matrice-analytics 0.1.70__py3-none-any.whl → 0.1.96__py3-none-any.whl

matrice_analytics/post_processing/utils/incident_manager_utils.py

@@ -0,0 +1,1754 @@
+"""
+incident_manager_utils.py
+
+Manages incident publishing to Redis/Kafka when severity levels change.
+Implements consecutive-frame validation before publishing:
+- 5 consecutive frames for medium/significant/critical
+- 10 consecutive frames for low (stricter)
+- 101 consecutive empty frames to send 'info' (incident ended)
+
+Polls 'incident_modification_config' topic for dynamic threshold settings.
+Publishes to 'incident_res' topic.
+
+PRODUCTION-READY VERSION
+"""
+
+import json
+import time
+import threading
+import logging
+import os
+import urllib.request
+import base64
+import re
+from typing import Dict, List, Optional, Any, Tuple
+from datetime import datetime, timezone
+from dataclasses import dataclass, field
+from pathlib import Path
+
+
+# Severity level ordering for comparison (none = no incident)
+SEVERITY_LEVELS = ["none", "info", "low", "medium", "significant", "critical"]
+
+# Default thresholds if none provided (same as fire_detection.py defaults)
+DEFAULT_THRESHOLDS = [
+    {"level": "low", "percentage": 0.0001},
+    {"level": "medium", "percentage": 3},
+    {"level": "significant", "percentage": 13},
+    {"level": "critical", "percentage": 30}
+]
+
+# Cache for location names to avoid repeated API calls
+_location_name_cache: Dict[str, str] = {}
+
+
+@dataclass
+class IncidentState:
+    """Tracks the current incident state for a camera/usecase."""
+    current_level: str = "none"        # Current confirmed severity level
+    pending_level: str = "none"        # Level being validated (needs consecutive frames)
+    consecutive_count: int = 0         # Consecutive frames with pending_level
+    last_published_level: str = "none" # Last level that was published (for spam prevention)
+    incident_cycle_id: int = 1         # Starts at 1, incremented when cycle resets (after info sent)
+    empty_frames_count: int = 0        # Consecutive empty incident frames (for "info" detection)
+    current_incident_id: str = ""      # Current incident_id for this cycle (managed per camera)
+    incident_active: bool = False      # Whether an incident is currently active in this cycle
+
+
+@dataclass
+class ThresholdConfig:
+    """Stores threshold configuration for a camera."""
+    camera_id: str
+    application_id: str = ""
+    app_deployment_id: str = ""
+    incident_type: str = ""
+    thresholds: List[Dict[str, Any]] = field(default_factory=lambda: DEFAULT_THRESHOLDS.copy())
+    last_updated: float = field(default_factory=time.time)
+    camera_name: str = ""  # Store camera_name from config
+
+
+class INCIDENT_MANAGER:
+    """
+    Manages incident severity level tracking and publishing.
+    
+    Key behaviors:
+    - Polls 'incident_modification_config' topic for dynamic threshold settings
+    - Calculates severity_level from incident_quant using thresholds
+    - Publishes incidents ONLY when severity level changes
+    - Requires different consecutive frames based on level:
+      - 5 frames for medium/significant/critical
+      - 10 frames for low (stricter to avoid false positives)
+      - 101 empty frames to send "info" (incident ended)
+    - Supports both Redis and Kafka transports
+    - Thread-safe operations
+    
+    Usage:
+        manager = INCIDENT_MANAGER(redis_client=..., kafka_client=...)
+        manager.start()  # Start config polling
+        manager.process_incident(camera_id, incident_data, stream_info)
+        manager.stop()   # Stop polling on shutdown
+    """
+    
+    # Frame thresholds for different severity levels
+    CONSECUTIVE_FRAMES_DEFAULT = 5       # For medium, significant, critical
+    CONSECUTIVE_FRAMES_LOW = 10          # For low level (stricter)
+    CONSECUTIVE_FRAMES_EMPTY = 101       # For sending "info" after no detections
+    
+    CONFIG_POLLING_INTERVAL = 10  # Poll every 10 seconds
+    CONFIG_TOPIC = "incident_modification_config"
+    INCIDENT_TOPIC = "incident_res"
+    
+    def __init__(
+        self,
+        redis_client: Optional[Any] = None,
+        kafka_client: Optional[Any] = None,
+        incident_topic: str = "incident_res",
+        config_topic: str = "incident_modification_config",
+        logger: Optional[logging.Logger] = None
+    ):
+        """
+        Initialize INCIDENT_MANAGER.
+        
+        Args:
+            redis_client: MatriceStream instance configured for Redis
+            kafka_client: MatriceStream instance configured for Kafka
+            incident_topic: Topic/stream name for publishing incidents
+            config_topic: Topic/stream name for receiving threshold configs
+            logger: Python logger instance
+        """
+        self.redis_client = redis_client
+        self.kafka_client = kafka_client
+        self.incident_topic = incident_topic
+        self.config_topic = config_topic
+        self.logger = logger or logging.getLogger(__name__)
+        
+        # Per-camera incident state tracking: {camera_id: IncidentState}
+        self._incident_states: Dict[str, IncidentState] = {}
+        self._states_lock = threading.Lock()
+        
+        # Per-camera threshold configuration: {camera_id: ThresholdConfig}
+        self._threshold_configs: Dict[str, ThresholdConfig] = {}
+        self._config_lock = threading.Lock()
+        
+        # Config polling thread control
+        self._polling_thread: Optional[threading.Thread] = None
+        self._stop_event = threading.Event()
+        self._running = False
+        
+        # Store factory reference for fetching camera info
+        self._factory_ref: Optional['IncidentManagerFactory'] = None
+        
+        self.logger.info(
+            f"[INCIDENT_MANAGER] Initialized with incident_topic={incident_topic}, "
+            f"config_topic={config_topic}, "
+            f"low_frames={self.CONSECUTIVE_FRAMES_LOW}, "
+            f"default_frames={self.CONSECUTIVE_FRAMES_DEFAULT}, "
+            f"empty_frames_for_info={self.CONSECUTIVE_FRAMES_EMPTY}, "
+            f"polling_interval={self.CONFIG_POLLING_INTERVAL}s"
+        )
+    
+    def set_factory_ref(self, factory: 'IncidentManagerFactory'):
+        """Set reference to factory for accessing deployment info."""
+        self._factory_ref = factory
+    
+    def start(self):
+        """Start the background config polling thread."""
+        if self._running:
+            self.logger.warning("[INCIDENT_MANAGER] Already running")
+            return
+        
+        self._running = True
+        self._stop_event.clear()
+        self._polling_thread = threading.Thread(
+            target=self._config_polling_loop,
+            daemon=True,
+            name="IncidentConfigPoller"
+        )
+        self._polling_thread.start()
+        self.logger.info("[INCIDENT_MANAGER] ✓ Started config polling thread")
+    
+    def stop(self):
+        """Stop the background polling thread gracefully."""
+        if not self._running:
+            return
+        
+        self.logger.info("[INCIDENT_MANAGER] Stopping...")
+        self._running = False
+        self._stop_event.set()
+        
+        if self._polling_thread and self._polling_thread.is_alive():
+            self._polling_thread.join(timeout=5)
+        
+        self.logger.info("[INCIDENT_MANAGER] ✓ Stopped")
+    
+    def _config_polling_loop(self):
+        """Background thread that polls for config updates every CONFIG_POLLING_INTERVAL seconds."""
+        self.logger.info(f"[INCIDENT_MANAGER] Config polling loop started (interval: {self.CONFIG_POLLING_INTERVAL}s)")
+        
+        while not self._stop_event.is_set():
+            try:
+                self._fetch_and_update_configs()
+            except Exception as e:
+                self.logger.error(f"[INCIDENT_MANAGER] Error in config polling loop: {e}", exc_info=True)
+            
+            # Sleep in small increments to allow quick shutdown
+            for _ in range(self.CONFIG_POLLING_INTERVAL):
+                if self._stop_event.is_set():
+                    break
+                time.sleep(1)
+        
+        self.logger.info("[INCIDENT_MANAGER] Config polling loop exited")
+    
+    def _fetch_and_update_configs(self):
+        """Fetch config messages from Redis (primary) or Kafka (fallback)."""
+        configs = []
+        
+        # Try Redis first (primary)
+        if self.redis_client:
+            try:
+                self.logger.debug(f"[INCIDENT_MANAGER] Fetching configs from Redis: {self.config_topic}")
+                configs = self._read_configs_from_redis(max_messages=100)
+                if configs:
+                    self.logger.info(f"[INCIDENT_MANAGER] Fetched {len(configs)} config(s) from Redis")
+            except Exception as e:
+                self.logger.debug(f"[INCIDENT_MANAGER] Redis config fetch: {e}")
+        
+        # Fallback to Kafka if Redis failed or no messages
+        if not configs and self.kafka_client:
+            try:
+                self.logger.debug(f"[INCIDENT_MANAGER] Fetching configs from Kafka: {self.config_topic}")
+                configs = self._read_configs_from_kafka(max_messages=100)
+                if configs:
+                    self.logger.info(f"[INCIDENT_MANAGER] Fetched {len(configs)} config(s) from Kafka")
+            except Exception as e:
+                self.logger.debug(f"[INCIDENT_MANAGER] Kafka config fetch: {e}")
+        
+        # Update in-memory threshold configs
+        for config_data in configs:
+            try:
+                self._handle_config_message(config_data)
+            except Exception as e:
+                self.logger.error(f"[INCIDENT_MANAGER] Error handling config message: {e}", exc_info=True)
+    
+    def _read_configs_from_redis(self, max_messages: int = 100) -> List[Dict[str, Any]]:
+        """Read config messages from Redis stream."""
+        messages = []
+        try:
+            for msg_count in range(max_messages):
+                msg = self.redis_client.get_message(timeout=0.1)
+                if not msg:
+                    break
+                
+                value = msg.get('value') or msg.get('data') or msg.get('message')
+                if value:
+                    parsed = self._parse_message_value(value)
+                    if parsed:
+                        messages.append(parsed)
+        except Exception as e:
+            self.logger.debug(f"[INCIDENT_MANAGER] Error reading from Redis: {e}")
+        
+        return messages
+    
+    def _read_configs_from_kafka(self, max_messages: int = 100) -> List[Dict[str, Any]]:
+        """Read config messages from Kafka topic."""
+        messages = []
+        try:
+            for msg_count in range(max_messages):
+                msg = self.kafka_client.get_message(timeout=0.1)
+                if not msg:
+                    break
+                
+                value = msg.get('value') or msg.get('data') or msg.get('message')
+                if value:
+                    parsed = self._parse_message_value(value)
+                    if parsed:
+                        messages.append(parsed)
+        except Exception as e:
+            self.logger.debug(f"[INCIDENT_MANAGER] Error reading from Kafka: {e}")
+        
+        return messages
+    
+    def _parse_message_value(self, value: Any) -> Optional[Dict[str, Any]]:
+        """Parse message value into a dictionary."""
+        try:
+            # Already a dict
+            if isinstance(value, dict):
+                if 'data' in value and isinstance(value['data'], dict):
+                    return value['data']
+                return value
+            
+            # Bytes to string
+            if isinstance(value, bytes):
+                value = value.decode('utf-8')
+            
+            # Parse JSON string
+            if isinstance(value, str):
+                try:
+                    return json.loads(value)
+                except json.JSONDecodeError:
+                    # Try fixing Python-style formatting
+                    fixed = value
+                    fixed = fixed.replace(": True", ": true").replace(": False", ": false")
+                    fixed = fixed.replace(":True", ":true").replace(":False", ":false")
+                    fixed = fixed.replace(": None", ": null").replace(":None", ":null")
+                    if "'" in fixed and '"' not in fixed:
+                        fixed = fixed.replace("'", '"')
+                    return json.loads(fixed)
+        except Exception as e:
+            self.logger.debug(f"[INCIDENT_MANAGER] Failed to parse message: {e}")
+        
+        return None
+    
+    def _handle_config_message(self, config_data: Dict[str, Any]):
+        """
+        Handle a threshold config message.
+        
+        Expected format:
+        {
+            "camera_id": "68f9d95cfaff6151c774e0e7",
+            "application_id": "...",
+            "app_deployment_id": "...",
+            "incident_type": "fire",
+            "camera_name": "camera_1",
+            "thresholds": [
+                {"level": "low", "percentage": 0.0001},
+                {"level": "medium", "percentage": 3},
+                {"level": "significant", "percentage": 13},
+                {"level": "critical", "percentage": 30}
+            ]
+        }
+        """
+        try:
+            camera_id = config_data.get("camera_id", "")
+            if not camera_id:
+                self.logger.debug("[INCIDENT_MANAGER] Config message missing camera_id, skipping")
+                return
+            
+            # Extract fields with defaults
+            application_id = config_data.get("application_id", "")
+            app_deployment_id = config_data.get("app_deployment_id", "")
+            incident_type = config_data.get("incident_type", "")
+            camera_name = config_data.get("camera_name", "")
+            thresholds = config_data.get("thresholds", [])
+            
+            # Validate thresholds - use defaults if invalid
+            if not thresholds or not isinstance(thresholds, list):
+                thresholds = DEFAULT_THRESHOLDS.copy()
+                self.logger.debug(f"[INCIDENT_MANAGER] Using default thresholds for camera: {camera_id}")
+            else:
+                # Validate each threshold has required fields
+                # Also map "high" -> "significant" (backend uses "high", we use "significant")
+                valid_thresholds = []
+                for t in thresholds:
+                    if isinstance(t, dict) and "level" in t and "percentage" in t:
+                        level = t.get("level", "").lower().strip()
+                        # Map "high" to "significant" when receiving from backend
+                        if level == "high":
+                            self.logger.debug(f"[INCIDENT_MANAGER] Mapping level 'high' -> 'significant' for camera {camera_id}")
+                            t = dict(t)  # Make a copy to avoid modifying original
+                            t["level"] = "significant"
+                        valid_thresholds.append(t)
+                
+                if not valid_thresholds:
+                    thresholds = DEFAULT_THRESHOLDS.copy()
+                else:
+                    thresholds = valid_thresholds
+            
+            # Create or update threshold config
+            with self._config_lock:
+                self._threshold_configs[camera_id] = ThresholdConfig(
+                    camera_id=camera_id,
+                    application_id=application_id,
+                    app_deployment_id=app_deployment_id,
+                    incident_type=incident_type,
+                    thresholds=thresholds,
+                    last_updated=time.time(),
+                    camera_name=camera_name
+                )
+            
+            self.logger.info(
+                f"[INCIDENT_MANAGER] ✓ Updated thresholds for camera: {camera_id}, "
+                f"thresholds: {thresholds}"
+            )
+            
+        except Exception as e:
+            self.logger.error(f"[INCIDENT_MANAGER] Error handling config message: {e}", exc_info=True)
+    
+    def _get_thresholds_for_camera(self, camera_id: str) -> Tuple[List[Dict[str, Any]], Optional[ThresholdConfig]]:
+        """
+        Get thresholds for a specific camera, or defaults if not configured.
+        
+        Returns:
+            Tuple of (thresholds list, ThresholdConfig or None)
+        """
+        with self._config_lock:
+            config = self._threshold_configs.get(camera_id)
+            if config:
+                return config.thresholds, config
+            return DEFAULT_THRESHOLDS, None
+    
+    def _calculate_severity_from_quant(
+        self, 
+        incident_quant: float, 
+        thresholds: List[Dict[str, Any]]
+    ) -> str:
+        """
+        Calculate severity level from incident_quant using thresholds.
+        
+        Args:
+            incident_quant: The quantitative value (e.g., intensity percentage)
+            thresholds: List of threshold configs sorted by percentage
+            
+        Returns:
+            Severity level string (none, low, medium, significant, critical)
+        """
+        if incident_quant is None or incident_quant < 0:
+            return "none"
+        
+        # Sort thresholds by percentage (ascending)
+        sorted_thresholds = sorted(thresholds, key=lambda x: float(x.get("percentage", 0)))
+        
+        # Find the highest level where percentage threshold is met
+        severity = "none"
+        for t in sorted_thresholds:
+            level = t.get("level", "").lower()
+            percentage = float(t.get("percentage", 0))
+            
+            if incident_quant >= percentage:
+                severity = level
+            else:
+                break  # Since sorted ascending, no need to check further
+        
+        # Validate severity
+        if severity not in SEVERITY_LEVELS:
+            severity = "none"
+        
+        return severity
+    
+    def _get_frames_required_for_level(self, level: str) -> int:
+        """
+        Get the number of consecutive frames required to confirm a level.
+        
+        Args:
+            level: Severity level string
+            
+        Returns:
+            Number of consecutive frames required
+        """
+        if level == "low":
+            return self.CONSECUTIVE_FRAMES_LOW  # 10 frames for low (stricter)
+        return self.CONSECUTIVE_FRAMES_DEFAULT  # 5 frames for others
+    
+    def _extract_camera_info_from_stream(
+        self, 
+        stream_info: Optional[Dict[str, Any]]
+    ) -> Dict[str, str]:
+        """
+        Extract camera info from stream_info (similar to ResultsIngestor pattern).
+        
+        Stream info structure example:
+        {
+            'broker': 'localhost:9092', 
+            'topic': '692d7bde42582ffde3611908_input_topic',  # camera_id is prefix before _input_topic
+            'stream_time': '2025-12-02-05:09:53.914224 UTC', 
+            'camera_info': {
+                'camera_name': 'cusstomer-cam-1', 
+                'camera_group': 'staging-customer-1', 
+                'location': '6908756db129880c34f2e09a'
+            }, 
+            'frame_id': '7b94e2f668fb456f95b73c3084e17f8a'
+        }
+        
+        Args:
+            stream_info: Stream metadata from usecase
+            
+        Returns:
+            Dict with camera_id, camera_name, app_deployment_id, application_id, frame_id, location_id
+        """
+        result = {
+            "camera_id": "",
+            "camera_name": "",
+            "app_deployment_id": "",
+            "application_id": "",
+            "frame_id": "",
+            "location_id": ""
+        }
+        
+        if not stream_info:
+            return result
+        
+        try:
+            # Try multiple paths to get camera info (like ResultsIngestor)
+            # Path 1: Direct camera_info in stream_info
+            camera_info = stream_info.get("camera_info", {}) or {}
+            
+            # Path 2: From input_settings -> input_stream pattern
+            input_settings = stream_info.get("input_settings", {}) or {}
+            input_stream = input_settings.get("input_stream", {}) or {}
+            input_camera_info = input_stream.get("camera_info", {}) or {}
+            
+            # Path 3: From input_streams array (like ResultsIngestor)
+            input_streams = stream_info.get("input_streams", [])
+            if input_streams and len(input_streams) > 0:
+                input_data = input_streams[0] if isinstance(input_streams[0], dict) else {}
+                input_stream_inner = input_data.get("input_stream", input_data)
+                input_camera_info = input_stream_inner.get("camera_info", {}) or input_camera_info
+            
+            # Merge all sources, preferring non-empty values
+            # camera_name - check all possible locations
+            result["camera_name"] = (
+                camera_info.get("camera_name", "") or
+                camera_info.get("cameraName", "") or
+                input_camera_info.get("camera_name", "") or
+                input_camera_info.get("cameraName", "") or
+                stream_info.get("camera_name", "") or
+                stream_info.get("cameraName", "") or
+                input_settings.get("camera_name", "") or
+                input_settings.get("cameraName", "") or
+                ""
+            )
+            
+            # camera_id - check direct fields first
+            result["camera_id"] = (
+                camera_info.get("camera_id", "") or
+                camera_info.get("cameraId", "") or
+                input_camera_info.get("camera_id", "") or
+                input_camera_info.get("cameraId", "") or
+                stream_info.get("camera_id", "") or
+                stream_info.get("cameraId", "") or
+                input_settings.get("camera_id", "") or
+                input_settings.get("cameraId", "") or
+                ""
+            )
+            
+            # If camera_id still not found, extract from topic
+            # Topic format: {camera_id}_input_topic (e.g., "692d7bde42582ffde3611908_input_topic")
+            if not result["camera_id"]:
+                topic = stream_info.get("topic", "")
+                if topic:
+                    extracted_camera_id = ""
+                    if topic.endswith("_input_topic"):
+                        extracted_camera_id = topic[: -len("_input_topic")]
+                        self.logger.debug(f"[INCIDENT_MANAGER] Extracted camera_id from topic (underscore): {extracted_camera_id}")
+                    elif topic.endswith("_input-topic"):
+                        extracted_camera_id = topic[: -len("_input-topic")]
+                        self.logger.debug(f"[INCIDENT_MANAGER] Extracted camera_id from topic (hyphen): {extracted_camera_id}")
+                    else:
+                        if "_input_topic" in topic:
+                            extracted_camera_id = topic.split("_input_topic")[0]
+                            self.logger.debug(f"[INCIDENT_MANAGER] Extracted camera_id from topic split (underscore): {extracted_camera_id}")
+                        elif "_input-topic" in topic:
+                            extracted_camera_id = topic.split("_input-topic")[0]
+                            self.logger.debug(f"[INCIDENT_MANAGER] Extracted camera_id from topic split (hyphen): {extracted_camera_id}")
+                    if extracted_camera_id:
+                        result["camera_id"] = extracted_camera_id
+            
+            # app_deployment_id
+            result["app_deployment_id"] = (
+                stream_info.get("app_deployment_id", "") or
+                stream_info.get("appDeploymentId", "") or
+                stream_info.get("app_deploymentId", "") or
+                input_settings.get("app_deployment_id", "") or
+                input_settings.get("appDeploymentId", "") or
+                camera_info.get("app_deployment_id", "") or
+                camera_info.get("appDeploymentId", "") or
+                ""
+            )
+            
+            # application_id
+            result["application_id"] = (
+                stream_info.get("application_id", "") or
+                stream_info.get("applicationId", "") or
+                stream_info.get("app_id", "") or
+                stream_info.get("appId", "") or
+                input_settings.get("application_id", "") or
+                input_settings.get("applicationId", "") or
+                camera_info.get("application_id", "") or
+                camera_info.get("applicationId", "") or
+                ""
+            )
+            
+            # frame_id - at top level of stream_info
+            result["frame_id"] = (
+                stream_info.get("frame_id", "") or
+                stream_info.get("frameId", "") or
+                input_settings.get("frame_id", "") or
+                input_settings.get("frameId", "") or
+                ""
+            )
+            
+            # location_id - from camera_info.location
+            result["location_id"] = (
+                camera_info.get("location", "") or
+                camera_info.get("location_id", "") or
+                camera_info.get("locationId", "") or
+                input_camera_info.get("location", "") or
+                input_camera_info.get("location_id", "") or
+                ""
+            )
+            
+            self.logger.debug(
+                f"[INCIDENT_MANAGER] Extracted from stream_info - "
+                f"camera_id={result['camera_id']}, camera_name={result['camera_name']}, "
+                f"app_deployment_id={result['app_deployment_id']}, application_id={result['application_id']}, "
+                f"frame_id={result['frame_id']}, location_id={result['location_id']}"
+            )
+            
+        except Exception as e:
+            self.logger.debug(f"[INCIDENT_MANAGER] Error extracting camera info: {e}")
+        
+        return result
+    
+    def _map_level_from_backend(self, level: str) -> str:
+        """Map level from backend terminology to internal terminology.
+        
+        Backend uses 'high', we use 'significant' internally.
+        """
+        if level and level.lower().strip() == "high":
+            return "significant"
+        return level
+    
+    def _map_level_to_backend(self, level: str) -> str:
+        """Map level from internal terminology to backend terminology.
+        
+        We use 'significant' internally, backend expects 'high'.
+        """
+        if level and level.lower().strip() == "significant":
+            return "high"
+        return level
+    
+    def _fetch_location_name(self, location_id: str) -> str:
+        """
+        Fetch location name from API using location_id.
+        
+        Args:
+            location_id: The location ID to look up
+            
+        Returns:
+            Location name string, or 'Entry Reception' as default if API fails
+        """
+        global _location_name_cache
+        default_location = "Entry Reception"
+        
+        if not location_id:
+            self.logger.debug(f"[INCIDENT_MANAGER] No location_id provided, using default: '{default_location}'")
+            return default_location
+        
+        # Check cache first
+        if location_id in _location_name_cache:
+            cached_name = _location_name_cache[location_id]
+            self.logger.debug(f"[INCIDENT_MANAGER] Using cached location name for '{location_id}': '{cached_name}'")
+            return cached_name
+        
+        # Need factory reference with session to make API call
+        if not self._factory_ref or not self._factory_ref._session:
+            self.logger.warning(f"[INCIDENT_MANAGER] No session available for location API, using default: '{default_location}'")
+            return default_location
+        
+        try:
+            endpoint = f"/v1/inference/get_location/{location_id}"
+            self.logger.info(f"[INCIDENT_MANAGER] Fetching location name from API: {endpoint}")
+            
+            response = self._factory_ref._session.rpc.get(endpoint)
+            
+            if response and isinstance(response, dict):
+                success = response.get("success", False)
+                if success:
+                    data = response.get("data", {})
+                    location_name = data.get("locationName", default_location)
+                    self.logger.info(f"[INCIDENT_MANAGER] ✓ Fetched location name: '{location_name}' for location_id: '{location_id}'")
+                    
+                    # Cache the result
+                    _location_name_cache[location_id] = location_name
+                    return location_name
+                else:
+                    self.logger.warning(
+                        f"[INCIDENT_MANAGER] API returned success=false for location_id '{location_id}': "
+                        f"{response.get('message', 'Unknown error')}"
+                    )
+            else:
+                self.logger.warning(f"[INCIDENT_MANAGER] Invalid response format from API: {response}")
+                
+        except Exception as e:
+            self.logger.error(f"[INCIDENT_MANAGER] Error fetching location name for '{location_id}': {e}", exc_info=True)
+        
+        # Use default on any failure
+        self.logger.info(f"[INCIDENT_MANAGER] Using default location name: '{default_location}'")
+        _location_name_cache[location_id] = default_location
+        return default_location
+    
+    def _generate_incident_id(self, camera_id: str, cycle_id: int) -> str:
+        """Generate a unique incident_id for a camera's cycle."""
+        return f"incident_{camera_id}_{cycle_id}"
+    
+    def process_incident(
+        self,
+        camera_id: str,
+        incident_data: Dict[str, Any],
+        stream_info: Optional[Dict[str, Any]] = None
+    ) -> bool:
+        """
+        Process an incident and publish if severity level changed.
+        
+        This method:
+        1. Gets incident_quant from incident_data
+        2. Calculates severity_level using dynamic thresholds for this camera
+        3. Updates incident_data with new severity_level
+        4. Tracks level changes with consecutive-frame validation:
+           - 5 frames for medium/significant/critical
+           - 10 frames for low (stricter)
+        5. Tracks empty incidents and publishes "info" after 101 consecutive empty frames
+        6. Publishes on level change
+        7. Manages incident_id per camera per cycle (increments after info is sent)
+        
+        Args:
+            camera_id: Unique camera identifier
+            incident_data: Incident dictionary from usecase (must include incident_quant)
+            stream_info: Stream metadata
+            
+        Returns:
+            True if incident was published, False otherwise
+        """
+        try:
+            self.logger.debug(f"[INCIDENT_MANAGER] Processing incident for camera: {camera_id}")
+            
+            # Get or create state for this camera
+            with self._states_lock:
+                if camera_id not in self._incident_states:
+                    new_state = IncidentState()
+                    # Initialize incident_id for new camera
+                    new_state.current_incident_id = self._generate_incident_id(camera_id, new_state.incident_cycle_id)
+                    self._incident_states[camera_id] = new_state
+                    self.logger.info(
+                        f"[INCIDENT_MANAGER] Created new state for camera: {camera_id}, "
+                        f"initial incident_id: {new_state.current_incident_id}"
+                    )
+                
+                state = self._incident_states[camera_id]
+                
+                # Ensure incident_id is set (for existing states that may not have it)
+                if not state.current_incident_id:
+                    state.current_incident_id = self._generate_incident_id(camera_id, state.incident_cycle_id)
+                    self.logger.info(f"[INCIDENT_MANAGER] Generated incident_id for existing state: {state.current_incident_id}")
+            
+            # Handle empty incident data - track for "info" level
+            is_empty_incident = (not incident_data or incident_data == {})
+            
+            if is_empty_incident:
+                self.logger.debug("[INCIDENT_MANAGER] Empty incident data, tracking for info level")
+                return self._handle_empty_incident(camera_id, state, stream_info)
+            
+            # Step 1: Get thresholds for this camera
+            thresholds, threshold_config = self._get_thresholds_for_camera(camera_id)
+            
+            # Step 2: Get incident_quant and calculate severity level dynamically
+            incident_quant = incident_data.get("incident_quant")
+            
+            if incident_quant is not None:
+                # Calculate severity from quant using dynamic thresholds
+                severity_level = self._calculate_severity_from_quant(incident_quant, thresholds)
+                
+                # Update incident_data with new severity level
+                incident_data["severity_level"] = severity_level
+                
+                self.logger.debug(
+                    f"[INCIDENT_MANAGER] Calculated severity from incident_quant={incident_quant}: "
+                    f"severity_level={severity_level}"
+                )
+            else:
+                # Fallback to existing severity_level in incident_data
+                severity_level = incident_data.get("severity_level", "none")
+                if not severity_level or severity_level == "":
+                    severity_level = "none"
+            
+            # Store threshold config info in incident_data for output message
+            if threshold_config:
+                incident_data["_config_camera_id"] = threshold_config.camera_id
+                incident_data["_config_application_id"] = threshold_config.application_id
+                incident_data["_config_app_deployment_id"] = threshold_config.app_deployment_id
+                incident_data["_config_camera_name"] = threshold_config.camera_name
+            
+            severity_level = severity_level.lower().strip()
+            
+            self.logger.debug(f"[INCIDENT_MANAGER] Final severity_level: '{severity_level}'")
+            
+            # Validate severity level
+            if severity_level not in SEVERITY_LEVELS:
+                self.logger.warning(
+                    f"[INCIDENT_MANAGER] Unknown severity level '{severity_level}', treating as 'none'"
+                )
+                severity_level = "none"
+            
+            # If level is "none", treat as empty incident (DO NOT reset empty_frames_count here!)
+            if severity_level == "none":
+                return self._handle_empty_incident(camera_id, state, stream_info)
+            
+            # We have a real detection (severity != none), reset empty frame counter
+            with self._states_lock:
+                state.empty_frames_count = 0
+            
+            with self._states_lock:
+                self.logger.debug(
+                    f"[INCIDENT_MANAGER] Current state - "
+                    f"current_level={state.current_level}, "
+                    f"pending_level={state.pending_level}, "
+                    f"consecutive_count={state.consecutive_count}, "
+                    f"last_published_level={state.last_published_level}, "
+                    f"incident_id={state.current_incident_id}, "
+                    f"cycle_id={state.incident_cycle_id}, "
+                    f"incident_active={state.incident_active}"
+                )
+                
+                # Check if this is a new pending level or continuation
+                if severity_level == state.pending_level:
+                    # Same level, increment counter
+                    state.consecutive_count += 1
+                    self.logger.debug(
+                        f"[INCIDENT_MANAGER] Same pending level, "
+                        f"consecutive_count now: {state.consecutive_count}"
+                    )
+                else:
+                    # Different level, reset counter
+                    state.pending_level = severity_level
+                    state.consecutive_count = 1
+                    self.logger.debug(
+                        f"[INCIDENT_MANAGER] New pending level: {severity_level}, "
+                        f"reset consecutive_count to 1"
+                    )
+                
+                # Get required frames for this level
+                frames_required = self._get_frames_required_for_level(severity_level)
+                
+                # Check if we've reached the threshold for confirmation
+                if state.consecutive_count >= frames_required:
+                    # Level is confirmed after required consecutive frames
+                    old_level = state.current_level
+                    new_level = state.pending_level
+                    
+                    self.logger.info(
+                        f"[INCIDENT_MANAGER] Level confirmed after {state.consecutive_count} frames "
+                        f"(required: {frames_required}): {old_level} -> {new_level}"
+                    )
+                    
+                    # Check if level actually changed
+                    if new_level != state.current_level:
+                        state.current_level = new_level
+                        
+                        # Check if we should publish
+                        # 1. Don't publish "none" level (no incident)
+                        # 2. Don't publish same level again (spam prevention)
+                        should_publish = (
+                            new_level != "none" and
+                            new_level != state.last_published_level
+                        )
+                        
+                        self.logger.info(
+                            f"[INCIDENT_MANAGER] Level changed: {old_level} -> {new_level}, "
+                            f"should_publish={should_publish} "
+                            f"(last_published={state.last_published_level})"
+                        )
+                        
+                        if should_publish:
+                            # Mark incident as active for this cycle
+                            state.incident_active = True
+                            
+                            # Use the managed incident_id for this cycle
+                            incident_data["incident_id"] = state.current_incident_id
+                            
+                            # Publish the incident
+                            success = self._publish_incident(
+                                camera_id, incident_data, stream_info
+                            )
+                            if success:
+                                state.last_published_level = new_level
+                                self.logger.info(
+                                    f"[INCIDENT_MANAGER] ✓ Published incident for level: {new_level}, "
+                                    f"incident_id: {state.current_incident_id}"
+                                )
+                            return success
+                        else:
+                            self.logger.debug(
+                                f"[INCIDENT_MANAGER] Skipping publish - "
+                                f"level={new_level}, already published"
+                            )
+                    else:
+                        self.logger.debug(
+                            f"[INCIDENT_MANAGER] No level change, staying at: {state.current_level}"
+                        )
+                
+                return False
+            
+        except Exception as e:
+            self.logger.error(
+                f"[INCIDENT_MANAGER] Error processing incident: {e}", 
+                exc_info=True
+            )
+            return False
+    
+    def _handle_empty_incident(
+        self, 
+        camera_id: str, 
+        state: IncidentState,
+        stream_info: Optional[Dict[str, Any]] = None
+    ) -> bool:
+        """
+        Handle empty incident (no detection).
+        
+        After 101 consecutive empty frames, send "info" level if an incident was active.
+        Info uses the SAME incident_id as the current cycle, then starts a new cycle.
+        
+        Args:
+            camera_id: Camera identifier
+            state: Current incident state
+            stream_info: Stream metadata
+            
+        Returns:
+            True if "info" incident was published, False otherwise
+        """
+        with self._states_lock:
+            state.empty_frames_count += 1
+            
+            self.logger.debug(
+                f"[INCIDENT_MANAGER] Empty frame count for camera {camera_id}: "
+                f"{state.empty_frames_count}/{self.CONSECUTIVE_FRAMES_EMPTY}, "
+                f"incident_active={state.incident_active}, "
+                f"current_incident_id={state.current_incident_id}"
+            )
+            
+            # Reset pending level tracking when empty
+            if state.pending_level not in ("none", "info"):
+                state.pending_level = "none"
+                state.consecutive_count = 0
+            
+            # Check if we should send "info" (incident ended)
+            if state.empty_frames_count >= self.CONSECUTIVE_FRAMES_EMPTY:
+                # Only send "info" if:
+                # 1. An incident was actually active in this cycle (we published something)
+                # 2. Last published level was NOT "info" (don't send duplicate info)
+                should_send_info = (
+                    state.incident_active and 
+                    state.last_published_level not in ("info", "none")
+                )
+                
+                if should_send_info:
+                    self.logger.info(
+                        f"[INCIDENT_MANAGER] {self.CONSECUTIVE_FRAMES_EMPTY} consecutive empty frames for camera {camera_id}, "
+                        f"sending 'info' level to close incident cycle "
+                        f"(last_published={state.last_published_level}, incident_id={state.current_incident_id})"
+                    )
+                    
+                    # Get incident_type from threshold config if available
+                    incident_type = "fire_smoke_detection"  # Default
+                    with self._config_lock:
+                        config = self._threshold_configs.get(camera_id)
+                        if config and config.incident_type:
+                            incident_type = config.incident_type
+                    
+                    # Create info incident data - USE THE SAME incident_id from this cycle!
+                    info_incident = {
+                        "incident_id": state.current_incident_id,  # Same incident_id for this cycle
+                        "incident_type": incident_type,
+                        "severity_level": "info",
+                        "human_text": "Incident ended"
+                    }
+                    
+                    # Update state BEFORE publishing
+                    state.current_level = "info"
+                    state.empty_frames_count = 0  # Reset counter
+                    
+                    # Publish info incident
+                    success = self._publish_incident(camera_id, info_incident, stream_info)
+                    if success:
+                        state.last_published_level = "info"
+                        
+                        # END THIS CYCLE - Start a new cycle for future incidents
+                        old_cycle_id = state.incident_cycle_id
+                        old_incident_id = state.current_incident_id
+                        
+                        state.incident_cycle_id += 1  # Increment cycle
+                        state.current_incident_id = self._generate_incident_id(camera_id, state.incident_cycle_id)
+                        state.incident_active = False  # No active incident in new cycle yet
+                        state.current_level = "none"  # Reset level for new cycle
+                        state.pending_level = "none"
+                        state.consecutive_count = 0
+                        # Note: We keep last_published_level as "info" to prevent duplicate info sends
+                        
+                        self.logger.info(
+                            f"[INCIDENT_MANAGER] ✓ Published 'info' for camera {camera_id}, "
+                            f"closed incident_id={old_incident_id} (cycle {old_cycle_id}), "
+                            f"started new cycle {state.incident_cycle_id} with incident_id={state.current_incident_id}"
+                        )
+                    return success
+                else:
+                    # No active incident or already sent info
+                    if not state.incident_active:
+                        self.logger.debug(
+                            f"[INCIDENT_MANAGER] Skipping 'info' for camera {camera_id} - "
+                            f"no incident was active in this cycle"
+                        )
+                    else:
+                        self.logger.debug(
+                            f"[INCIDENT_MANAGER] Skipping 'info' for camera {camera_id} - "
+                            f"last_published is already '{state.last_published_level}'"
+                        )
+                    
+                    # Reset empty frame counter if we decide not to send info
+                    # to avoid repeated checks every frame after 101
+                    state.empty_frames_count = 0
+            
+            return False
+    
+    def _publish_incident(
+        self,
+        camera_id: str,
+        incident_data: Dict[str, Any],
+        stream_info: Optional[Dict[str, Any]] = None
+    ) -> bool:
+        """
+        Publish incident to Redis/Kafka topic.
+        
+        Args:
+            camera_id: Camera identifier
+            incident_data: Incident dictionary
+            stream_info: Stream metadata
+            
+        Returns:
+            True if published successfully, False otherwise
+        """
+        self.logger.info(f"[INCIDENT_MANAGER] ========== PUBLISHING INCIDENT ==========")
+        
+        try:
+            # Build the incident message
+            message = self._build_incident_message(camera_id, incident_data, stream_info)
+            
+            self.logger.info(f"[INCIDENT_MANAGER] Built incident message: {json.dumps(message, default=str)[:500]}...")
+            
+            success = False
+            
+            # Try Redis first (primary)
+            if self.redis_client:
+                try:
+                    self.logger.debug(
+                        f"[INCIDENT_MANAGER] Publishing to Redis stream: {self.incident_topic}"
+                    )
+                    self._publish_to_redis(self.incident_topic, message)
+                    self.logger.info(
+                        f"[INCIDENT_MANAGER] ✓ Incident published to Redis"
+                    )
+                    success = True
+                except Exception as e:
+                    self.logger.error(
+                        f"[INCIDENT_MANAGER] ❌ Redis publish failed: {e}", 
+                        exc_info=True
+                    )
+            
+            # Fallback to Kafka if Redis failed or no Redis client
+            if not success and self.kafka_client:
+                try:
+                    self.logger.debug(
+                        f"[INCIDENT_MANAGER] Publishing to Kafka topic: {self.incident_topic}"
+                    )
+                    self._publish_to_kafka(self.incident_topic, message)
+                    self.logger.info(
+                        f"[INCIDENT_MANAGER] ✓ Incident published to Kafka"
+                    )
+                    success = True
+                except Exception as e:
+                    self.logger.error(
+                        f"[INCIDENT_MANAGER] ❌ Kafka publish failed: {e}", 
+                        exc_info=True
+                    )
+            
+            if success:
+                self.logger.info(f"[INCIDENT_MANAGER] ========== INCIDENT PUBLISHED ==========")
+            else:
+                self.logger.error(
+                    f"[INCIDENT_MANAGER] ❌ INCIDENT NOT PUBLISHED (both transports failed)"
+                )
+            
+            return success
+            
+        except Exception as e:
+            self.logger.error(
+                f"[INCIDENT_MANAGER] Error publishing incident: {e}", 
+                exc_info=True
+            )
+            return False
+    
+    def _build_incident_message(
+        self,
+        camera_id: str,
+        incident_data: Dict[str, Any],
+        stream_info: Optional[Dict[str, Any]] = None
+    ) -> Dict[str, Any]:
+        """
+        Build the incident message in the required format.
+        
+        Output format (STRICT):
+        {
+            "camera_id": "...",
+            "app_deployment_id": "...",
+            "application_id": "...",
+            "camera_name": "...",
+            "frame_id": "...",
+            "location_name": "...",
+            "incidents": [{
+                "incident_id": "...",
+                "incident_type": "...",
+                "severity_level": "...",
+                "human_text": "..."
+            }]
+        }
+        
+        Keys to REMOVE: "alerts", "alert_settings", "duration", "incident_quant", 
+                        "start_time", "end_time", "camera_info", "level_settings"
+        """
+        
+        # Extract camera info from multiple sources
+        stream_camera_info = self._extract_camera_info_from_stream(stream_info)
+        
+        # Get IDs from threshold config (if available - set by config polling)
+        config_camera_id = incident_data.get("_config_camera_id", "")
+        config_application_id = incident_data.get("_config_application_id", "")
+        config_app_deployment_id = incident_data.get("_config_app_deployment_id", "")
+        config_camera_name = incident_data.get("_config_camera_name", "")
+        
+        # Get IDs from factory (from action_details)
+        factory_app_deployment_id = ""
+        factory_application_id = ""
+        if self._factory_ref:
+            factory_app_deployment_id = self._factory_ref._app_deployment_id or ""
+            factory_application_id = self._factory_ref._application_id or ""
+        
+        # Priority: stream_info > threshold_config > factory > camera_id param
+        final_camera_id = (
+            stream_camera_info.get("camera_id") or 
+            config_camera_id or 
+            camera_id or 
+            ""
+        )
+        
+        final_camera_name = (
+            stream_camera_info.get("camera_name") or 
+            config_camera_name or 
+            ""
+        )
+        
+        final_app_deployment_id = (
+            stream_camera_info.get("app_deployment_id") or 
+            config_app_deployment_id or 
+            factory_app_deployment_id or 
+            ""
+        )
+        
+        final_application_id = (
+            stream_camera_info.get("application_id") or 
+            config_application_id or 
+            factory_application_id or 
+            ""
+        )
+        
+        # Extract frame_id from stream_info
+        final_frame_id = stream_camera_info.get("frame_id", "")
+        
+        # Fetch location_name from API using location_id
+        location_id = stream_camera_info.get("location_id", "")
+        final_location_name = self._fetch_location_name(location_id)
+        
+        self.logger.info(
+            f"[INCIDENT_MANAGER] Building message with - "
+            f"camera_id={final_camera_id}, camera_name={final_camera_name}, "
+            f"app_deployment_id={final_app_deployment_id}, application_id={final_application_id}, "
+            f"frame_id={final_frame_id}, location_name={final_location_name}"
+        )
+        
+        # Build incident - ONLY include required fields
+        # Map "significant" -> "high" for backend (we use "significant" internally, backend expects "high")
+        severity_level = incident_data.get("severity_level", "")
+        if severity_level.lower().strip() == "significant":
+            severity_level = "high"
+            self.logger.debug(f"[INCIDENT_MANAGER] Mapped severity_level 'significant' -> 'high' for publishing")
+        
+        incident = {
+            "incident_id": incident_data.get("incident_id", ""),
+            "incident_type": incident_data.get("incident_type", "fire_smoke_detection"),
+            "severity_level": severity_level,
+            "human_text": incident_data.get("human_text", "")
+        }
+        
+        # Build final message with all required fields
+        message = {
+            "camera_id": final_camera_id,
+            "app_deployment_id": final_app_deployment_id,
+            "application_id": final_application_id,
+            "camera_name": final_camera_name,
+            "frame_id": final_frame_id,
+            "location_name": final_location_name,
+            "incidents": [incident]
+        }
+        
+        return message
+    
+    def _publish_to_redis(self, topic: str, message: Dict[str, Any]):
+        """Publish message to Redis stream."""
+        try:
+            self.redis_client.add_message(
+                topic_or_channel=topic,
+                message=json.dumps(message),
+                key=message.get("camera_id", "")
+            )
+        except Exception as e:
+            self.logger.error(f"[INCIDENT_MANAGER] Redis publish error: {e}")
+            raise
+    
+    def _publish_to_kafka(self, topic: str, message: Dict[str, Any]):
+        """Publish message to Kafka topic."""
+        try:
+            self.kafka_client.add_message(
+                topic_or_channel=topic,
+                message=json.dumps(message),
+                key=message.get("camera_id", "")
+            )
+        except Exception as e:
+            self.logger.error(f"[INCIDENT_MANAGER] Kafka publish error: {e}")
+            raise
+    
+    def reset_camera_state(self, camera_id: str):
+        """Reset incident state for a specific camera."""
+        with self._states_lock:
+            if camera_id in self._incident_states:
+                self._incident_states[camera_id] = IncidentState()
+                self.logger.info(f"[INCIDENT_MANAGER] Reset state for camera: {camera_id}")
+    
+    def get_camera_state(self, camera_id: str) -> Optional[Dict[str, Any]]:
+        """Get current incident state for a camera (for debugging)."""
+        with self._states_lock:
+            state = self._incident_states.get(camera_id)
+            if state:
+                return {
+                    "current_level": state.current_level,
+                    "pending_level": state.pending_level,
+                    "consecutive_count": state.consecutive_count,
+                    "last_published_level": state.last_published_level,
+                    "incident_cycle_id": state.incident_cycle_id,
+                    "empty_frames_count": state.empty_frames_count,
+                    "current_incident_id": state.current_incident_id,
+                    "incident_active": state.incident_active
+                }
+            return None
+    
+    def get_all_camera_states(self) -> Dict[str, Dict[str, Any]]:
+        """Get all camera states for debugging/monitoring."""
+        with self._states_lock:
+            return {
+                cam_id: {
+                    "current_level": state.current_level,
+                    "pending_level": state.pending_level,
+                    "consecutive_count": state.consecutive_count,
+                    "last_published_level": state.last_published_level,
+                    "incident_cycle_id": state.incident_cycle_id,
+                    "empty_frames_count": state.empty_frames_count,
+                    "current_incident_id": state.current_incident_id,
+                    "incident_active": state.incident_active
+                }
+                for cam_id, state in self._incident_states.items()
+            }
+    
+    def get_threshold_config(self, camera_id: str) -> Optional[Dict[str, Any]]:
+        """Get threshold configuration for a camera (for debugging)."""
+        with self._config_lock:
+            config = self._threshold_configs.get(camera_id)
+            if config:
+                return {
+                    "camera_id": config.camera_id,
+                    "application_id": config.application_id,
+                    "app_deployment_id": config.app_deployment_id,
+                    "incident_type": config.incident_type,
+                    "thresholds": config.thresholds,
+                    "last_updated": config.last_updated,
+                    "camera_name": config.camera_name
+                }
+            return None
+    
+    def set_thresholds_for_camera(
+        self, 
+        camera_id: str, 
+        thresholds: List[Dict[str, Any]],
+        application_id: str = "",
+        app_deployment_id: str = "",
+        incident_type: str = "",
+        camera_name: str = ""
+    ):
+        """
+        Manually set thresholds for a camera (useful for testing or direct config).
+        
+        Args:
+            camera_id: Camera identifier
+            thresholds: List of threshold configs
+            application_id: Application ID
+            app_deployment_id: App deployment ID
+            incident_type: Incident type (e.g., "fire")
+            camera_name: Camera name
+        """
+        # Map "high" -> "significant" in thresholds (backend uses "high", we use "significant")
+        mapped_thresholds = []
+        if thresholds:
+            for t in thresholds:
+                if isinstance(t, dict):
+                    level = t.get("level", "").lower().strip()
+                    if level == "high":
+                        t = dict(t)  # Copy to avoid modifying original
+                        t["level"] = "significant"
+                        self.logger.debug(f"[INCIDENT_MANAGER] Mapped threshold level 'high' -> 'significant'")
+                    mapped_thresholds.append(t)
+        
+        with self._config_lock:
+            self._threshold_configs[camera_id] = ThresholdConfig(
+                camera_id=camera_id,
+                application_id=application_id,
+                app_deployment_id=app_deployment_id,
+                incident_type=incident_type,
+                thresholds=mapped_thresholds if mapped_thresholds else DEFAULT_THRESHOLDS.copy(),
+                last_updated=time.time(),
+                camera_name=camera_name
+            )
+        self.logger.info(f"[INCIDENT_MANAGER] Manually set thresholds for camera: {camera_id}")
+
+
+class IncidentManagerFactory:
+    """
+    Factory class for creating INCIDENT_MANAGER instances.
+    
+    Handles session initialization and Redis/Kafka client creation
+    following the same pattern as license_plate_monitoring.py.
+    """
+    
+    ACTION_ID_PATTERN = re.compile(r"^[0-9a-f]{8,}$", re.IGNORECASE)
+    
+    def __init__(self, logger: Optional[logging.Logger] = None):
+        self.logger = logger or logging.getLogger(__name__)
+        self._initialized = False
+        self._incident_manager: Optional[INCIDENT_MANAGER] = None
+        
+        # Store these for later access
+        self._session = None
+        self._action_id: Optional[str] = None
+        self._instance_id: Optional[str] = None
+        self._deployment_id: Optional[str] = None
+        self._app_deployment_id: Optional[str] = None
+        self._application_id: Optional[str] = None  # Store application_id from action_details
+        self._external_ip: Optional[str] = None
+    
+    def initialize(self, config: Any) -> Optional[INCIDENT_MANAGER]:
+        """
+        Initialize and return INCIDENT_MANAGER with Redis/Kafka clients.
+        
+        This follows the same pattern as license_plate_monitoring.py for
+        session initialization and Redis/Kafka client creation.
+        
+        Args:
+            config: Configuration object with session, server_id, etc.
+            
+        Returns:
+            INCIDENT_MANAGER instance or None if initialization failed
+        """
+        if self._initialized and self._incident_manager is not None:
+            self.logger.debug("[INCIDENT_MANAGER_FACTORY] Already initialized, returning existing instance")
+            return self._incident_manager
+        
+        try:
+            # Import required modules
+            from matrice_common.stream.matrice_stream import MatriceStream, StreamType
+            from matrice_common.session import Session
+            
+            self.logger.info("[INCIDENT_MANAGER_FACTORY] ===== STARTING INITIALIZATION =====")
+            
+            # Get or create session
+            self._session = getattr(config, 'session', None)
+            if not self._session:
+                self.logger.info("[INCIDENT_MANAGER_FACTORY] No session in config, creating from environment...")
+                account_number = os.getenv("MATRICE_ACCOUNT_NUMBER", "")
+                access_key_id = os.getenv("MATRICE_ACCESS_KEY_ID", "")
+                secret_key = os.getenv("MATRICE_SECRET_ACCESS_KEY", "")
+                project_id = os.getenv("MATRICE_PROJECT_ID", "")
+                
+                self.logger.debug(f"[INCIDENT_MANAGER_FACTORY] Env vars - account: {'SET' if account_number else 'NOT SET'}, "
+                                  f"access_key: {'SET' if access_key_id else 'NOT SET'}, "
+                                  f"secret: {'SET' if secret_key else 'NOT SET'}")
+                
+                
+                self._session = Session(
+                    account_number=account_number,
+                    access_key=access_key_id,
+                    secret_key=secret_key,
+                    project_id=project_id,
+                )
+                self.logger.info("[INCIDENT_MANAGER_FACTORY] ✓ Created session from environment")
+            else:
+                self.logger.info("[INCIDENT_MANAGER_FACTORY] ✓ Using session from config")
+            
+            rpc = self._session.rpc
+            
+            # Discover action_id
+            self._action_id = self._discover_action_id()
+            if not self._action_id:
+                self.logger.error("[INCIDENT_MANAGER_FACTORY] ❌ Could not discover action_id")
+                print("----- INCIDENT MANAGER ACTION DISCOVERY -----")
+                print("action_id: NOT FOUND")
+                print("---------------------------------------------")
+                self._initialized = True
+                return None
+            
+            self.logger.info(f"[INCIDENT_MANAGER_FACTORY] ✓ Discovered action_id: {self._action_id}")
+            
+            # Fetch action details
+            action_details = {}
+            try:
+                action_url = f"/v1/actions/action/{self._action_id}/details"
+                action_resp = rpc.get(action_url)
+                if not (action_resp and action_resp.get("success", False)):
+                    raise RuntimeError(
+                        action_resp.get("message", "Unknown error") 
+                        if isinstance(action_resp, dict) else "Unknown error"
+                    )
+                action_doc = action_resp.get("data", {}) if isinstance(action_resp, dict) else {}
+                action_details = action_doc.get("actionDetails", {}) if isinstance(action_doc, dict) else {}
+                
+                # IMPORTANT: jobParams contains application_id
+                # Structure: response['data']['jobParams']['application_id']
+                job_params = action_doc.get("jobParams", {}) if isinstance(action_doc, dict) else {}
+                
+                # Extract server details
+                server_id = (
+                    action_details.get("serverId")
+                    or action_details.get("server_id")
+                    or action_details.get("serverID")
+                    or action_details.get("redis_server_id")
+                    or action_details.get("kafka_server_id")
+                )
+                server_type = (
+                    action_details.get("serverType")
+                    or action_details.get("server_type")
+                    or action_details.get("type")
+                )
+                
+                # Store identifiers
+                self._deployment_id = action_details.get("_idDeployment") or action_details.get("deployment_id")
+                
+                # app_deployment_id: check actionDetails first, then jobParams
+                self._app_deployment_id = (
+                    action_details.get("app_deployment_id") or 
+                    action_details.get("appDeploymentId") or
+                    action_details.get("app_deploymentId") or
+                    job_params.get("app_deployment_id") or
+                    job_params.get("appDeploymentId") or
+                    job_params.get("app_deploymentId") or
+                    ""
+                )
+                
+                # application_id: PRIMARILY from jobParams (this is where it lives!)
+                # response['data']['jobParams'].get('application_id', '')
+                self._application_id = (
+                    job_params.get("application_id") or
+                    job_params.get("applicationId") or
+                    job_params.get("app_id") or
+                    job_params.get("appId") or
+                    action_details.get("application_id") or
+                    action_details.get("applicationId") or
+                    ""
+                )
+                
+                self._instance_id = action_details.get("instanceID") or action_details.get("instanceId")
+                self._external_ip = action_details.get("externalIP") or action_details.get("externalIp")
+                
+                print("----- INCIDENT MANAGER ACTION DETAILS -----")
+                print(f"action_id: {self._action_id}")
+                print(f"server_type: {server_type}")
+                print(f"server_id: {server_id}")
+                print(f"deployment_id: {self._deployment_id}")
+                print(f"app_deployment_id: {self._app_deployment_id}")
+                print(f"application_id: {self._application_id}")
+                print(f"instance_id: {self._instance_id}")
+                print(f"external_ip: {self._external_ip}")
+                print(f"jobParams keys: {list(job_params.keys()) if job_params else []}")
+                print("--------------------------------------------")
+                
+                self.logger.info(
+                    f"[INCIDENT_MANAGER_FACTORY] Action details - server_type={server_type}, "
+                    f"instance_id={self._instance_id}, "
+                    f"app_deployment_id={self._app_deployment_id}, application_id={self._application_id}"
+                )
+                
+                # Log all available keys for debugging
+                self.logger.debug(f"[INCIDENT_MANAGER_FACTORY] actionDetails keys: {list(action_details.keys())}")
+                self.logger.debug(f"[INCIDENT_MANAGER_FACTORY] jobParams keys: {list(job_params.keys()) if job_params else []}")
+                
+            except Exception as e:
+                self.logger.error(f"[INCIDENT_MANAGER_FACTORY] ❌ Failed to fetch action details: {e}", exc_info=True)
+                print("----- INCIDENT MANAGER ACTION DETAILS ERROR -----")
+                print(f"action_id: {self._action_id}")
+                print(f"error: {e}")
+                print("-------------------------------------------------")
+                self._initialized = True
+                return None
+            
+            # Determine localhost vs cloud using externalIP from action_details
+            is_localhost = False
+            public_ip = self._get_public_ip()
+            
+            # Get server host from action_details (user's method - no dependency on server_id)
+            server_host = (
+                action_details.get("externalIP")
+                or action_details.get("external_IP")
+                or action_details.get("externalip")
+                or action_details.get("external_ip")
+                or action_details.get("externalIp")
+                or action_details.get("external_Ip")
+            )
+            print(f"server_host: {server_host}")
+            self.logger.info(f"[INCIDENT_MANAGER_FACTORY] DEBUG - server_host: {server_host}")
+            
+            localhost_indicators = ["localhost", "127.0.0.1", "0.0.0.0"]
+            if server_host in localhost_indicators:
+                is_localhost = True
+                self.logger.info(
+                    f"[INCIDENT_MANAGER_FACTORY] Detected Localhost environment "
+                    f"(Public IP={public_ip}, Server IP={server_host})"
+                )
+            else:
+                is_localhost = False
+                self.logger.info(
+                    f"[INCIDENT_MANAGER_FACTORY] Detected Cloud environment "
+                    f"(Public IP={public_ip}, Server IP={server_host})"
+                )
+            
+            redis_client = None
+            kafka_client = None
+            
+            # STRICT SWITCH: Only Redis if localhost, Only Kafka if cloud
+            if is_localhost:
+                # Initialize Redis client (ONLY) using instance_id
+                if not self._instance_id:
+                    self.logger.error("[INCIDENT_MANAGER_FACTORY] ❌ Localhost mode but instance_id missing")
+                else:
+                    try:
+                        url = f"/v1/actions/get_redis_server_by_instance_id/{self._instance_id}"
+                        self.logger.info(f"[INCIDENT_MANAGER_FACTORY] Fetching Redis server info for instance: {self._instance_id}")
+                        response = rpc.get(url)
+                        
+                        if isinstance(response, dict) and response.get("success", False):
+                            data = response.get("data", {})
+                            host = data.get("host")
+                            port = data.get("port")
+                            username = data.get("username")
+                            password = data.get("password", "")
+                            db_index = data.get("db", 0)
+                            conn_timeout = data.get("connection_timeout", 120)
+                            
+                            print("----- INCIDENT MANAGER REDIS SERVER PARAMS -----")
+                            print(f"instance_id: {self._instance_id}")
+                            print(f"host: {host}")
+                            print(f"port: {port}")
+                            print(f"username: {username}")
+                            print(f"password: {'*' * len(password) if password else ''}")
+                            print(f"db: {db_index}")
+                            print(f"connection_timeout: {conn_timeout}")
+                            print("------------------------------------------------")
+                            
+                            self.logger.info(
+                                f"[INCIDENT_MANAGER_FACTORY] Redis params - host={host}, port={port}, user={username}"
+                            )
+                            
+                            redis_client = MatriceStream(
+                                StreamType.REDIS,
+                                host=host,
+                                port=int(port),
+                                password=password,
+                                username=username,
+                                db=db_index,
+                                connection_timeout=conn_timeout
+                            )
+                            # Setup for both config polling and incident publishing
+                            redis_client.setup("incident_modification_config")
+                            self.logger.info("[INCIDENT_MANAGER_FACTORY] ✓ Redis client initialized")
+                        else:
+                            self.logger.warning(
+                                f"[INCIDENT_MANAGER_FACTORY] Failed to fetch Redis server info: "
+                                f"{response.get('message', 'Unknown error') if isinstance(response, dict) else 'Unknown error'}"
+                            )
+                    except Exception as e:
+                        self.logger.warning(f"[INCIDENT_MANAGER_FACTORY] Redis initialization failed: {e}")
+            
+            else:
+                # Initialize Kafka client (ONLY) using global info endpoint
+                try:
+                    url = f"/v1/actions/get_kafka_info"
+                    self.logger.info("[INCIDENT_MANAGER_FACTORY] Fetching Kafka server info for Cloud mode")
+                    response = rpc.get(url)
+                    
+                    if isinstance(response, dict) and response.get("success", False):
+                        data = response.get("data", {})
+                        enc_ip = data.get("ip")
+                        enc_port = data.get("port")
+                        
+                        # Decode base64 encoded values
+                        ip_addr = None
+                        port = None
+                        try:
+                            ip_addr = base64.b64decode(str(enc_ip)).decode("utf-8")
+                        except Exception:
+                            ip_addr = enc_ip
+                        try:
+                            port = base64.b64decode(str(enc_port)).decode("utf-8")
+                        except Exception:
+                            port = enc_port
+                        
+                        print("----- INCIDENT MANAGER KAFKA SERVER PARAMS -----")
+                        print(f"ipAddress: {ip_addr}")
+                        print(f"port: {port}")
+                        print("------------------------------------------------")
+                        
+                        self.logger.info(f"[INCIDENT_MANAGER_FACTORY] Kafka params - ip={ip_addr}, port={port}")
+                        
+                        bootstrap_servers = f"{ip_addr}:{port}"
+                        kafka_client = MatriceStream(
+                            StreamType.KAFKA,
+                            bootstrap_servers=bootstrap_servers,
+                            sasl_mechanism="SCRAM-SHA-256",
+                            sasl_username="matrice-sdk-user",
+                            sasl_password="matrice-sdk-password",
+                            security_protocol="SASL_PLAINTEXT"
+                        )
+                        # Setup for both config polling and incident publishing
+                        kafka_client.setup("incident_modification_config", consumer_group_id="py_analytics_incidents")
+                        self.logger.info(f"[INCIDENT_MANAGER_FACTORY] ✓ Kafka client initialized (servers={bootstrap_servers})")
+                    else:
+                        self.logger.warning(
+                            f"[INCIDENT_MANAGER_FACTORY] Failed to fetch Kafka server info: "
+                            f"{response.get('message', 'Unknown error') if isinstance(response, dict) else 'Unknown error'}"
+                        )
+                except Exception as e:
+                    self.logger.warning(f"[INCIDENT_MANAGER_FACTORY] Kafka initialization failed: {e}")
+            
+            # Create incident manager if we have at least one transport
+            if redis_client or kafka_client:
+                self._incident_manager = INCIDENT_MANAGER(
+                    redis_client=redis_client,
+                    kafka_client=kafka_client,
+                    incident_topic="incident_res",
+                    config_topic="incident_modification_config",
+                    logger=self.logger
+                )
+                # Set factory reference for accessing deployment info
+                self._incident_manager.set_factory_ref(self)
+                # Start the config polling thread
+                self._incident_manager.start()
+                
+                transport = "Redis" if redis_client else "Kafka"
+                self.logger.info(f"[INCIDENT_MANAGER_FACTORY] ✓ Incident manager created with {transport}")
+                print(f"----- INCIDENT MANAGER INITIALIZED ({transport}) -----")
+            else:
+                self.logger.warning(
+                    f"[INCIDENT_MANAGER_FACTORY] No {'Redis' if is_localhost else 'Kafka'} client available, "
+                    f"incident manager not created"
+                )
+            
+            self._initialized = True
+            self.logger.info("[INCIDENT_MANAGER_FACTORY] ===== INITIALIZATION COMPLETE =====")
+            return self._incident_manager
+            
+        except ImportError as e:
+            self.logger.error(f"[INCIDENT_MANAGER_FACTORY] Import error: {e}")
+            self._initialized = True
+            return None
+        except Exception as e:
+            self.logger.error(f"[INCIDENT_MANAGER_FACTORY] Initialization failed: {e}", exc_info=True)
+            self._initialized = True
+            return None
+    
+    def _discover_action_id(self) -> Optional[str]:
+        """Discover action_id from current working directory name (and parents)."""
+        try:
+            candidates: List[str] = []
+            
+            try:
+                cwd = Path.cwd()
+                candidates.append(cwd.name)
+                for parent in cwd.parents:
+                    candidates.append(parent.name)
+            except Exception:
+                pass
+            
+            try:
+                usr_src = Path("/usr/src")
+                if usr_src.exists():
+                    for child in usr_src.iterdir():
+                        if child.is_dir():
+                            candidates.append(child.name)
+            except Exception:
+                pass
+            
+            for candidate in candidates:
+                if candidate and len(candidate) >= 8 and self.ACTION_ID_PATTERN.match(candidate):
+                    return candidate
+        except Exception:
+            pass
+        return None
+    
+    def _get_public_ip(self) -> str:
+        """Get the public IP address of this machine."""
+        self.logger.info("[INCIDENT_MANAGER_FACTORY] Fetching public IP address...")
+        try:
+            public_ip = urllib.request.urlopen(
+                "https://v4.ident.me", timeout=120
+            ).read().decode("utf8").strip()
+            self.logger.debug(f"[INCIDENT_MANAGER_FACTORY] Public IP: {public_ip}")
+            return public_ip
+        except Exception as e:
+            self.logger.warning(f"[INCIDENT_MANAGER_FACTORY] Error fetching public IP: {e}")
+            return "localhost"
+    
+    def _get_backend_base_url(self) -> str:
+        """Resolve backend base URL based on ENV variable."""
+        env = os.getenv("ENV", "prod").strip().lower()
+        if env in ("prod", "production"):
+            host = "prod.backend.app.matrice.ai"
+        elif env in ("dev", "development"):
+            host = "dev.backend.app.matrice.ai"
+        else:
+            host = "staging.backend.app.matrice.ai"
+        return f"https://{host}"
+    
+    @property
+    def is_initialized(self) -> bool:
+        return self._initialized
+    
+    @property
+    def incident_manager(self) -> Optional[INCIDENT_MANAGER]:
+        return self._incident_manager
+
+
+# Module-level factory instance for convenience
+_default_factory: Optional[IncidentManagerFactory] = None
+
+
+def get_incident_manager(config: Any, logger: Optional[logging.Logger] = None) -> Optional[INCIDENT_MANAGER]:
+    """
+    Get or create INCIDENT_MANAGER instance.
+    
+    This is a convenience function that uses a module-level factory.
+    For more control, use IncidentManagerFactory directly.
+    
+    Args:
+        config: Configuration object with session, server_id, etc.
+        logger: Logger instance
+        
+    Returns:
+        INCIDENT_MANAGER instance or None
+    """
+    global _default_factory
+    
+    if _default_factory is None:
+        _default_factory = IncidentManagerFactory(logger=logger)
+    
+    return _default_factory.initialize(config)