+        except (OSError, ValueError):
+            # Can't read PID file - treat as stale
+            return True
+
+        # Check if process is running
+        try:
+            # Sending signal 0 checks if process exists without killing it
+            os.kill(pid, 0)
+            # Process is running - not stale
+            return False
+        except ProcessLookupError:
+            # Process not running - stale
+            return True
+        except PermissionError:
+            # Process exists but we can't signal it - assume not stale
+            return False
+
+    def _cleanup_stale(self) -> None:
+        """Remove stale socket and PID files."""
+        # Remove socket file
+        if self._config.socket_path.exists():
+            try:
+                self._config.socket_path.unlink()
+                logger.info(f"Removed stale socket: {self._config.socket_path}")
+            except OSError as e:
+                logger.warning(f"Failed to remove stale socket: {e}")
+
+        # Remove PID file
+        if self._config.pid_path.exists():
+            try:
+                self._config.pid_path.unlink()
+                logger.info(f"Removed stale PID file: {self._config.pid_path}")
+            except OSError as e:
+                logger.warning(f"Failed to remove stale PID file: {e}")
+
+
+__all__: list[str] = ["EmitDaemon"]