frigate/frigate/app.py

738 lines
26 KiB
Python
Raw Normal View History

import datetime
2020-11-04 15:28:07 +03:00
import logging
import multiprocessing as mp
import os
import secrets
import shutil
from multiprocessing import Queue
Use Fork-Server As Spawn Method (#18682) * Set runtime * Use count correctly * Don't assume camera sizes * Use separate zmq proxy for object detection * Correct order * Use forkserver * Only store PID instead of entire process reference * Cleanup * Catch correct errors * Fix typing * Remove before_run from process util The before_run never actually ran because: You're right to suspect an issue with before_run not being called and a potential deadlock. The way you've implemented the run_wrapper using __getattribute__ for the run method of BaseProcess is a common pitfall in Python's multiprocessing, especially when combined with how multiprocessing.Process works internally. Here's a breakdown of why before_run isn't being called and why you might be experiencing a deadlock: The Problem: __getattribute__ and Process Serialization When you create a multiprocessing.Process object and call start(), the multiprocessing module needs to serialize the process object (or at least enough of it to re-create the process in the new interpreter). It then pickles this serialized object and sends it to the newly spawned process. The issue with your __getattribute__ implementation for run is that: run is retrieved during serialization: When multiprocessing tries to pickle your Process object to send to the new process, it will likely access the run attribute. This triggers your __getattribute__ wrapper, which then tries to bind run_wrapper to self. run_wrapper is bound to the parent process's self: The run_wrapper closure, when created in the parent process, captures the self (the Process instance) from the parent's memory space. Deserialization creates a new object: In the child process, a new Process object is created by deserializing the pickled data. However, the run_wrapper method that was pickled still holds a reference to the self from the parent process. This is a subtle but critical distinction. The child's run is not your wrapped run: When the child process starts, it internally calls its own run method. Because of the serialization and deserialization process, the run method that's ultimately executed in the child process is the original multiprocessing.Process.run or the Process.run if you had directly overridden it. Your __getattribute__ magic, which wraps run, isn't correctly applied to the Process object within the child's context. * Cleanup * Logging bugfix (#18465) * use mp Manager to handle logging queues A Python bug (https://github.com/python/cpython/issues/91555) was preventing logs from the embeddings maintainer process from printing. The bug is fixed in Python 3.14, but a viable workaround is to use the multiprocessing Manager, which better manages mp queues and causes the logging to work correctly. * consolidate * fix typing * Fix typing * Use global log queue * Move to using process for logging * Convert camera tracking to process * Add more processes * Finalize process * Cleanup * Cleanup typing * Formatting * Remove daemon --------- Co-authored-by: Josh Hawkins <32435876+hawkeye217@users.noreply.github.com>
2025-06-12 21:12:34 +03:00
from multiprocessing.managers import DictProxy, SyncManager
from multiprocessing.synchronize import Event as MpEvent
2025-05-30 05:58:31 +03:00
from pathlib import Path
from typing import Callable, Optional
2020-11-04 15:28:07 +03:00
import psutil
Frigate HTTP API using FastAPI (#13871) * POC: Added FastAPI with one endpoint (get /logs/service) * POC: Revert error_log * POC: Converted preview related endpoints to FastAPI * POC: Converted two more endpoints to FastAPI * POC: lint * Convert all media endpoints to FastAPI. Added /media prefix (/media/camera && media/events && /media/preview) * Convert all notifications API endpoints to FastAPI * Convert first review API endpoints to FastAPI * Convert remaining review API endpoints to FastAPI * Convert export endpoints to FastAPI * Fix path parameters * Convert events endpoints to FastAPI * Use body for multiple events endpoints * Use body for multiple events endpoints (create and end event) * Convert app endpoints to FastAPI * Convert app endpoints to FastAPI * Convert auth endpoints to FastAPI * Removed flask app in favour of FastAPI app. Implemented FastAPI middleware to check CSRF, connect and disconnect from DB. Added middleware x-forwared-for headers * Added starlette plugin to expose custom headers * Use slowapi as the limiter * Use query parameters for the frame latest endpoint * Use query parameters for the media snapshot.jpg endpoint * Use query parameters for the media MJPEG feed endpoint * Revert initial nginx.conf change * Added missing even_id for /events/search endpoint * Removed left over comment * Use FastAPI TestClient * severity query parameter should be a string * Use the same pattern for all tests * Fix endpoint * Revert media routers to old names. Order routes to make sure the dynamic ones from media.py are only used whenever there's no match on auth/etc * Reverted paths for media on tsx files * Deleted file * Fix test_http to use TestClient * Formatting * Bind timeline to DB * Fix http tests * Replace filename with pathvalidate * Fix latest.ext handling and disable uvicorn access logs * Add cosntraints to api provided values * Formatting * Remove unused * Remove unused * Get rate limiter working --------- Co-authored-by: Nicolas Mowen <nickmowen213@gmail.com>
2024-09-24 16:05:30 +03:00
import uvicorn
2020-12-24 16:47:27 +03:00
from peewee_migrate import Router
2020-11-04 15:28:07 +03:00
from playhouse.sqlite_ext import SqliteExtDatabase
from frigate.api.auth import hash_password
Frigate HTTP API using FastAPI (#13871) * POC: Added FastAPI with one endpoint (get /logs/service) * POC: Revert error_log * POC: Converted preview related endpoints to FastAPI * POC: Converted two more endpoints to FastAPI * POC: lint * Convert all media endpoints to FastAPI. Added /media prefix (/media/camera && media/events && /media/preview) * Convert all notifications API endpoints to FastAPI * Convert first review API endpoints to FastAPI * Convert remaining review API endpoints to FastAPI * Convert export endpoints to FastAPI * Fix path parameters * Convert events endpoints to FastAPI * Use body for multiple events endpoints * Use body for multiple events endpoints (create and end event) * Convert app endpoints to FastAPI * Convert app endpoints to FastAPI * Convert auth endpoints to FastAPI * Removed flask app in favour of FastAPI app. Implemented FastAPI middleware to check CSRF, connect and disconnect from DB. Added middleware x-forwared-for headers * Added starlette plugin to expose custom headers * Use slowapi as the limiter * Use query parameters for the frame latest endpoint * Use query parameters for the media snapshot.jpg endpoint * Use query parameters for the media MJPEG feed endpoint * Revert initial nginx.conf change * Added missing even_id for /events/search endpoint * Removed left over comment * Use FastAPI TestClient * severity query parameter should be a string * Use the same pattern for all tests * Fix endpoint * Revert media routers to old names. Order routes to make sure the dynamic ones from media.py are only used whenever there's no match on auth/etc * Reverted paths for media on tsx files * Deleted file * Fix test_http to use TestClient * Formatting * Bind timeline to DB * Fix http tests * Replace filename with pathvalidate * Fix latest.ext handling and disable uvicorn access logs * Add cosntraints to api provided values * Formatting * Remove unused * Remove unused * Get rate limiter working --------- Co-authored-by: Nicolas Mowen <nickmowen213@gmail.com>
2024-09-24 16:05:30 +03:00
from frigate.api.fastapi_app import create_fastapi_app
from frigate.camera import CameraMetrics, PTZMetrics
from frigate.camera.maintainer import CameraMaintainer
from frigate.comms.base_communicator import Communicator
from frigate.comms.dispatcher import Dispatcher
from frigate.comms.event_metadata_updater import EventMetadataPublisher
from frigate.comms.inter_process import InterProcessCommunicator
from frigate.comms.mqtt import MqttClient
Use Fork-Server As Spawn Method (#18682) * Set runtime * Use count correctly * Don't assume camera sizes * Use separate zmq proxy for object detection * Correct order * Use forkserver * Only store PID instead of entire process reference * Cleanup * Catch correct errors * Fix typing * Remove before_run from process util The before_run never actually ran because: You're right to suspect an issue with before_run not being called and a potential deadlock. The way you've implemented the run_wrapper using __getattribute__ for the run method of BaseProcess is a common pitfall in Python's multiprocessing, especially when combined with how multiprocessing.Process works internally. Here's a breakdown of why before_run isn't being called and why you might be experiencing a deadlock: The Problem: __getattribute__ and Process Serialization When you create a multiprocessing.Process object and call start(), the multiprocessing module needs to serialize the process object (or at least enough of it to re-create the process in the new interpreter). It then pickles this serialized object and sends it to the newly spawned process. The issue with your __getattribute__ implementation for run is that: run is retrieved during serialization: When multiprocessing tries to pickle your Process object to send to the new process, it will likely access the run attribute. This triggers your __getattribute__ wrapper, which then tries to bind run_wrapper to self. run_wrapper is bound to the parent process's self: The run_wrapper closure, when created in the parent process, captures the self (the Process instance) from the parent's memory space. Deserialization creates a new object: In the child process, a new Process object is created by deserializing the pickled data. However, the run_wrapper method that was pickled still holds a reference to the self from the parent process. This is a subtle but critical distinction. The child's run is not your wrapped run: When the child process starts, it internally calls its own run method. Because of the serialization and deserialization process, the run method that's ultimately executed in the child process is the original multiprocessing.Process.run or the Process.run if you had directly overridden it. Your __getattribute__ magic, which wraps run, isn't correctly applied to the Process object within the child's context. * Cleanup * Logging bugfix (#18465) * use mp Manager to handle logging queues A Python bug (https://github.com/python/cpython/issues/91555) was preventing logs from the embeddings maintainer process from printing. The bug is fixed in Python 3.14, but a viable workaround is to use the multiprocessing Manager, which better manages mp queues and causes the logging to work correctly. * consolidate * fix typing * Fix typing * Use global log queue * Move to using process for logging * Convert camera tracking to process * Add more processes * Finalize process * Cleanup * Cleanup typing * Formatting * Remove daemon --------- Co-authored-by: Josh Hawkins <32435876+hawkeye217@users.noreply.github.com>
2025-06-12 21:12:34 +03:00
from frigate.comms.object_detector_signaler import DetectorProxy
from frigate.comms.webpush import WebPushClient
from frigate.comms.ws import WebSocketClient
from frigate.comms.zmq_proxy import ZmqProxy
from frigate.config.camera.updater import CameraConfigUpdatePublisher
from frigate.config.config import FrigateConfig
Camera profile support (#22482) * add CameraProfileConfig model for named config overrides * add profiles field to CameraConfig * add active_profile field to FrigateConfig Runtime-only field excluded from YAML serialization, tracks which profile is currently active. * add ProfileManager for profile activation and persistence Handles snapshotting base configs, applying profile overrides via deep_merge + apply_section_update, publishing ZMQ updates, and persisting active profile to /config/.active_profile. * add profile API endpoints (GET /profiles, GET/PUT /profile) * add MQTT and dispatcher integration for profiles - Subscribe to frigate/profile/set MQTT topic - Publish profile/state and profiles/available on connect - Add _on_profile_command handler to dispatcher - Broadcast active profile state on WebSocket connect * wire ProfileManager into app startup and FastAPI - Create ProfileManager after dispatcher init - Restore persisted profile on startup - Pass dispatcher and profile_manager to FastAPI app * add tests for invalid profile values and keys Tests that Pydantic rejects: invalid field values (fps: "not_a_number"), unknown section keys (ffmpeg in profile), invalid nested values, and invalid profiles in full config parsing. * formatting * fix CameraLiveConfig JSON serialization error on profile activation refactor _publish_updates to only publish ZMQ updates for sections that actually changed, not all sections on affected cameras. * consolidate * add enabled field to camera profiles for enabling/disabling cameras * add zones support to camera profiles * add frontend profile types, color utility, and config save support * add profile state management and save preview support * add profileName prop to BaseSection for profile-aware config editing * add profile section dropdown and wire into camera settings pages * add per-profile camera enable/disable to Camera Management view * add profiles summary page with card-based layout and fix backend zone comparison bug * add active profile badge to settings toolbar * i18n * add red dot for any pending changes including profiles * profile support for mask and zone editor * fix hidden field validation errors caused by lodash wildcard and schema gaps lodash unset does not support wildcard (*) segments, so hidden fields like filters.*.mask were never stripped from form data, leaving null raw_coordinates that fail RJSF anyOf validation. Add unsetWithWildcard helper and also strip hidden fields from the JSON schema itself as defense-in-depth. * add face_recognition and lpr to profile-eligible sections * move profile dropdown from section panes to settings header * add profiles enable toggle and improve empty state * formatting * tweaks * tweak colors and switch * fix profile save diff, masksAndZones delete, and config sync * ui tweaks * ensure profile manager gets updated config * rename profile settings to ui settings * refactor profilesview and add dots/border colors when overridden * implement an update_config method for profile manager * fix mask deletion * more unique colors * add top-level profiles config section with friendly names * implement profile friendly names and improve profile UI - Add ProfileDefinitionConfig type and profiles field to FrigateConfig - Use ProfilesApiResponse type with friendly_name support throughout - Replace Record<string, unknown> with proper JsonObject/JsonValue types - Add profile creation form matching zone pattern (Zod + NameAndIdFields) - Add pencil icon for renaming profile friendly names in ProfilesView - Move Profiles menu item to first under Camera Configuration - Add activity indicators on save/rename/delete buttons - Display friendly names in CameraManagementView profile selector - Fix duplicate colored dots in management profile dropdown - Fix i18n namespace for overridden base config tooltips - Move profile override deletion from dropdown trash icon to footer button with confirmation dialog, matching Reset to Global pattern - Remove Add Profile from section header dropdown to prevent saving camera overrides before top-level profile definition exists - Clean up newProfiles state after API profile deletion - Refresh profiles SWR cache after saving profile definitions * remove profile badge in settings and add profiles to main menu * use icon only on mobile * change color order * docs * show activity indicator on trash icon while deleting a profile * tweak language * immediately create profiles on backend instead of deferring to Save All * hide restart-required fields when editing a profile section fields that require a restart cannot take effect via profile switching, so they are merged into hiddenFields when profileName is set * show active profile indicator in desktop status bar * fix profile config inheritance bug where Pydantic defaults override base values The /config API was dumping profile overrides with model_dump() which included all Pydantic defaults. When the frontend merged these over the camera's base config, explicitly-set base values were lost. Now profile overrides are re-dumped with exclude_unset=True so only user-specified fields are returned. Also fixes the Save All path generating spurious deletion markers for restart-required fields that are hidden during profile editing but not excluded from the raw data sanitization in prepareSectionSavePayload. * docs tweaks * docs tweak * formatting * formatting * fix typing * fix test pollution test_maintainer was injecting MagicMock() into sys.modules["frigate.config.camera.updater"] at module load time and never restoring it. When the profile tests later imported CameraConfigUpdateEnum and CameraConfigUpdateTopic from that module, they got mock objects instead of the real dataclass/enum, so equality comparisons always failed * remove * fix settings showing profile-merged values when editing base config When a profile is active, the in-memory config contains effective (profile-merged) values. The settings UI was displaying these merged values even when the "Base Config" view was selected. Backend: snapshot pre-profile base configs in ProfileManager and expose them via a `base_config` key in the /api/config camera response when a profile is active. The top-level sections continue to reflect the effective running config. Frontend: read from `base_config` when available in BaseSection, useConfigOverride, useAllCameraOverrides, and prepareSectionSavePayload. Include formData labels in Object/Audio switches widgets so that labels added only by a profile override remain visible when editing that profile. * use rasterized_mask as field makes it easier to exclude from the schema with exclude=True prevents leaking of the field when using model_dump for profiles * fix zones - Fix zone colors not matching across profiles by falling back to base zone color when profile zone data lacks a color field - Use base_config for base-layer values in masks/zones view so profile-merged values don't pollute the base config editing view - Handle zones separately in profile manager snapshot/restore since ZoneConfig requires special serialization (color as private attr, contour generation) - Inherit base zone color and generate contours for profile zone overrides in profile manager * formatting * don't require restart for camera enabled change for profiles * publish camera state when changing profiles * formatting * remove available profiles from mqtt * improve typing
2026-03-19 17:47:57 +03:00
from frigate.config.profile_manager import ProfileManager
from frigate.const import (
CACHE_DIR,
CLIPS_DIR,
CONFIG_DIR,
EXPORT_DIR,
FACE_DIR,
MODEL_CACHE_DIR,
RECORD_DIR,
THUMB_DIR,
TRIGGER_DIR,
)
from frigate.data_processing.types import DataProcessorMetrics
Use sqlite-vec extension instead of chromadb for embeddings (#14163) * swap sqlite_vec for chroma in requirements * load sqlite_vec in embeddings manager * remove chroma and revamp Embeddings class for sqlite_vec * manual minilm onnx inference * remove chroma in clip model * migrate api from chroma to sqlite_vec * migrate event cleanup from chroma to sqlite_vec * migrate embedding maintainer from chroma to sqlite_vec * genai description for sqlite_vec * load sqlite_vec in main thread db * extend the SqliteQueueDatabase class and use peewee db.execute_sql * search with Event type for similarity * fix similarity search * install and add comment about transformers * fix normalization * add id filter * clean up * clean up * fully remove chroma and add transformers env var * readd uvicorn for fastapi * readd tokenizer parallelism env var * remove chroma from docs * remove chroma from UI * try removing custom pysqlite3 build * hard code limit * optimize queries * revert explore query * fix query * keep building pysqlite3 * single pass fetch and process * remove unnecessary re-embed * update deps * move SqliteVecQueueDatabase to db directory * make search thumbnail take up full size of results box * improve typing * improve model downloading and add status screen * daemon downloading thread * catch case when semantic search is disabled * fix typing * build sqlite_vec from source * resolve conflict * file permissions * try build deps * remove sources * sources * fix thread start * include git in build * reorder embeddings after detectors are started * build with sqlite amalgamation * non-platform specific * use wget instead of curl * remove unzip -d * remove sqlite_vec from requirements and load the compiled version * fix build * avoid race in db connection * add scale_factor and bias to description zscore normalization
2024-10-07 23:30:45 +03:00
from frigate.db.sqlitevecq import SqliteVecQueueDatabase
from frigate.debug_replay import (
DebugReplayManager,
cleanup_replay_cameras,
)
Use Fork-Server As Spawn Method (#18682) * Set runtime * Use count correctly * Don't assume camera sizes * Use separate zmq proxy for object detection * Correct order * Use forkserver * Only store PID instead of entire process reference * Cleanup * Catch correct errors * Fix typing * Remove before_run from process util The before_run never actually ran because: You're right to suspect an issue with before_run not being called and a potential deadlock. The way you've implemented the run_wrapper using __getattribute__ for the run method of BaseProcess is a common pitfall in Python's multiprocessing, especially when combined with how multiprocessing.Process works internally. Here's a breakdown of why before_run isn't being called and why you might be experiencing a deadlock: The Problem: __getattribute__ and Process Serialization When you create a multiprocessing.Process object and call start(), the multiprocessing module needs to serialize the process object (or at least enough of it to re-create the process in the new interpreter). It then pickles this serialized object and sends it to the newly spawned process. The issue with your __getattribute__ implementation for run is that: run is retrieved during serialization: When multiprocessing tries to pickle your Process object to send to the new process, it will likely access the run attribute. This triggers your __getattribute__ wrapper, which then tries to bind run_wrapper to self. run_wrapper is bound to the parent process's self: The run_wrapper closure, when created in the parent process, captures the self (the Process instance) from the parent's memory space. Deserialization creates a new object: In the child process, a new Process object is created by deserializing the pickled data. However, the run_wrapper method that was pickled still holds a reference to the self from the parent process. This is a subtle but critical distinction. The child's run is not your wrapped run: When the child process starts, it internally calls its own run method. Because of the serialization and deserialization process, the run method that's ultimately executed in the child process is the original multiprocessing.Process.run or the Process.run if you had directly overridden it. Your __getattribute__ magic, which wraps run, isn't correctly applied to the Process object within the child's context. * Cleanup * Logging bugfix (#18465) * use mp Manager to handle logging queues A Python bug (https://github.com/python/cpython/issues/91555) was preventing logs from the embeddings maintainer process from printing. The bug is fixed in Python 3.14, but a viable workaround is to use the multiprocessing Manager, which better manages mp queues and causes the logging to work correctly. * consolidate * fix typing * Fix typing * Use global log queue * Move to using process for logging * Convert camera tracking to process * Add more processes * Finalize process * Cleanup * Cleanup typing * Formatting * Remove daemon --------- Co-authored-by: Josh Hawkins <32435876+hawkeye217@users.noreply.github.com>
2025-06-12 21:12:34 +03:00
from frigate.embeddings import EmbeddingProcess, EmbeddingsContext
from frigate.events.audio import AudioProcessor
from frigate.events.cleanup import EventCleanup
from frigate.events.maintainer import EventProcessor
from frigate.jobs.motion_search import stop_all_motion_search_jobs
from frigate.log import _stop_logging
from frigate.models import (
Event,
Export,
Previews,
Recordings,
RecordingsToDelete,
Regions,
ReviewSegment,
Timeline,
Trigger,
User,
)
from frigate.object_detection.base import ObjectDetectProcess
Use Fork-Server As Spawn Method (#18682) * Set runtime * Use count correctly * Don't assume camera sizes * Use separate zmq proxy for object detection * Correct order * Use forkserver * Only store PID instead of entire process reference * Cleanup * Catch correct errors * Fix typing * Remove before_run from process util The before_run never actually ran because: You're right to suspect an issue with before_run not being called and a potential deadlock. The way you've implemented the run_wrapper using __getattribute__ for the run method of BaseProcess is a common pitfall in Python's multiprocessing, especially when combined with how multiprocessing.Process works internally. Here's a breakdown of why before_run isn't being called and why you might be experiencing a deadlock: The Problem: __getattribute__ and Process Serialization When you create a multiprocessing.Process object and call start(), the multiprocessing module needs to serialize the process object (or at least enough of it to re-create the process in the new interpreter). It then pickles this serialized object and sends it to the newly spawned process. The issue with your __getattribute__ implementation for run is that: run is retrieved during serialization: When multiprocessing tries to pickle your Process object to send to the new process, it will likely access the run attribute. This triggers your __getattribute__ wrapper, which then tries to bind run_wrapper to self. run_wrapper is bound to the parent process's self: The run_wrapper closure, when created in the parent process, captures the self (the Process instance) from the parent's memory space. Deserialization creates a new object: In the child process, a new Process object is created by deserializing the pickled data. However, the run_wrapper method that was pickled still holds a reference to the self from the parent process. This is a subtle but critical distinction. The child's run is not your wrapped run: When the child process starts, it internally calls its own run method. Because of the serialization and deserialization process, the run method that's ultimately executed in the child process is the original multiprocessing.Process.run or the Process.run if you had directly overridden it. Your __getattribute__ magic, which wraps run, isn't correctly applied to the Process object within the child's context. * Cleanup * Logging bugfix (#18465) * use mp Manager to handle logging queues A Python bug (https://github.com/python/cpython/issues/91555) was preventing logs from the embeddings maintainer process from printing. The bug is fixed in Python 3.14, but a viable workaround is to use the multiprocessing Manager, which better manages mp queues and causes the logging to work correctly. * consolidate * fix typing * Fix typing * Use global log queue * Move to using process for logging * Convert camera tracking to process * Add more processes * Finalize process * Cleanup * Cleanup typing * Formatting * Remove daemon --------- Co-authored-by: Josh Hawkins <32435876+hawkeye217@users.noreply.github.com>
2025-06-12 21:12:34 +03:00
from frigate.output.output import OutputProcess
from frigate.ptz.autotrack import PtzAutoTrackerThread
from frigate.ptz.onvif import OnvifController
from frigate.record.cleanup import RecordingCleanup
from frigate.record.export import migrate_exports
Use Fork-Server As Spawn Method (#18682) * Set runtime * Use count correctly * Don't assume camera sizes * Use separate zmq proxy for object detection * Correct order * Use forkserver * Only store PID instead of entire process reference * Cleanup * Catch correct errors * Fix typing * Remove before_run from process util The before_run never actually ran because: You're right to suspect an issue with before_run not being called and a potential deadlock. The way you've implemented the run_wrapper using __getattribute__ for the run method of BaseProcess is a common pitfall in Python's multiprocessing, especially when combined with how multiprocessing.Process works internally. Here's a breakdown of why before_run isn't being called and why you might be experiencing a deadlock: The Problem: __getattribute__ and Process Serialization When you create a multiprocessing.Process object and call start(), the multiprocessing module needs to serialize the process object (or at least enough of it to re-create the process in the new interpreter). It then pickles this serialized object and sends it to the newly spawned process. The issue with your __getattribute__ implementation for run is that: run is retrieved during serialization: When multiprocessing tries to pickle your Process object to send to the new process, it will likely access the run attribute. This triggers your __getattribute__ wrapper, which then tries to bind run_wrapper to self. run_wrapper is bound to the parent process's self: The run_wrapper closure, when created in the parent process, captures the self (the Process instance) from the parent's memory space. Deserialization creates a new object: In the child process, a new Process object is created by deserializing the pickled data. However, the run_wrapper method that was pickled still holds a reference to the self from the parent process. This is a subtle but critical distinction. The child's run is not your wrapped run: When the child process starts, it internally calls its own run method. Because of the serialization and deserialization process, the run method that's ultimately executed in the child process is the original multiprocessing.Process.run or the Process.run if you had directly overridden it. Your __getattribute__ magic, which wraps run, isn't correctly applied to the Process object within the child's context. * Cleanup * Logging bugfix (#18465) * use mp Manager to handle logging queues A Python bug (https://github.com/python/cpython/issues/91555) was preventing logs from the embeddings maintainer process from printing. The bug is fixed in Python 3.14, but a viable workaround is to use the multiprocessing Manager, which better manages mp queues and causes the logging to work correctly. * consolidate * fix typing * Fix typing * Use global log queue * Move to using process for logging * Convert camera tracking to process * Add more processes * Finalize process * Cleanup * Cleanup typing * Formatting * Remove daemon --------- Co-authored-by: Josh Hawkins <32435876+hawkeye217@users.noreply.github.com>
2025-06-12 21:12:34 +03:00
from frigate.record.record import RecordProcess
from frigate.review.review import ReviewProcess
from frigate.stats.emitter import StatsEmitter
from frigate.stats.util import stats_init
Limit recording retention to available storage (#3942) * Add field and migration for segment size * Store the segment size in db * Add comment * Add default * Fix size parsing * Include segment size in recordings endpoint * Start adding storage maintainer * Add storage maintainer and calculate average sizes * Update comment * Store segment and hour avg sizes per camera * Formatting * Keep track of total segment and hour averages * Remove unused files * Cleanup 2 hours of recordings at a time * Formatting * Fix bug * Round segment size * Cleanup some comments * Handle case where segments are not deleted on initial run or is only retained segments * Improve cleanup log * Formatting * Fix typo and improve logging * Catch case where no recordings exist for camera * Specifically define sort * Handle edge case for cameras that only record part time * Increase definition of part time recorder * Remove warning about not supported storage based retention * Add note about storage based retention to recording docs * Add tests for storage maintenance calculation and cleanup * Format tests * Don't run for a camera with no recording segments * Get size of file from cache * Rework camera stats to be more efficient * Remove total and other inefficencies * Rewrite storage cleanup logic to be much more efficient * Fix existing tests * Fix bugs from tests * Add another test * Improve logging * Formatting * Set back correct loop time * Update name * Update comment * Only include segments that have a nonzero size * Catch case where camera has 0 nonzero segment durations * Add test to cover zero bandwidth migration case * Fix test * Incorrect boolean logic * Formatting * Explicity re-define iterator
2022-10-09 14:28:26 +03:00
from frigate.storage import StorageMaintainer
from frigate.timeline import TimelineProcessor
from frigate.track.object_processing import TrackedObjectProcessor
from frigate.util.builtin import empty_and_close_queue
from frigate.util.image import UntrackedSharedMemory
from frigate.util.process import FrigateProcess
from frigate.util.services import set_file_limit
2021-09-14 06:02:23 +03:00
from frigate.version import VERSION
2020-11-04 15:28:07 +03:00
from frigate.watchdog import FrigateWatchdog
logger = logging.getLogger(__name__)
2021-02-17 16:23:32 +03:00
class FrigateApp:
def __init__(
self, config: FrigateConfig, manager: SyncManager, stop_event: MpEvent
) -> None:
Use Fork-Server As Spawn Method (#18682) * Set runtime * Use count correctly * Don't assume camera sizes * Use separate zmq proxy for object detection * Correct order * Use forkserver * Only store PID instead of entire process reference * Cleanup * Catch correct errors * Fix typing * Remove before_run from process util The before_run never actually ran because: You're right to suspect an issue with before_run not being called and a potential deadlock. The way you've implemented the run_wrapper using __getattribute__ for the run method of BaseProcess is a common pitfall in Python's multiprocessing, especially when combined with how multiprocessing.Process works internally. Here's a breakdown of why before_run isn't being called and why you might be experiencing a deadlock: The Problem: __getattribute__ and Process Serialization When you create a multiprocessing.Process object and call start(), the multiprocessing module needs to serialize the process object (or at least enough of it to re-create the process in the new interpreter). It then pickles this serialized object and sends it to the newly spawned process. The issue with your __getattribute__ implementation for run is that: run is retrieved during serialization: When multiprocessing tries to pickle your Process object to send to the new process, it will likely access the run attribute. This triggers your __getattribute__ wrapper, which then tries to bind run_wrapper to self. run_wrapper is bound to the parent process's self: The run_wrapper closure, when created in the parent process, captures the self (the Process instance) from the parent's memory space. Deserialization creates a new object: In the child process, a new Process object is created by deserializing the pickled data. However, the run_wrapper method that was pickled still holds a reference to the self from the parent process. This is a subtle but critical distinction. The child's run is not your wrapped run: When the child process starts, it internally calls its own run method. Because of the serialization and deserialization process, the run method that's ultimately executed in the child process is the original multiprocessing.Process.run or the Process.run if you had directly overridden it. Your __getattribute__ magic, which wraps run, isn't correctly applied to the Process object within the child's context. * Cleanup * Logging bugfix (#18465) * use mp Manager to handle logging queues A Python bug (https://github.com/python/cpython/issues/91555) was preventing logs from the embeddings maintainer process from printing. The bug is fixed in Python 3.14, but a viable workaround is to use the multiprocessing Manager, which better manages mp queues and causes the logging to work correctly. * consolidate * fix typing * Fix typing * Use global log queue * Move to using process for logging * Convert camera tracking to process * Add more processes * Finalize process * Cleanup * Cleanup typing * Formatting * Remove daemon --------- Co-authored-by: Josh Hawkins <32435876+hawkeye217@users.noreply.github.com>
2025-06-12 21:12:34 +03:00
self.metrics_manager = manager
self.audio_process: Optional[mp.Process] = None
self.stop_event = stop_event
self.detection_queue: Queue = mp.Queue()
self.detectors: dict[str, ObjectDetectProcess] = {}
self.detection_shms: list[mp.shared_memory.SharedMemory] = []
self.log_queue: Queue = mp.Queue()
Use Fork-Server As Spawn Method (#18682) * Set runtime * Use count correctly * Don't assume camera sizes * Use separate zmq proxy for object detection * Correct order * Use forkserver * Only store PID instead of entire process reference * Cleanup * Catch correct errors * Fix typing * Remove before_run from process util The before_run never actually ran because: You're right to suspect an issue with before_run not being called and a potential deadlock. The way you've implemented the run_wrapper using __getattribute__ for the run method of BaseProcess is a common pitfall in Python's multiprocessing, especially when combined with how multiprocessing.Process works internally. Here's a breakdown of why before_run isn't being called and why you might be experiencing a deadlock: The Problem: __getattribute__ and Process Serialization When you create a multiprocessing.Process object and call start(), the multiprocessing module needs to serialize the process object (or at least enough of it to re-create the process in the new interpreter). It then pickles this serialized object and sends it to the newly spawned process. The issue with your __getattribute__ implementation for run is that: run is retrieved during serialization: When multiprocessing tries to pickle your Process object to send to the new process, it will likely access the run attribute. This triggers your __getattribute__ wrapper, which then tries to bind run_wrapper to self. run_wrapper is bound to the parent process's self: The run_wrapper closure, when created in the parent process, captures the self (the Process instance) from the parent's memory space. Deserialization creates a new object: In the child process, a new Process object is created by deserializing the pickled data. However, the run_wrapper method that was pickled still holds a reference to the self from the parent process. This is a subtle but critical distinction. The child's run is not your wrapped run: When the child process starts, it internally calls its own run method. Because of the serialization and deserialization process, the run method that's ultimately executed in the child process is the original multiprocessing.Process.run or the Process.run if you had directly overridden it. Your __getattribute__ magic, which wraps run, isn't correctly applied to the Process object within the child's context. * Cleanup * Logging bugfix (#18465) * use mp Manager to handle logging queues A Python bug (https://github.com/python/cpython/issues/91555) was preventing logs from the embeddings maintainer process from printing. The bug is fixed in Python 3.14, but a viable workaround is to use the multiprocessing Manager, which better manages mp queues and causes the logging to work correctly. * consolidate * fix typing * Fix typing * Use global log queue * Move to using process for logging * Convert camera tracking to process * Add more processes * Finalize process * Cleanup * Cleanup typing * Formatting * Remove daemon --------- Co-authored-by: Josh Hawkins <32435876+hawkeye217@users.noreply.github.com>
2025-06-12 21:12:34 +03:00
self.camera_metrics: DictProxy = self.metrics_manager.dict()
self.embeddings_metrics: DataProcessorMetrics | None = (
Use Fork-Server As Spawn Method (#18682) * Set runtime * Use count correctly * Don't assume camera sizes * Use separate zmq proxy for object detection * Correct order * Use forkserver * Only store PID instead of entire process reference * Cleanup * Catch correct errors * Fix typing * Remove before_run from process util The before_run never actually ran because: You're right to suspect an issue with before_run not being called and a potential deadlock. The way you've implemented the run_wrapper using __getattribute__ for the run method of BaseProcess is a common pitfall in Python's multiprocessing, especially when combined with how multiprocessing.Process works internally. Here's a breakdown of why before_run isn't being called and why you might be experiencing a deadlock: The Problem: __getattribute__ and Process Serialization When you create a multiprocessing.Process object and call start(), the multiprocessing module needs to serialize the process object (or at least enough of it to re-create the process in the new interpreter). It then pickles this serialized object and sends it to the newly spawned process. The issue with your __getattribute__ implementation for run is that: run is retrieved during serialization: When multiprocessing tries to pickle your Process object to send to the new process, it will likely access the run attribute. This triggers your __getattribute__ wrapper, which then tries to bind run_wrapper to self. run_wrapper is bound to the parent process's self: The run_wrapper closure, when created in the parent process, captures the self (the Process instance) from the parent's memory space. Deserialization creates a new object: In the child process, a new Process object is created by deserializing the pickled data. However, the run_wrapper method that was pickled still holds a reference to the self from the parent process. This is a subtle but critical distinction. The child's run is not your wrapped run: When the child process starts, it internally calls its own run method. Because of the serialization and deserialization process, the run method that's ultimately executed in the child process is the original multiprocessing.Process.run or the Process.run if you had directly overridden it. Your __getattribute__ magic, which wraps run, isn't correctly applied to the Process object within the child's context. * Cleanup * Logging bugfix (#18465) * use mp Manager to handle logging queues A Python bug (https://github.com/python/cpython/issues/91555) was preventing logs from the embeddings maintainer process from printing. The bug is fixed in Python 3.14, but a viable workaround is to use the multiprocessing Manager, which better manages mp queues and causes the logging to work correctly. * consolidate * fix typing * Fix typing * Use global log queue * Move to using process for logging * Convert camera tracking to process * Add more processes * Finalize process * Cleanup * Cleanup typing * Formatting * Remove daemon --------- Co-authored-by: Josh Hawkins <32435876+hawkeye217@users.noreply.github.com>
2025-06-12 21:12:34 +03:00
DataProcessorMetrics(
self.metrics_manager, list(config.classification.custom.keys())
)
if (
config.semantic_search.enabled
or any(
c.objects.genai.enabled or c.review.genai.enabled
for c in config.cameras.values()
)
or config.lpr.enabled
or config.face_recognition.enabled
or len(config.classification.custom) > 0
)
else None
)
self.ptz_metrics: dict[str, PTZMetrics] = {}
self.processes: dict[str, int] = {}
Use sqlite-vec extension instead of chromadb for embeddings (#14163) * swap sqlite_vec for chroma in requirements * load sqlite_vec in embeddings manager * remove chroma and revamp Embeddings class for sqlite_vec * manual minilm onnx inference * remove chroma in clip model * migrate api from chroma to sqlite_vec * migrate event cleanup from chroma to sqlite_vec * migrate embedding maintainer from chroma to sqlite_vec * genai description for sqlite_vec * load sqlite_vec in main thread db * extend the SqliteQueueDatabase class and use peewee db.execute_sql * search with Event type for similarity * fix similarity search * install and add comment about transformers * fix normalization * add id filter * clean up * clean up * fully remove chroma and add transformers env var * readd uvicorn for fastapi * readd tokenizer parallelism env var * remove chroma from docs * remove chroma from UI * try removing custom pysqlite3 build * hard code limit * optimize queries * revert explore query * fix query * keep building pysqlite3 * single pass fetch and process * remove unnecessary re-embed * update deps * move SqliteVecQueueDatabase to db directory * make search thumbnail take up full size of results box * improve typing * improve model downloading and add status screen * daemon downloading thread * catch case when semantic search is disabled * fix typing * build sqlite_vec from source * resolve conflict * file permissions * try build deps * remove sources * sources * fix thread start * include git in build * reorder embeddings after detectors are started * build with sqlite amalgamation * non-platform specific * use wget instead of curl * remove unzip -d * remove sqlite_vec from requirements and load the compiled version * fix build * avoid race in db connection * add scale_factor and bias to description zscore normalization
2024-10-07 23:30:45 +03:00
self.embeddings: Optional[EmbeddingsContext] = None
Camera profile support (#22482) * add CameraProfileConfig model for named config overrides * add profiles field to CameraConfig * add active_profile field to FrigateConfig Runtime-only field excluded from YAML serialization, tracks which profile is currently active. * add ProfileManager for profile activation and persistence Handles snapshotting base configs, applying profile overrides via deep_merge + apply_section_update, publishing ZMQ updates, and persisting active profile to /config/.active_profile. * add profile API endpoints (GET /profiles, GET/PUT /profile) * add MQTT and dispatcher integration for profiles - Subscribe to frigate/profile/set MQTT topic - Publish profile/state and profiles/available on connect - Add _on_profile_command handler to dispatcher - Broadcast active profile state on WebSocket connect * wire ProfileManager into app startup and FastAPI - Create ProfileManager after dispatcher init - Restore persisted profile on startup - Pass dispatcher and profile_manager to FastAPI app * add tests for invalid profile values and keys Tests that Pydantic rejects: invalid field values (fps: "not_a_number"), unknown section keys (ffmpeg in profile), invalid nested values, and invalid profiles in full config parsing. * formatting * fix CameraLiveConfig JSON serialization error on profile activation refactor _publish_updates to only publish ZMQ updates for sections that actually changed, not all sections on affected cameras. * consolidate * add enabled field to camera profiles for enabling/disabling cameras * add zones support to camera profiles * add frontend profile types, color utility, and config save support * add profile state management and save preview support * add profileName prop to BaseSection for profile-aware config editing * add profile section dropdown and wire into camera settings pages * add per-profile camera enable/disable to Camera Management view * add profiles summary page with card-based layout and fix backend zone comparison bug * add active profile badge to settings toolbar * i18n * add red dot for any pending changes including profiles * profile support for mask and zone editor * fix hidden field validation errors caused by lodash wildcard and schema gaps lodash unset does not support wildcard (*) segments, so hidden fields like filters.*.mask were never stripped from form data, leaving null raw_coordinates that fail RJSF anyOf validation. Add unsetWithWildcard helper and also strip hidden fields from the JSON schema itself as defense-in-depth. * add face_recognition and lpr to profile-eligible sections * move profile dropdown from section panes to settings header * add profiles enable toggle and improve empty state * formatting * tweaks * tweak colors and switch * fix profile save diff, masksAndZones delete, and config sync * ui tweaks * ensure profile manager gets updated config * rename profile settings to ui settings * refactor profilesview and add dots/border colors when overridden * implement an update_config method for profile manager * fix mask deletion * more unique colors * add top-level profiles config section with friendly names * implement profile friendly names and improve profile UI - Add ProfileDefinitionConfig type and profiles field to FrigateConfig - Use ProfilesApiResponse type with friendly_name support throughout - Replace Record<string, unknown> with proper JsonObject/JsonValue types - Add profile creation form matching zone pattern (Zod + NameAndIdFields) - Add pencil icon for renaming profile friendly names in ProfilesView - Move Profiles menu item to first under Camera Configuration - Add activity indicators on save/rename/delete buttons - Display friendly names in CameraManagementView profile selector - Fix duplicate colored dots in management profile dropdown - Fix i18n namespace for overridden base config tooltips - Move profile override deletion from dropdown trash icon to footer button with confirmation dialog, matching Reset to Global pattern - Remove Add Profile from section header dropdown to prevent saving camera overrides before top-level profile definition exists - Clean up newProfiles state after API profile deletion - Refresh profiles SWR cache after saving profile definitions * remove profile badge in settings and add profiles to main menu * use icon only on mobile * change color order * docs * show activity indicator on trash icon while deleting a profile * tweak language * immediately create profiles on backend instead of deferring to Save All * hide restart-required fields when editing a profile section fields that require a restart cannot take effect via profile switching, so they are merged into hiddenFields when profileName is set * show active profile indicator in desktop status bar * fix profile config inheritance bug where Pydantic defaults override base values The /config API was dumping profile overrides with model_dump() which included all Pydantic defaults. When the frontend merged these over the camera's base config, explicitly-set base values were lost. Now profile overrides are re-dumped with exclude_unset=True so only user-specified fields are returned. Also fixes the Save All path generating spurious deletion markers for restart-required fields that are hidden during profile editing but not excluded from the raw data sanitization in prepareSectionSavePayload. * docs tweaks * docs tweak * formatting * formatting * fix typing * fix test pollution test_maintainer was injecting MagicMock() into sys.modules["frigate.config.camera.updater"] at module load time and never restoring it. When the profile tests later imported CameraConfigUpdateEnum and CameraConfigUpdateTopic from that module, they got mock objects instead of the real dataclass/enum, so equality comparisons always failed * remove * fix settings showing profile-merged values when editing base config When a profile is active, the in-memory config contains effective (profile-merged) values. The settings UI was displaying these merged values even when the "Base Config" view was selected. Backend: snapshot pre-profile base configs in ProfileManager and expose them via a `base_config` key in the /api/config camera response when a profile is active. The top-level sections continue to reflect the effective running config. Frontend: read from `base_config` when available in BaseSection, useConfigOverride, useAllCameraOverrides, and prepareSectionSavePayload. Include formData labels in Object/Audio switches widgets so that labels added only by a profile override remain visible when editing that profile. * use rasterized_mask as field makes it easier to exclude from the schema with exclude=True prevents leaking of the field when using model_dump for profiles * fix zones - Fix zone colors not matching across profiles by falling back to base zone color when profile zone data lacks a color field - Use base_config for base-layer values in masks/zones view so profile-merged values don't pollute the base config editing view - Handle zones separately in profile manager snapshot/restore since ZoneConfig requires special serialization (color as private attr, contour generation) - Inherit base zone color and generate contours for profile zone overrides in profile manager * formatting * don't require restart for camera enabled change for profiles * publish camera state when changing profiles * formatting * remove available profiles from mqtt * improve typing
2026-03-19 17:47:57 +03:00
self.profile_manager: Optional[ProfileManager] = None
self.config = config
2021-01-16 06:33:53 +03:00
def ensure_dirs(self) -> None:
dirs = [
CONFIG_DIR,
RECORD_DIR,
THUMB_DIR,
f"{CLIPS_DIR}/cache",
CACHE_DIR,
MODEL_CACHE_DIR,
EXPORT_DIR,
]
if self.config.face_recognition.enabled:
dirs.append(FACE_DIR)
if self.config.semantic_search.enabled:
dirs.append(TRIGGER_DIR)
for d in dirs:
2020-12-01 16:22:23 +03:00
if not os.path.exists(d) and not os.path.islink(d):
2020-12-03 17:01:22 +03:00
logger.info(f"Creating directory: {d}")
2020-12-01 16:22:23 +03:00
os.makedirs(d)
2020-12-03 17:01:22 +03:00
else:
logger.debug(f"Skipping directory: {d}")
2020-12-21 16:37:42 +03:00
def init_debug_replay_manager(self) -> None:
self.replay_manager = DebugReplayManager()
def init_camera_metrics(self) -> None:
# create camera_metrics
for camera_name in self.config.cameras.keys():
Use Fork-Server As Spawn Method (#18682) * Set runtime * Use count correctly * Don't assume camera sizes * Use separate zmq proxy for object detection * Correct order * Use forkserver * Only store PID instead of entire process reference * Cleanup * Catch correct errors * Fix typing * Remove before_run from process util The before_run never actually ran because: You're right to suspect an issue with before_run not being called and a potential deadlock. The way you've implemented the run_wrapper using __getattribute__ for the run method of BaseProcess is a common pitfall in Python's multiprocessing, especially when combined with how multiprocessing.Process works internally. Here's a breakdown of why before_run isn't being called and why you might be experiencing a deadlock: The Problem: __getattribute__ and Process Serialization When you create a multiprocessing.Process object and call start(), the multiprocessing module needs to serialize the process object (or at least enough of it to re-create the process in the new interpreter). It then pickles this serialized object and sends it to the newly spawned process. The issue with your __getattribute__ implementation for run is that: run is retrieved during serialization: When multiprocessing tries to pickle your Process object to send to the new process, it will likely access the run attribute. This triggers your __getattribute__ wrapper, which then tries to bind run_wrapper to self. run_wrapper is bound to the parent process's self: The run_wrapper closure, when created in the parent process, captures the self (the Process instance) from the parent's memory space. Deserialization creates a new object: In the child process, a new Process object is created by deserializing the pickled data. However, the run_wrapper method that was pickled still holds a reference to the self from the parent process. This is a subtle but critical distinction. The child's run is not your wrapped run: When the child process starts, it internally calls its own run method. Because of the serialization and deserialization process, the run method that's ultimately executed in the child process is the original multiprocessing.Process.run or the Process.run if you had directly overridden it. Your __getattribute__ magic, which wraps run, isn't correctly applied to the Process object within the child's context. * Cleanup * Logging bugfix (#18465) * use mp Manager to handle logging queues A Python bug (https://github.com/python/cpython/issues/91555) was preventing logs from the embeddings maintainer process from printing. The bug is fixed in Python 3.14, but a viable workaround is to use the multiprocessing Manager, which better manages mp queues and causes the logging to work correctly. * consolidate * fix typing * Fix typing * Use global log queue * Move to using process for logging * Convert camera tracking to process * Add more processes * Finalize process * Cleanup * Cleanup typing * Formatting * Remove daemon --------- Co-authored-by: Josh Hawkins <32435876+hawkeye217@users.noreply.github.com>
2025-06-12 21:12:34 +03:00
self.camera_metrics[camera_name] = CameraMetrics(self.metrics_manager)
self.ptz_metrics[camera_name] = PTZMetrics(
autotracker_enabled=self.config.cameras[
camera_name
].onvif.autotracking.enabled
)
2021-02-17 16:23:32 +03:00
def init_queues(self) -> None:
2020-11-04 15:28:07 +03:00
# Queue for cameras to push tracked objects to
# leaving room for 2 extra cameras to be added
self.detected_frames_queue: Queue = mp.Queue(
maxsize=(
sum(
camera.enabled_in_config == True
for camera in self.config.cameras.values()
)
+ 2
)
* 2
2021-02-17 16:23:32 +03:00
)
2020-11-04 15:28:07 +03:00
# Queue for timeline events
self.timeline_queue: Queue = mp.Queue()
def init_database(self) -> None:
def vacuum_db(db: SqliteExtDatabase) -> None:
logger.info("Running database vacuum")
db.execute_sql("VACUUM;")
try:
with open(f"{CONFIG_DIR}/.vacuum", "w") as f:
f.write(str(datetime.datetime.now().timestamp()))
except PermissionError:
logger.error("Unable to write to /config to save DB state")
def cleanup_timeline_db(db: SqliteExtDatabase) -> None:
db.execute_sql(
"DELETE FROM timeline WHERE source_id NOT IN (SELECT id FROM event);"
)
try:
with open(f"{CONFIG_DIR}/.timeline", "w") as f:
f.write(str(datetime.datetime.now().timestamp()))
except PermissionError:
logger.error("Unable to write to /config to save DB state")
# Migrate DB schema
2021-01-24 15:53:01 +03:00
migrate_db = SqliteExtDatabase(self.config.database.path)
2020-12-24 16:47:27 +03:00
# Run migrations
2021-02-17 16:23:32 +03:00
del logging.getLogger("peewee_migrate").handlers[:]
2021-01-24 15:53:01 +03:00
router = Router(migrate_db)
if len(router.diff) > 0:
logger.info("Making backup of DB before migrations...")
shutil.copyfile(
self.config.database.path,
self.config.database.path.replace("frigate.db", "backup.db"),
)
2020-12-24 16:47:27 +03:00
router.run()
# this is a temporary check to clean up user DB from beta
# will be removed before final release
if not os.path.exists(f"{CONFIG_DIR}/.timeline"):
cleanup_timeline_db(migrate_db)
# check if vacuum needs to be run
if os.path.exists(f"{CONFIG_DIR}/.vacuum"):
with open(f"{CONFIG_DIR}/.vacuum") as f:
try:
timestamp = round(float(f.readline()))
except Exception:
timestamp = 0
if (
timestamp
< (
datetime.datetime.now() - datetime.timedelta(weeks=2)
).timestamp()
):
vacuum_db(migrate_db)
else:
vacuum_db(migrate_db)
2021-01-24 15:53:01 +03:00
migrate_db.close()
def init_go2rtc(self) -> None:
for proc in psutil.process_iter(["pid", "name"]):
if proc.info["name"] == "go2rtc":
logger.info(f"go2rtc process pid: {proc.info['pid']}")
self.processes["go2rtc"] = proc.info["pid"]
def init_recording_manager(self) -> None:
recording_process = RecordProcess(self.config, self.stop_event)
self.recording_process = recording_process
recording_process.start()
self.processes["recording"] = recording_process.pid or 0
logger.info(f"Recording process started: {recording_process.pid}")
def init_review_segment_manager(self) -> None:
review_segment_process = ReviewProcess(self.config, self.stop_event)
self.review_segment_process = review_segment_process
review_segment_process.start()
self.processes["review_segment"] = review_segment_process.pid or 0
logger.info(f"Review process started: {review_segment_process.pid}")
def init_embeddings_manager(self) -> None:
# always start the embeddings process
Use Fork-Server As Spawn Method (#18682) * Set runtime * Use count correctly * Don't assume camera sizes * Use separate zmq proxy for object detection * Correct order * Use forkserver * Only store PID instead of entire process reference * Cleanup * Catch correct errors * Fix typing * Remove before_run from process util The before_run never actually ran because: You're right to suspect an issue with before_run not being called and a potential deadlock. The way you've implemented the run_wrapper using __getattribute__ for the run method of BaseProcess is a common pitfall in Python's multiprocessing, especially when combined with how multiprocessing.Process works internally. Here's a breakdown of why before_run isn't being called and why you might be experiencing a deadlock: The Problem: __getattribute__ and Process Serialization When you create a multiprocessing.Process object and call start(), the multiprocessing module needs to serialize the process object (or at least enough of it to re-create the process in the new interpreter). It then pickles this serialized object and sends it to the newly spawned process. The issue with your __getattribute__ implementation for run is that: run is retrieved during serialization: When multiprocessing tries to pickle your Process object to send to the new process, it will likely access the run attribute. This triggers your __getattribute__ wrapper, which then tries to bind run_wrapper to self. run_wrapper is bound to the parent process's self: The run_wrapper closure, when created in the parent process, captures the self (the Process instance) from the parent's memory space. Deserialization creates a new object: In the child process, a new Process object is created by deserializing the pickled data. However, the run_wrapper method that was pickled still holds a reference to the self from the parent process. This is a subtle but critical distinction. The child's run is not your wrapped run: When the child process starts, it internally calls its own run method. Because of the serialization and deserialization process, the run method that's ultimately executed in the child process is the original multiprocessing.Process.run or the Process.run if you had directly overridden it. Your __getattribute__ magic, which wraps run, isn't correctly applied to the Process object within the child's context. * Cleanup * Logging bugfix (#18465) * use mp Manager to handle logging queues A Python bug (https://github.com/python/cpython/issues/91555) was preventing logs from the embeddings maintainer process from printing. The bug is fixed in Python 3.14, but a viable workaround is to use the multiprocessing Manager, which better manages mp queues and causes the logging to work correctly. * consolidate * fix typing * Fix typing * Use global log queue * Move to using process for logging * Convert camera tracking to process * Add more processes * Finalize process * Cleanup * Cleanup typing * Formatting * Remove daemon --------- Co-authored-by: Josh Hawkins <32435876+hawkeye217@users.noreply.github.com>
2025-06-12 21:12:34 +03:00
embedding_process = EmbeddingProcess(
self.config, self.embeddings_metrics, self.stop_event
)
self.embedding_process = embedding_process
embedding_process.start()
self.processes["embeddings"] = embedding_process.pid or 0
logger.info(f"Embedding process started: {embedding_process.pid}")
def bind_database(self) -> None:
"""Bind db to the main process."""
# NOTE: all db accessing processes need to be created before the db can be bound to the main process
Use sqlite-vec extension instead of chromadb for embeddings (#14163) * swap sqlite_vec for chroma in requirements * load sqlite_vec in embeddings manager * remove chroma and revamp Embeddings class for sqlite_vec * manual minilm onnx inference * remove chroma in clip model * migrate api from chroma to sqlite_vec * migrate event cleanup from chroma to sqlite_vec * migrate embedding maintainer from chroma to sqlite_vec * genai description for sqlite_vec * load sqlite_vec in main thread db * extend the SqliteQueueDatabase class and use peewee db.execute_sql * search with Event type for similarity * fix similarity search * install and add comment about transformers * fix normalization * add id filter * clean up * clean up * fully remove chroma and add transformers env var * readd uvicorn for fastapi * readd tokenizer parallelism env var * remove chroma from docs * remove chroma from UI * try removing custom pysqlite3 build * hard code limit * optimize queries * revert explore query * fix query * keep building pysqlite3 * single pass fetch and process * remove unnecessary re-embed * update deps * move SqliteVecQueueDatabase to db directory * make search thumbnail take up full size of results box * improve typing * improve model downloading and add status screen * daemon downloading thread * catch case when semantic search is disabled * fix typing * build sqlite_vec from source * resolve conflict * file permissions * try build deps * remove sources * sources * fix thread start * include git in build * reorder embeddings after detectors are started * build with sqlite amalgamation * non-platform specific * use wget instead of curl * remove unzip -d * remove sqlite_vec from requirements and load the compiled version * fix build * avoid race in db connection * add scale_factor and bias to description zscore normalization
2024-10-07 23:30:45 +03:00
self.db = SqliteVecQueueDatabase(
self.config.database.path,
pragmas={
"auto_vacuum": "FULL", # Does not defragment database
"cache_size": -512 * 1000, # 512MB of cache,
"synchronous": "NORMAL", # Safe when using WAL https://www.sqlite.org/pragma.html#pragma_synchronous
},
timeout=max(
60,
10
* len([c for c in self.config.cameras.values() if c.enabled_in_config]),
),
Use sqlite-vec extension instead of chromadb for embeddings (#14163) * swap sqlite_vec for chroma in requirements * load sqlite_vec in embeddings manager * remove chroma and revamp Embeddings class for sqlite_vec * manual minilm onnx inference * remove chroma in clip model * migrate api from chroma to sqlite_vec * migrate event cleanup from chroma to sqlite_vec * migrate embedding maintainer from chroma to sqlite_vec * genai description for sqlite_vec * load sqlite_vec in main thread db * extend the SqliteQueueDatabase class and use peewee db.execute_sql * search with Event type for similarity * fix similarity search * install and add comment about transformers * fix normalization * add id filter * clean up * clean up * fully remove chroma and add transformers env var * readd uvicorn for fastapi * readd tokenizer parallelism env var * remove chroma from docs * remove chroma from UI * try removing custom pysqlite3 build * hard code limit * optimize queries * revert explore query * fix query * keep building pysqlite3 * single pass fetch and process * remove unnecessary re-embed * update deps * move SqliteVecQueueDatabase to db directory * make search thumbnail take up full size of results box * improve typing * improve model downloading and add status screen * daemon downloading thread * catch case when semantic search is disabled * fix typing * build sqlite_vec from source * resolve conflict * file permissions * try build deps * remove sources * sources * fix thread start * include git in build * reorder embeddings after detectors are started * build with sqlite amalgamation * non-platform specific * use wget instead of curl * remove unzip -d * remove sqlite_vec from requirements and load the compiled version * fix build * avoid race in db connection * add scale_factor and bias to description zscore normalization
2024-10-07 23:30:45 +03:00
load_vec_extension=self.config.semantic_search.enabled,
)
models = [
Event,
Export,
Previews,
Recordings,
RecordingsToDelete,
Regions,
ReviewSegment,
Timeline,
User,
Trigger,
]
2020-11-04 15:28:07 +03:00
self.db.bind(models)
def check_db_data_migrations(self) -> None:
# check if vacuum needs to be run
if not os.path.exists(f"{CONFIG_DIR}/.exports"):
try:
with open(f"{CONFIG_DIR}/.exports", "w") as f:
f.write(str(datetime.datetime.now().timestamp()))
except PermissionError:
logger.error("Unable to write to /config to save export state")
migrate_exports(self.config.ffmpeg, list(self.config.cameras.keys()))
Use sqlite-vec extension instead of chromadb for embeddings (#14163) * swap sqlite_vec for chroma in requirements * load sqlite_vec in embeddings manager * remove chroma and revamp Embeddings class for sqlite_vec * manual minilm onnx inference * remove chroma in clip model * migrate api from chroma to sqlite_vec * migrate event cleanup from chroma to sqlite_vec * migrate embedding maintainer from chroma to sqlite_vec * genai description for sqlite_vec * load sqlite_vec in main thread db * extend the SqliteQueueDatabase class and use peewee db.execute_sql * search with Event type for similarity * fix similarity search * install and add comment about transformers * fix normalization * add id filter * clean up * clean up * fully remove chroma and add transformers env var * readd uvicorn for fastapi * readd tokenizer parallelism env var * remove chroma from docs * remove chroma from UI * try removing custom pysqlite3 build * hard code limit * optimize queries * revert explore query * fix query * keep building pysqlite3 * single pass fetch and process * remove unnecessary re-embed * update deps * move SqliteVecQueueDatabase to db directory * make search thumbnail take up full size of results box * improve typing * improve model downloading and add status screen * daemon downloading thread * catch case when semantic search is disabled * fix typing * build sqlite_vec from source * resolve conflict * file permissions * try build deps * remove sources * sources * fix thread start * include git in build * reorder embeddings after detectors are started * build with sqlite amalgamation * non-platform specific * use wget instead of curl * remove unzip -d * remove sqlite_vec from requirements and load the compiled version * fix build * avoid race in db connection * add scale_factor and bias to description zscore normalization
2024-10-07 23:30:45 +03:00
def init_embeddings_client(self) -> None:
# Create a client for other processes to use
self.embeddings = EmbeddingsContext(self.db)
Use sqlite-vec extension instead of chromadb for embeddings (#14163) * swap sqlite_vec for chroma in requirements * load sqlite_vec in embeddings manager * remove chroma and revamp Embeddings class for sqlite_vec * manual minilm onnx inference * remove chroma in clip model * migrate api from chroma to sqlite_vec * migrate event cleanup from chroma to sqlite_vec * migrate embedding maintainer from chroma to sqlite_vec * genai description for sqlite_vec * load sqlite_vec in main thread db * extend the SqliteQueueDatabase class and use peewee db.execute_sql * search with Event type for similarity * fix similarity search * install and add comment about transformers * fix normalization * add id filter * clean up * clean up * fully remove chroma and add transformers env var * readd uvicorn for fastapi * readd tokenizer parallelism env var * remove chroma from docs * remove chroma from UI * try removing custom pysqlite3 build * hard code limit * optimize queries * revert explore query * fix query * keep building pysqlite3 * single pass fetch and process * remove unnecessary re-embed * update deps * move SqliteVecQueueDatabase to db directory * make search thumbnail take up full size of results box * improve typing * improve model downloading and add status screen * daemon downloading thread * catch case when semantic search is disabled * fix typing * build sqlite_vec from source * resolve conflict * file permissions * try build deps * remove sources * sources * fix thread start * include git in build * reorder embeddings after detectors are started * build with sqlite amalgamation * non-platform specific * use wget instead of curl * remove unzip -d * remove sqlite_vec from requirements and load the compiled version * fix build * avoid race in db connection * add scale_factor and bias to description zscore normalization
2024-10-07 23:30:45 +03:00
def init_inter_process_communicator(self) -> None:
self.inter_process_communicator = InterProcessCommunicator()
self.inter_config_updater = CameraConfigUpdatePublisher()
self.event_metadata_updater = EventMetadataPublisher()
self.inter_zmq_proxy = ZmqProxy()
Use Fork-Server As Spawn Method (#18682) * Set runtime * Use count correctly * Don't assume camera sizes * Use separate zmq proxy for object detection * Correct order * Use forkserver * Only store PID instead of entire process reference * Cleanup * Catch correct errors * Fix typing * Remove before_run from process util The before_run never actually ran because: You're right to suspect an issue with before_run not being called and a potential deadlock. The way you've implemented the run_wrapper using __getattribute__ for the run method of BaseProcess is a common pitfall in Python's multiprocessing, especially when combined with how multiprocessing.Process works internally. Here's a breakdown of why before_run isn't being called and why you might be experiencing a deadlock: The Problem: __getattribute__ and Process Serialization When you create a multiprocessing.Process object and call start(), the multiprocessing module needs to serialize the process object (or at least enough of it to re-create the process in the new interpreter). It then pickles this serialized object and sends it to the newly spawned process. The issue with your __getattribute__ implementation for run is that: run is retrieved during serialization: When multiprocessing tries to pickle your Process object to send to the new process, it will likely access the run attribute. This triggers your __getattribute__ wrapper, which then tries to bind run_wrapper to self. run_wrapper is bound to the parent process's self: The run_wrapper closure, when created in the parent process, captures the self (the Process instance) from the parent's memory space. Deserialization creates a new object: In the child process, a new Process object is created by deserializing the pickled data. However, the run_wrapper method that was pickled still holds a reference to the self from the parent process. This is a subtle but critical distinction. The child's run is not your wrapped run: When the child process starts, it internally calls its own run method. Because of the serialization and deserialization process, the run method that's ultimately executed in the child process is the original multiprocessing.Process.run or the Process.run if you had directly overridden it. Your __getattribute__ magic, which wraps run, isn't correctly applied to the Process object within the child's context. * Cleanup * Logging bugfix (#18465) * use mp Manager to handle logging queues A Python bug (https://github.com/python/cpython/issues/91555) was preventing logs from the embeddings maintainer process from printing. The bug is fixed in Python 3.14, but a viable workaround is to use the multiprocessing Manager, which better manages mp queues and causes the logging to work correctly. * consolidate * fix typing * Fix typing * Use global log queue * Move to using process for logging * Convert camera tracking to process * Add more processes * Finalize process * Cleanup * Cleanup typing * Formatting * Remove daemon --------- Co-authored-by: Josh Hawkins <32435876+hawkeye217@users.noreply.github.com>
2025-06-12 21:12:34 +03:00
self.detection_proxy = DetectorProxy()
def init_onvif(self) -> None:
self.onvif_controller = OnvifController(self.config, self.ptz_metrics)
def init_dispatcher(self) -> None:
comms: list[Communicator] = []
2020-11-04 15:28:07 +03:00
if self.config.mqtt.enabled:
comms.append(MqttClient(self.config))
notification_cameras = [
c
for c in self.config.cameras.values()
if c.enabled and c.notifications.enabled_in_config
]
if notification_cameras:
comms.append(WebPushClient(self.config, self.stop_event))
comms.append(WebSocketClient(self.config))
comms.append(self.inter_process_communicator)
self.dispatcher = Dispatcher(
self.config,
self.inter_config_updater,
self.onvif_controller,
self.ptz_metrics,
comms,
)
2021-06-14 15:31:13 +03:00
Camera profile support (#22482) * add CameraProfileConfig model for named config overrides * add profiles field to CameraConfig * add active_profile field to FrigateConfig Runtime-only field excluded from YAML serialization, tracks which profile is currently active. * add ProfileManager for profile activation and persistence Handles snapshotting base configs, applying profile overrides via deep_merge + apply_section_update, publishing ZMQ updates, and persisting active profile to /config/.active_profile. * add profile API endpoints (GET /profiles, GET/PUT /profile) * add MQTT and dispatcher integration for profiles - Subscribe to frigate/profile/set MQTT topic - Publish profile/state and profiles/available on connect - Add _on_profile_command handler to dispatcher - Broadcast active profile state on WebSocket connect * wire ProfileManager into app startup and FastAPI - Create ProfileManager after dispatcher init - Restore persisted profile on startup - Pass dispatcher and profile_manager to FastAPI app * add tests for invalid profile values and keys Tests that Pydantic rejects: invalid field values (fps: "not_a_number"), unknown section keys (ffmpeg in profile), invalid nested values, and invalid profiles in full config parsing. * formatting * fix CameraLiveConfig JSON serialization error on profile activation refactor _publish_updates to only publish ZMQ updates for sections that actually changed, not all sections on affected cameras. * consolidate * add enabled field to camera profiles for enabling/disabling cameras * add zones support to camera profiles * add frontend profile types, color utility, and config save support * add profile state management and save preview support * add profileName prop to BaseSection for profile-aware config editing * add profile section dropdown and wire into camera settings pages * add per-profile camera enable/disable to Camera Management view * add profiles summary page with card-based layout and fix backend zone comparison bug * add active profile badge to settings toolbar * i18n * add red dot for any pending changes including profiles * profile support for mask and zone editor * fix hidden field validation errors caused by lodash wildcard and schema gaps lodash unset does not support wildcard (*) segments, so hidden fields like filters.*.mask were never stripped from form data, leaving null raw_coordinates that fail RJSF anyOf validation. Add unsetWithWildcard helper and also strip hidden fields from the JSON schema itself as defense-in-depth. * add face_recognition and lpr to profile-eligible sections * move profile dropdown from section panes to settings header * add profiles enable toggle and improve empty state * formatting * tweaks * tweak colors and switch * fix profile save diff, masksAndZones delete, and config sync * ui tweaks * ensure profile manager gets updated config * rename profile settings to ui settings * refactor profilesview and add dots/border colors when overridden * implement an update_config method for profile manager * fix mask deletion * more unique colors * add top-level profiles config section with friendly names * implement profile friendly names and improve profile UI - Add ProfileDefinitionConfig type and profiles field to FrigateConfig - Use ProfilesApiResponse type with friendly_name support throughout - Replace Record<string, unknown> with proper JsonObject/JsonValue types - Add profile creation form matching zone pattern (Zod + NameAndIdFields) - Add pencil icon for renaming profile friendly names in ProfilesView - Move Profiles menu item to first under Camera Configuration - Add activity indicators on save/rename/delete buttons - Display friendly names in CameraManagementView profile selector - Fix duplicate colored dots in management profile dropdown - Fix i18n namespace for overridden base config tooltips - Move profile override deletion from dropdown trash icon to footer button with confirmation dialog, matching Reset to Global pattern - Remove Add Profile from section header dropdown to prevent saving camera overrides before top-level profile definition exists - Clean up newProfiles state after API profile deletion - Refresh profiles SWR cache after saving profile definitions * remove profile badge in settings and add profiles to main menu * use icon only on mobile * change color order * docs * show activity indicator on trash icon while deleting a profile * tweak language * immediately create profiles on backend instead of deferring to Save All * hide restart-required fields when editing a profile section fields that require a restart cannot take effect via profile switching, so they are merged into hiddenFields when profileName is set * show active profile indicator in desktop status bar * fix profile config inheritance bug where Pydantic defaults override base values The /config API was dumping profile overrides with model_dump() which included all Pydantic defaults. When the frontend merged these over the camera's base config, explicitly-set base values were lost. Now profile overrides are re-dumped with exclude_unset=True so only user-specified fields are returned. Also fixes the Save All path generating spurious deletion markers for restart-required fields that are hidden during profile editing but not excluded from the raw data sanitization in prepareSectionSavePayload. * docs tweaks * docs tweak * formatting * formatting * fix typing * fix test pollution test_maintainer was injecting MagicMock() into sys.modules["frigate.config.camera.updater"] at module load time and never restoring it. When the profile tests later imported CameraConfigUpdateEnum and CameraConfigUpdateTopic from that module, they got mock objects instead of the real dataclass/enum, so equality comparisons always failed * remove * fix settings showing profile-merged values when editing base config When a profile is active, the in-memory config contains effective (profile-merged) values. The settings UI was displaying these merged values even when the "Base Config" view was selected. Backend: snapshot pre-profile base configs in ProfileManager and expose them via a `base_config` key in the /api/config camera response when a profile is active. The top-level sections continue to reflect the effective running config. Frontend: read from `base_config` when available in BaseSection, useConfigOverride, useAllCameraOverrides, and prepareSectionSavePayload. Include formData labels in Object/Audio switches widgets so that labels added only by a profile override remain visible when editing that profile. * use rasterized_mask as field makes it easier to exclude from the schema with exclude=True prevents leaking of the field when using model_dump for profiles * fix zones - Fix zone colors not matching across profiles by falling back to base zone color when profile zone data lacks a color field - Use base_config for base-layer values in masks/zones view so profile-merged values don't pollute the base config editing view - Handle zones separately in profile manager snapshot/restore since ZoneConfig requires special serialization (color as private attr, contour generation) - Inherit base zone color and generate contours for profile zone overrides in profile manager * formatting * don't require restart for camera enabled change for profiles * publish camera state when changing profiles * formatting * remove available profiles from mqtt * improve typing
2026-03-19 17:47:57 +03:00
def init_profile_manager(self) -> None:
self.profile_manager = ProfileManager(
self.config, self.inter_config_updater, self.dispatcher
)
self.dispatcher.profile_manager = self.profile_manager
persisted = ProfileManager.load_persisted_profile()
if persisted and any(
persisted in cam.profiles for cam in self.config.cameras.values()
):
logger.info("Restoring persisted profile '%s'", persisted)
self.profile_manager.activate_profile(persisted)
def start_detectors(self) -> None:
2020-11-04 15:28:07 +03:00
for name in self.config.cameras.keys():
try:
largest_frame = max(
[
det.model.height * det.model.width * 3
if det.model is not None
else 320
for det in self.config.detectors.values()
]
)
shm_in = UntrackedSharedMemory(
2021-06-24 08:45:27 +03:00
name=name,
create=True,
size=largest_frame,
)
except FileExistsError:
shm_in = UntrackedSharedMemory(name=name)
try:
shm_out = UntrackedSharedMemory(
name=f"out-{name}", create=True, size=20 * 6 * 4
)
except FileExistsError:
shm_out = UntrackedSharedMemory(name=f"out-{name}")
2020-11-04 15:28:07 +03:00
self.detection_shms.append(shm_in)
self.detection_shms.append(shm_out)
for name, detector_config in self.config.detectors.items():
self.detectors[name] = ObjectDetectProcess(
name,
self.detection_queue,
list(self.config.cameras.keys()),
self.config,
detector_config,
self.stop_event,
)
2020-11-04 15:28:07 +03:00
def start_ptz_autotracker(self) -> None:
self.ptz_autotracker_thread = PtzAutoTrackerThread(
self.config,
self.onvif_controller,
self.ptz_metrics,
self.dispatcher,
self.stop_event,
)
self.ptz_autotracker_thread.start()
def start_detected_frames_processor(self) -> None:
2021-02-17 16:23:32 +03:00
self.detected_frames_processor = TrackedObjectProcessor(
self.config,
self.dispatcher,
2021-02-17 16:23:32 +03:00
self.detected_frames_queue,
self.ptz_autotracker_thread,
2021-02-17 16:23:32 +03:00
self.stop_event,
)
2020-11-04 15:28:07 +03:00
self.detected_frames_processor.start()
def start_video_output_processor(self) -> None:
output_processor = OutputProcess(self.config, self.stop_event)
self.output_processor = output_processor
output_processor.start()
2021-05-30 14:45:37 +03:00
logger.info(f"Output process started: {output_processor.pid}")
def start_camera_processor(self) -> None:
self.camera_maintainer = CameraMaintainer(
self.config,
self.detection_queue,
self.detected_frames_queue,
self.camera_metrics,
self.ptz_metrics,
self.stop_event,
self.metrics_manager,
)
self.camera_maintainer.start()
2021-02-17 16:23:32 +03:00
def start_audio_processor(self) -> None:
audio_cameras = [
c
for c in self.config.cameras.values()
if c.enabled and c.audio.enabled_in_config
]
if audio_cameras:
self.audio_process = AudioProcessor(
self.config, audio_cameras, self.camera_metrics, self.stop_event
)
self.audio_process.start()
self.processes["audio_detector"] = self.audio_process.pid or 0
def start_timeline_processor(self) -> None:
self.timeline_processor = TimelineProcessor(
self.config, self.timeline_queue, self.stop_event
)
self.timeline_processor.start()
def start_event_processor(self) -> None:
2021-02-17 16:23:32 +03:00
self.event_processor = EventProcessor(
self.config,
self.timeline_queue,
2021-02-17 16:23:32 +03:00
self.stop_event,
)
2020-11-04 15:28:07 +03:00
self.event_processor.start()
2021-02-17 16:23:32 +03:00
def start_event_cleanup(self) -> None:
Use sqlite-vec extension instead of chromadb for embeddings (#14163) * swap sqlite_vec for chroma in requirements * load sqlite_vec in embeddings manager * remove chroma and revamp Embeddings class for sqlite_vec * manual minilm onnx inference * remove chroma in clip model * migrate api from chroma to sqlite_vec * migrate event cleanup from chroma to sqlite_vec * migrate embedding maintainer from chroma to sqlite_vec * genai description for sqlite_vec * load sqlite_vec in main thread db * extend the SqliteQueueDatabase class and use peewee db.execute_sql * search with Event type for similarity * fix similarity search * install and add comment about transformers * fix normalization * add id filter * clean up * clean up * fully remove chroma and add transformers env var * readd uvicorn for fastapi * readd tokenizer parallelism env var * remove chroma from docs * remove chroma from UI * try removing custom pysqlite3 build * hard code limit * optimize queries * revert explore query * fix query * keep building pysqlite3 * single pass fetch and process * remove unnecessary re-embed * update deps * move SqliteVecQueueDatabase to db directory * make search thumbnail take up full size of results box * improve typing * improve model downloading and add status screen * daemon downloading thread * catch case when semantic search is disabled * fix typing * build sqlite_vec from source * resolve conflict * file permissions * try build deps * remove sources * sources * fix thread start * include git in build * reorder embeddings after detectors are started * build with sqlite amalgamation * non-platform specific * use wget instead of curl * remove unzip -d * remove sqlite_vec from requirements and load the compiled version * fix build * avoid race in db connection * add scale_factor and bias to description zscore normalization
2024-10-07 23:30:45 +03:00
self.event_cleanup = EventCleanup(self.config, self.stop_event, self.db)
2020-11-24 16:27:51 +03:00
self.event_cleanup.start()
2021-02-17 16:23:32 +03:00
def start_record_cleanup(self) -> None:
self.record_cleanup = RecordingCleanup(self.config, self.stop_event)
self.record_cleanup.start()
Limit recording retention to available storage (#3942) * Add field and migration for segment size * Store the segment size in db * Add comment * Add default * Fix size parsing * Include segment size in recordings endpoint * Start adding storage maintainer * Add storage maintainer and calculate average sizes * Update comment * Store segment and hour avg sizes per camera * Formatting * Keep track of total segment and hour averages * Remove unused files * Cleanup 2 hours of recordings at a time * Formatting * Fix bug * Round segment size * Cleanup some comments * Handle case where segments are not deleted on initial run or is only retained segments * Improve cleanup log * Formatting * Fix typo and improve logging * Catch case where no recordings exist for camera * Specifically define sort * Handle edge case for cameras that only record part time * Increase definition of part time recorder * Remove warning about not supported storage based retention * Add note about storage based retention to recording docs * Add tests for storage maintenance calculation and cleanup * Format tests * Don't run for a camera with no recording segments * Get size of file from cache * Rework camera stats to be more efficient * Remove total and other inefficencies * Rewrite storage cleanup logic to be much more efficient * Fix existing tests * Fix bugs from tests * Add another test * Improve logging * Formatting * Set back correct loop time * Update name * Update comment * Only include segments that have a nonzero size * Catch case where camera has 0 nonzero segment durations * Add test to cover zero bandwidth migration case * Fix test * Incorrect boolean logic * Formatting * Explicity re-define iterator
2022-10-09 14:28:26 +03:00
def start_storage_maintainer(self) -> None:
self.storage_maintainer = StorageMaintainer(self.config, self.stop_event)
self.storage_maintainer.start()
def start_stats_emitter(self) -> None:
2021-02-17 16:23:32 +03:00
self.stats_emitter = StatsEmitter(
self.config,
stats_init(
self.config,
self.camera_metrics,
self.embeddings_metrics,
self.detectors,
self.processes,
),
2021-02-17 16:23:32 +03:00
self.stop_event,
)
self.stats_emitter.start()
def start_watchdog(self) -> None:
2020-11-04 15:28:07 +03:00
self.frigate_watchdog = FrigateWatchdog(self.detectors, self.stop_event)
# (attribute on self, key in self.processes, factory)
specs: list[tuple[str, str, Callable[[], FrigateProcess]]] = [
(
"embedding_process",
"embeddings",
lambda: EmbeddingProcess(
self.config, self.embeddings_metrics, self.stop_event
),
),
(
"recording_process",
"recording",
lambda: RecordProcess(self.config, self.stop_event),
),
(
"review_segment_process",
"review_segment",
lambda: ReviewProcess(self.config, self.stop_event),
),
(
"output_processor",
"output",
lambda: OutputProcess(self.config, self.stop_event),
),
]
for attr, key, factory in specs:
if not hasattr(self, attr):
continue
def on_restart(
proc: FrigateProcess, _attr: str = attr, _key: str = key
) -> None:
setattr(self, _attr, proc)
self.processes[_key] = proc.pid or 0
self.frigate_watchdog.register(
key, getattr(self, attr), factory, on_restart
)
2020-11-04 15:28:07 +03:00
self.frigate_watchdog.start()
def init_auth(self) -> None:
if self.config.auth.enabled:
if User.select().count() == 0:
password = secrets.token_hex(16)
password_hash = hash_password(
password, iterations=self.config.auth.hash_iterations
)
User.insert(
{
User.username: "admin",
2025-03-10 06:59:07 +03:00
User.role: "admin",
User.password_hash: password_hash,
User.notification_tokens: [],
}
).execute()
self.config.auth.admin_first_time_login = True
logger.info("********************************************************")
logger.info("********************************************************")
logger.info("*** Auth is enabled, but no users exist. ***")
logger.info("*** Created a default user: ***")
logger.info("*** User: admin ***")
logger.info(f"*** Password: {password} ***")
logger.info("********************************************************")
logger.info("********************************************************")
elif self.config.auth.reset_admin_password:
password = secrets.token_hex(16)
password_hash = hash_password(
password, iterations=self.config.auth.hash_iterations
)
User.replace(
username="admin",
role="admin",
password_hash=password_hash,
notification_tokens=[],
).execute()
logger.info("********************************************************")
logger.info("********************************************************")
logger.info("*** Reset admin password set in the config. ***")
logger.info(f"*** Password: {password} ***")
logger.info("********************************************************")
logger.info("********************************************************")
def start(self) -> None:
2021-09-14 06:02:23 +03:00
logger.info(f"Starting Frigate ({VERSION})")
Streamline live view (#9772) * Break out live page * Improving layouts and add chip component * Improve default camera player sizing * Improve live updating * Cleanup and fit figma * Use fixed height * Masonry layout * Fix stuff * Don't force heights * Adjust scaling * Cleanup * remove sidebar (#9731) * remove sidebar * keep sidebar on mobile for now and add icons * Fix revalidation * Cleanup * Cleanup width * Add chips for activity on cameras * Remove dashboard from header * Use Inter font (#9735) * Show still image when no activity is occurring * remove unused search params * add playing check for webrtc * Don't use grid at all for single column * Fix height on mobile * a few style updates to better match figma (#9745) * Remove active objects when they become stationary * Move to sidebar only and make settings separate component * Fix layout * Animate visibility of chips * Sidebar is full screen * Fix tall aspect ratio cameras * Fix complicated aspect logic * remove * Adjust thumbnail aspect and add text * margin on single column layout * Smaller event thumb text * Simplify basic image view * Only show the red dot when camera is recording * Improve typing for camera toggles * animate chips with react-transition-group (#9763) * don't flash when going to still image * revalidate * tooltips and active tracking outline (#9766) * tooltips * fix tooltip provider and add active tracking outline * remove unused icon * remove figma comment * Get live mode working for jsmpeg * add small gradient below timeago on event thumbnails (#9767) * Create live mode hook and make sure jsmpeg can be used * Enforce env var * Use print * Remove unstable * Add tooltips to thumbnails * Put back vite * Format * Update web/src/components/player/JSMpegPlayer.tsx --------- Co-authored-by: Josh Hawkins <32435876+hawkeye217@users.noreply.github.com> Co-authored-by: Blake Blackshear <blake@frigate.video>
2024-02-10 15:30:53 +03:00
# Ensure global state.
self.ensure_dirs()
# Set soft file limits.
set_file_limit()
# Start frigate services.
self.init_debug_replay_manager()
self.init_camera_metrics()
self.init_queues()
self.init_database()
self.init_onvif()
self.init_recording_manager()
self.init_review_segment_manager()
self.init_go2rtc()
self.init_embeddings_manager()
self.bind_database()
self.check_db_data_migrations()
# Clean up any stale replay camera artifacts (filesystem + DB)
cleanup_replay_cameras()
self.init_inter_process_communicator()
Use Fork-Server As Spawn Method (#18682) * Set runtime * Use count correctly * Don't assume camera sizes * Use separate zmq proxy for object detection * Correct order * Use forkserver * Only store PID instead of entire process reference * Cleanup * Catch correct errors * Fix typing * Remove before_run from process util The before_run never actually ran because: You're right to suspect an issue with before_run not being called and a potential deadlock. The way you've implemented the run_wrapper using __getattribute__ for the run method of BaseProcess is a common pitfall in Python's multiprocessing, especially when combined with how multiprocessing.Process works internally. Here's a breakdown of why before_run isn't being called and why you might be experiencing a deadlock: The Problem: __getattribute__ and Process Serialization When you create a multiprocessing.Process object and call start(), the multiprocessing module needs to serialize the process object (or at least enough of it to re-create the process in the new interpreter). It then pickles this serialized object and sends it to the newly spawned process. The issue with your __getattribute__ implementation for run is that: run is retrieved during serialization: When multiprocessing tries to pickle your Process object to send to the new process, it will likely access the run attribute. This triggers your __getattribute__ wrapper, which then tries to bind run_wrapper to self. run_wrapper is bound to the parent process's self: The run_wrapper closure, when created in the parent process, captures the self (the Process instance) from the parent's memory space. Deserialization creates a new object: In the child process, a new Process object is created by deserializing the pickled data. However, the run_wrapper method that was pickled still holds a reference to the self from the parent process. This is a subtle but critical distinction. The child's run is not your wrapped run: When the child process starts, it internally calls its own run method. Because of the serialization and deserialization process, the run method that's ultimately executed in the child process is the original multiprocessing.Process.run or the Process.run if you had directly overridden it. Your __getattribute__ magic, which wraps run, isn't correctly applied to the Process object within the child's context. * Cleanup * Logging bugfix (#18465) * use mp Manager to handle logging queues A Python bug (https://github.com/python/cpython/issues/91555) was preventing logs from the embeddings maintainer process from printing. The bug is fixed in Python 3.14, but a viable workaround is to use the multiprocessing Manager, which better manages mp queues and causes the logging to work correctly. * consolidate * fix typing * Fix typing * Use global log queue * Move to using process for logging * Convert camera tracking to process * Add more processes * Finalize process * Cleanup * Cleanup typing * Formatting * Remove daemon --------- Co-authored-by: Josh Hawkins <32435876+hawkeye217@users.noreply.github.com>
2025-06-12 21:12:34 +03:00
self.start_detectors()
self.init_dispatcher()
Camera profile support (#22482) * add CameraProfileConfig model for named config overrides * add profiles field to CameraConfig * add active_profile field to FrigateConfig Runtime-only field excluded from YAML serialization, tracks which profile is currently active. * add ProfileManager for profile activation and persistence Handles snapshotting base configs, applying profile overrides via deep_merge + apply_section_update, publishing ZMQ updates, and persisting active profile to /config/.active_profile. * add profile API endpoints (GET /profiles, GET/PUT /profile) * add MQTT and dispatcher integration for profiles - Subscribe to frigate/profile/set MQTT topic - Publish profile/state and profiles/available on connect - Add _on_profile_command handler to dispatcher - Broadcast active profile state on WebSocket connect * wire ProfileManager into app startup and FastAPI - Create ProfileManager after dispatcher init - Restore persisted profile on startup - Pass dispatcher and profile_manager to FastAPI app * add tests for invalid profile values and keys Tests that Pydantic rejects: invalid field values (fps: "not_a_number"), unknown section keys (ffmpeg in profile), invalid nested values, and invalid profiles in full config parsing. * formatting * fix CameraLiveConfig JSON serialization error on profile activation refactor _publish_updates to only publish ZMQ updates for sections that actually changed, not all sections on affected cameras. * consolidate * add enabled field to camera profiles for enabling/disabling cameras * add zones support to camera profiles * add frontend profile types, color utility, and config save support * add profile state management and save preview support * add profileName prop to BaseSection for profile-aware config editing * add profile section dropdown and wire into camera settings pages * add per-profile camera enable/disable to Camera Management view * add profiles summary page with card-based layout and fix backend zone comparison bug * add active profile badge to settings toolbar * i18n * add red dot for any pending changes including profiles * profile support for mask and zone editor * fix hidden field validation errors caused by lodash wildcard and schema gaps lodash unset does not support wildcard (*) segments, so hidden fields like filters.*.mask were never stripped from form data, leaving null raw_coordinates that fail RJSF anyOf validation. Add unsetWithWildcard helper and also strip hidden fields from the JSON schema itself as defense-in-depth. * add face_recognition and lpr to profile-eligible sections * move profile dropdown from section panes to settings header * add profiles enable toggle and improve empty state * formatting * tweaks * tweak colors and switch * fix profile save diff, masksAndZones delete, and config sync * ui tweaks * ensure profile manager gets updated config * rename profile settings to ui settings * refactor profilesview and add dots/border colors when overridden * implement an update_config method for profile manager * fix mask deletion * more unique colors * add top-level profiles config section with friendly names * implement profile friendly names and improve profile UI - Add ProfileDefinitionConfig type and profiles field to FrigateConfig - Use ProfilesApiResponse type with friendly_name support throughout - Replace Record<string, unknown> with proper JsonObject/JsonValue types - Add profile creation form matching zone pattern (Zod + NameAndIdFields) - Add pencil icon for renaming profile friendly names in ProfilesView - Move Profiles menu item to first under Camera Configuration - Add activity indicators on save/rename/delete buttons - Display friendly names in CameraManagementView profile selector - Fix duplicate colored dots in management profile dropdown - Fix i18n namespace for overridden base config tooltips - Move profile override deletion from dropdown trash icon to footer button with confirmation dialog, matching Reset to Global pattern - Remove Add Profile from section header dropdown to prevent saving camera overrides before top-level profile definition exists - Clean up newProfiles state after API profile deletion - Refresh profiles SWR cache after saving profile definitions * remove profile badge in settings and add profiles to main menu * use icon only on mobile * change color order * docs * show activity indicator on trash icon while deleting a profile * tweak language * immediately create profiles on backend instead of deferring to Save All * hide restart-required fields when editing a profile section fields that require a restart cannot take effect via profile switching, so they are merged into hiddenFields when profileName is set * show active profile indicator in desktop status bar * fix profile config inheritance bug where Pydantic defaults override base values The /config API was dumping profile overrides with model_dump() which included all Pydantic defaults. When the frontend merged these over the camera's base config, explicitly-set base values were lost. Now profile overrides are re-dumped with exclude_unset=True so only user-specified fields are returned. Also fixes the Save All path generating spurious deletion markers for restart-required fields that are hidden during profile editing but not excluded from the raw data sanitization in prepareSectionSavePayload. * docs tweaks * docs tweak * formatting * formatting * fix typing * fix test pollution test_maintainer was injecting MagicMock() into sys.modules["frigate.config.camera.updater"] at module load time and never restoring it. When the profile tests later imported CameraConfigUpdateEnum and CameraConfigUpdateTopic from that module, they got mock objects instead of the real dataclass/enum, so equality comparisons always failed * remove * fix settings showing profile-merged values when editing base config When a profile is active, the in-memory config contains effective (profile-merged) values. The settings UI was displaying these merged values even when the "Base Config" view was selected. Backend: snapshot pre-profile base configs in ProfileManager and expose them via a `base_config` key in the /api/config camera response when a profile is active. The top-level sections continue to reflect the effective running config. Frontend: read from `base_config` when available in BaseSection, useConfigOverride, useAllCameraOverrides, and prepareSectionSavePayload. Include formData labels in Object/Audio switches widgets so that labels added only by a profile override remain visible when editing that profile. * use rasterized_mask as field makes it easier to exclude from the schema with exclude=True prevents leaking of the field when using model_dump for profiles * fix zones - Fix zone colors not matching across profiles by falling back to base zone color when profile zone data lacks a color field - Use base_config for base-layer values in masks/zones view so profile-merged values don't pollute the base config editing view - Handle zones separately in profile manager snapshot/restore since ZoneConfig requires special serialization (color as private attr, contour generation) - Inherit base zone color and generate contours for profile zone overrides in profile manager * formatting * don't require restart for camera enabled change for profiles * publish camera state when changing profiles * formatting * remove available profiles from mqtt * improve typing
2026-03-19 17:47:57 +03:00
self.init_profile_manager()
Use sqlite-vec extension instead of chromadb for embeddings (#14163) * swap sqlite_vec for chroma in requirements * load sqlite_vec in embeddings manager * remove chroma and revamp Embeddings class for sqlite_vec * manual minilm onnx inference * remove chroma in clip model * migrate api from chroma to sqlite_vec * migrate event cleanup from chroma to sqlite_vec * migrate embedding maintainer from chroma to sqlite_vec * genai description for sqlite_vec * load sqlite_vec in main thread db * extend the SqliteQueueDatabase class and use peewee db.execute_sql * search with Event type for similarity * fix similarity search * install and add comment about transformers * fix normalization * add id filter * clean up * clean up * fully remove chroma and add transformers env var * readd uvicorn for fastapi * readd tokenizer parallelism env var * remove chroma from docs * remove chroma from UI * try removing custom pysqlite3 build * hard code limit * optimize queries * revert explore query * fix query * keep building pysqlite3 * single pass fetch and process * remove unnecessary re-embed * update deps * move SqliteVecQueueDatabase to db directory * make search thumbnail take up full size of results box * improve typing * improve model downloading and add status screen * daemon downloading thread * catch case when semantic search is disabled * fix typing * build sqlite_vec from source * resolve conflict * file permissions * try build deps * remove sources * sources * fix thread start * include git in build * reorder embeddings after detectors are started * build with sqlite amalgamation * non-platform specific * use wget instead of curl * remove unzip -d * remove sqlite_vec from requirements and load the compiled version * fix build * avoid race in db connection * add scale_factor and bias to description zscore normalization
2024-10-07 23:30:45 +03:00
self.init_embeddings_client()
self.start_video_output_processor()
self.start_ptz_autotracker()
2020-11-04 15:28:07 +03:00
self.start_detected_frames_processor()
self.start_camera_processor()
self.start_audio_processor()
self.start_storage_maintainer()
self.start_stats_emitter()
self.start_timeline_processor()
2020-11-04 15:28:07 +03:00
self.start_event_processor()
2020-11-24 16:27:51 +03:00
self.start_event_cleanup()
self.start_record_cleanup()
2020-11-04 15:28:07 +03:00
self.start_watchdog()
2021-02-17 16:23:32 +03:00
self.init_auth()
2020-12-05 20:14:18 +03:00
2021-05-18 08:52:08 +03:00
try:
Frigate HTTP API using FastAPI (#13871) * POC: Added FastAPI with one endpoint (get /logs/service) * POC: Revert error_log * POC: Converted preview related endpoints to FastAPI * POC: Converted two more endpoints to FastAPI * POC: lint * Convert all media endpoints to FastAPI. Added /media prefix (/media/camera && media/events && /media/preview) * Convert all notifications API endpoints to FastAPI * Convert first review API endpoints to FastAPI * Convert remaining review API endpoints to FastAPI * Convert export endpoints to FastAPI * Fix path parameters * Convert events endpoints to FastAPI * Use body for multiple events endpoints * Use body for multiple events endpoints (create and end event) * Convert app endpoints to FastAPI * Convert app endpoints to FastAPI * Convert auth endpoints to FastAPI * Removed flask app in favour of FastAPI app. Implemented FastAPI middleware to check CSRF, connect and disconnect from DB. Added middleware x-forwared-for headers * Added starlette plugin to expose custom headers * Use slowapi as the limiter * Use query parameters for the frame latest endpoint * Use query parameters for the media snapshot.jpg endpoint * Use query parameters for the media MJPEG feed endpoint * Revert initial nginx.conf change * Added missing even_id for /events/search endpoint * Removed left over comment * Use FastAPI TestClient * severity query parameter should be a string * Use the same pattern for all tests * Fix endpoint * Revert media routers to old names. Order routes to make sure the dynamic ones from media.py are only used whenever there's no match on auth/etc * Reverted paths for media on tsx files * Deleted file * Fix test_http to use TestClient * Formatting * Bind timeline to DB * Fix http tests * Replace filename with pathvalidate * Fix latest.ext handling and disable uvicorn access logs * Add cosntraints to api provided values * Formatting * Remove unused * Remove unused * Get rate limiter working --------- Co-authored-by: Nicolas Mowen <nickmowen213@gmail.com>
2024-09-24 16:05:30 +03:00
uvicorn.run(
create_fastapi_app(
self.config,
self.db,
self.embeddings,
self.detected_frames_processor,
self.storage_maintainer,
self.onvif_controller,
self.stats_emitter,
self.event_metadata_updater,
self.inter_config_updater,
self.replay_manager,
Camera profile support (#22482) * add CameraProfileConfig model for named config overrides * add profiles field to CameraConfig * add active_profile field to FrigateConfig Runtime-only field excluded from YAML serialization, tracks which profile is currently active. * add ProfileManager for profile activation and persistence Handles snapshotting base configs, applying profile overrides via deep_merge + apply_section_update, publishing ZMQ updates, and persisting active profile to /config/.active_profile. * add profile API endpoints (GET /profiles, GET/PUT /profile) * add MQTT and dispatcher integration for profiles - Subscribe to frigate/profile/set MQTT topic - Publish profile/state and profiles/available on connect - Add _on_profile_command handler to dispatcher - Broadcast active profile state on WebSocket connect * wire ProfileManager into app startup and FastAPI - Create ProfileManager after dispatcher init - Restore persisted profile on startup - Pass dispatcher and profile_manager to FastAPI app * add tests for invalid profile values and keys Tests that Pydantic rejects: invalid field values (fps: "not_a_number"), unknown section keys (ffmpeg in profile), invalid nested values, and invalid profiles in full config parsing. * formatting * fix CameraLiveConfig JSON serialization error on profile activation refactor _publish_updates to only publish ZMQ updates for sections that actually changed, not all sections on affected cameras. * consolidate * add enabled field to camera profiles for enabling/disabling cameras * add zones support to camera profiles * add frontend profile types, color utility, and config save support * add profile state management and save preview support * add profileName prop to BaseSection for profile-aware config editing * add profile section dropdown and wire into camera settings pages * add per-profile camera enable/disable to Camera Management view * add profiles summary page with card-based layout and fix backend zone comparison bug * add active profile badge to settings toolbar * i18n * add red dot for any pending changes including profiles * profile support for mask and zone editor * fix hidden field validation errors caused by lodash wildcard and schema gaps lodash unset does not support wildcard (*) segments, so hidden fields like filters.*.mask were never stripped from form data, leaving null raw_coordinates that fail RJSF anyOf validation. Add unsetWithWildcard helper and also strip hidden fields from the JSON schema itself as defense-in-depth. * add face_recognition and lpr to profile-eligible sections * move profile dropdown from section panes to settings header * add profiles enable toggle and improve empty state * formatting * tweaks * tweak colors and switch * fix profile save diff, masksAndZones delete, and config sync * ui tweaks * ensure profile manager gets updated config * rename profile settings to ui settings * refactor profilesview and add dots/border colors when overridden * implement an update_config method for profile manager * fix mask deletion * more unique colors * add top-level profiles config section with friendly names * implement profile friendly names and improve profile UI - Add ProfileDefinitionConfig type and profiles field to FrigateConfig - Use ProfilesApiResponse type with friendly_name support throughout - Replace Record<string, unknown> with proper JsonObject/JsonValue types - Add profile creation form matching zone pattern (Zod + NameAndIdFields) - Add pencil icon for renaming profile friendly names in ProfilesView - Move Profiles menu item to first under Camera Configuration - Add activity indicators on save/rename/delete buttons - Display friendly names in CameraManagementView profile selector - Fix duplicate colored dots in management profile dropdown - Fix i18n namespace for overridden base config tooltips - Move profile override deletion from dropdown trash icon to footer button with confirmation dialog, matching Reset to Global pattern - Remove Add Profile from section header dropdown to prevent saving camera overrides before top-level profile definition exists - Clean up newProfiles state after API profile deletion - Refresh profiles SWR cache after saving profile definitions * remove profile badge in settings and add profiles to main menu * use icon only on mobile * change color order * docs * show activity indicator on trash icon while deleting a profile * tweak language * immediately create profiles on backend instead of deferring to Save All * hide restart-required fields when editing a profile section fields that require a restart cannot take effect via profile switching, so they are merged into hiddenFields when profileName is set * show active profile indicator in desktop status bar * fix profile config inheritance bug where Pydantic defaults override base values The /config API was dumping profile overrides with model_dump() which included all Pydantic defaults. When the frontend merged these over the camera's base config, explicitly-set base values were lost. Now profile overrides are re-dumped with exclude_unset=True so only user-specified fields are returned. Also fixes the Save All path generating spurious deletion markers for restart-required fields that are hidden during profile editing but not excluded from the raw data sanitization in prepareSectionSavePayload. * docs tweaks * docs tweak * formatting * formatting * fix typing * fix test pollution test_maintainer was injecting MagicMock() into sys.modules["frigate.config.camera.updater"] at module load time and never restoring it. When the profile tests later imported CameraConfigUpdateEnum and CameraConfigUpdateTopic from that module, they got mock objects instead of the real dataclass/enum, so equality comparisons always failed * remove * fix settings showing profile-merged values when editing base config When a profile is active, the in-memory config contains effective (profile-merged) values. The settings UI was displaying these merged values even when the "Base Config" view was selected. Backend: snapshot pre-profile base configs in ProfileManager and expose them via a `base_config` key in the /api/config camera response when a profile is active. The top-level sections continue to reflect the effective running config. Frontend: read from `base_config` when available in BaseSection, useConfigOverride, useAllCameraOverrides, and prepareSectionSavePayload. Include formData labels in Object/Audio switches widgets so that labels added only by a profile override remain visible when editing that profile. * use rasterized_mask as field makes it easier to exclude from the schema with exclude=True prevents leaking of the field when using model_dump for profiles * fix zones - Fix zone colors not matching across profiles by falling back to base zone color when profile zone data lacks a color field - Use base_config for base-layer values in masks/zones view so profile-merged values don't pollute the base config editing view - Handle zones separately in profile manager snapshot/restore since ZoneConfig requires special serialization (color as private attr, contour generation) - Inherit base zone color and generate contours for profile zone overrides in profile manager * formatting * don't require restart for camera enabled change for profiles * publish camera state when changing profiles * formatting * remove available profiles from mqtt * improve typing
2026-03-19 17:47:57 +03:00
self.dispatcher,
self.profile_manager,
Frigate HTTP API using FastAPI (#13871) * POC: Added FastAPI with one endpoint (get /logs/service) * POC: Revert error_log * POC: Converted preview related endpoints to FastAPI * POC: Converted two more endpoints to FastAPI * POC: lint * Convert all media endpoints to FastAPI. Added /media prefix (/media/camera && media/events && /media/preview) * Convert all notifications API endpoints to FastAPI * Convert first review API endpoints to FastAPI * Convert remaining review API endpoints to FastAPI * Convert export endpoints to FastAPI * Fix path parameters * Convert events endpoints to FastAPI * Use body for multiple events endpoints * Use body for multiple events endpoints (create and end event) * Convert app endpoints to FastAPI * Convert app endpoints to FastAPI * Convert auth endpoints to FastAPI * Removed flask app in favour of FastAPI app. Implemented FastAPI middleware to check CSRF, connect and disconnect from DB. Added middleware x-forwared-for headers * Added starlette plugin to expose custom headers * Use slowapi as the limiter * Use query parameters for the frame latest endpoint * Use query parameters for the media snapshot.jpg endpoint * Use query parameters for the media MJPEG feed endpoint * Revert initial nginx.conf change * Added missing even_id for /events/search endpoint * Removed left over comment * Use FastAPI TestClient * severity query parameter should be a string * Use the same pattern for all tests * Fix endpoint * Revert media routers to old names. Order routes to make sure the dynamic ones from media.py are only used whenever there's no match on auth/etc * Reverted paths for media on tsx files * Deleted file * Fix test_http to use TestClient * Formatting * Bind timeline to DB * Fix http tests * Replace filename with pathvalidate * Fix latest.ext handling and disable uvicorn access logs * Add cosntraints to api provided values * Formatting * Remove unused * Remove unused * Get rate limiter working --------- Co-authored-by: Nicolas Mowen <nickmowen213@gmail.com>
2024-09-24 16:05:30 +03:00
),
host="127.0.0.1",
port=5001,
log_level="error",
)
finally:
self.stop()
2021-02-17 16:23:32 +03:00
def stop(self) -> None:
logger.info("Stopping...")
2025-05-30 05:58:31 +03:00
# used by the docker healthcheck
Path("/dev/shm/.frigate-is-stopping").touch()
# Cancel any running motion search jobs before setting stop_event
stop_all_motion_search_jobs()
2020-11-04 15:28:07 +03:00
self.stop_event.set()
# set an end_time on entries without an end_time before exiting
Event.update(
end_time=datetime.datetime.now().timestamp(), has_snapshot=False
).where(Event.end_time == None).execute()
ReviewSegment.update(end_time=datetime.datetime.now().timestamp()).where(
ReviewSegment.end_time == None
).execute()
# stop the audio process
if self.audio_process:
self.audio_process.terminate()
self.audio_process.join()
# stop the onvif controller
if self.onvif_controller:
self.onvif_controller.close()
# ensure the detectors are done
for detector in self.detectors.values():
detector.stop()
empty_and_close_queue(self.detection_queue)
logger.info("Detection queue closed")
self.detected_frames_processor.join()
empty_and_close_queue(self.detected_frames_queue)
logger.info("Detected frames queue closed")
self.timeline_processor.join()
self.event_processor.join()
empty_and_close_queue(self.timeline_queue)
logger.info("Timeline queue closed")
self.output_processor.terminate()
self.output_processor.join()
self.recording_process.terminate()
self.recording_process.join()
self.review_segment_process.terminate()
self.review_segment_process.join()
self.dispatcher.stop()
self.ptz_autotracker_thread.join()
2020-11-24 16:27:51 +03:00
self.event_cleanup.join()
self.record_cleanup.join()
self.stats_emitter.join()
2020-11-04 15:28:07 +03:00
self.frigate_watchdog.join()
self.camera_maintainer.join()
2021-02-07 17:38:35 +03:00
self.db.stop()
2020-11-04 15:28:07 +03:00
# Save embeddings stats to disk
if self.embeddings:
self.embeddings.stop()
# Stop Communicators
self.inter_process_communicator.stop()
self.inter_config_updater.stop()
self.event_metadata_updater.stop()
self.inter_zmq_proxy.stop()
Use Fork-Server As Spawn Method (#18682) * Set runtime * Use count correctly * Don't assume camera sizes * Use separate zmq proxy for object detection * Correct order * Use forkserver * Only store PID instead of entire process reference * Cleanup * Catch correct errors * Fix typing * Remove before_run from process util The before_run never actually ran because: You're right to suspect an issue with before_run not being called and a potential deadlock. The way you've implemented the run_wrapper using __getattribute__ for the run method of BaseProcess is a common pitfall in Python's multiprocessing, especially when combined with how multiprocessing.Process works internally. Here's a breakdown of why before_run isn't being called and why you might be experiencing a deadlock: The Problem: __getattribute__ and Process Serialization When you create a multiprocessing.Process object and call start(), the multiprocessing module needs to serialize the process object (or at least enough of it to re-create the process in the new interpreter). It then pickles this serialized object and sends it to the newly spawned process. The issue with your __getattribute__ implementation for run is that: run is retrieved during serialization: When multiprocessing tries to pickle your Process object to send to the new process, it will likely access the run attribute. This triggers your __getattribute__ wrapper, which then tries to bind run_wrapper to self. run_wrapper is bound to the parent process's self: The run_wrapper closure, when created in the parent process, captures the self (the Process instance) from the parent's memory space. Deserialization creates a new object: In the child process, a new Process object is created by deserializing the pickled data. However, the run_wrapper method that was pickled still holds a reference to the self from the parent process. This is a subtle but critical distinction. The child's run is not your wrapped run: When the child process starts, it internally calls its own run method. Because of the serialization and deserialization process, the run method that's ultimately executed in the child process is the original multiprocessing.Process.run or the Process.run if you had directly overridden it. Your __getattribute__ magic, which wraps run, isn't correctly applied to the Process object within the child's context. * Cleanup * Logging bugfix (#18465) * use mp Manager to handle logging queues A Python bug (https://github.com/python/cpython/issues/91555) was preventing logs from the embeddings maintainer process from printing. The bug is fixed in Python 3.14, but a viable workaround is to use the multiprocessing Manager, which better manages mp queues and causes the logging to work correctly. * consolidate * fix typing * Fix typing * Use global log queue * Move to using process for logging * Convert camera tracking to process * Add more processes * Finalize process * Cleanup * Cleanup typing * Formatting * Remove daemon --------- Co-authored-by: Josh Hawkins <32435876+hawkeye217@users.noreply.github.com>
2025-06-12 21:12:34 +03:00
self.detection_proxy.stop()
2020-11-04 15:28:07 +03:00
while len(self.detection_shms) > 0:
shm = self.detection_shms.pop()
shm.close()
shm.unlink()
_stop_logging()
Use Fork-Server As Spawn Method (#18682) * Set runtime * Use count correctly * Don't assume camera sizes * Use separate zmq proxy for object detection * Correct order * Use forkserver * Only store PID instead of entire process reference * Cleanup * Catch correct errors * Fix typing * Remove before_run from process util The before_run never actually ran because: You're right to suspect an issue with before_run not being called and a potential deadlock. The way you've implemented the run_wrapper using __getattribute__ for the run method of BaseProcess is a common pitfall in Python's multiprocessing, especially when combined with how multiprocessing.Process works internally. Here's a breakdown of why before_run isn't being called and why you might be experiencing a deadlock: The Problem: __getattribute__ and Process Serialization When you create a multiprocessing.Process object and call start(), the multiprocessing module needs to serialize the process object (or at least enough of it to re-create the process in the new interpreter). It then pickles this serialized object and sends it to the newly spawned process. The issue with your __getattribute__ implementation for run is that: run is retrieved during serialization: When multiprocessing tries to pickle your Process object to send to the new process, it will likely access the run attribute. This triggers your __getattribute__ wrapper, which then tries to bind run_wrapper to self. run_wrapper is bound to the parent process's self: The run_wrapper closure, when created in the parent process, captures the self (the Process instance) from the parent's memory space. Deserialization creates a new object: In the child process, a new Process object is created by deserializing the pickled data. However, the run_wrapper method that was pickled still holds a reference to the self from the parent process. This is a subtle but critical distinction. The child's run is not your wrapped run: When the child process starts, it internally calls its own run method. Because of the serialization and deserialization process, the run method that's ultimately executed in the child process is the original multiprocessing.Process.run or the Process.run if you had directly overridden it. Your __getattribute__ magic, which wraps run, isn't correctly applied to the Process object within the child's context. * Cleanup * Logging bugfix (#18465) * use mp Manager to handle logging queues A Python bug (https://github.com/python/cpython/issues/91555) was preventing logs from the embeddings maintainer process from printing. The bug is fixed in Python 3.14, but a viable workaround is to use the multiprocessing Manager, which better manages mp queues and causes the logging to work correctly. * consolidate * fix typing * Fix typing * Use global log queue * Move to using process for logging * Convert camera tracking to process * Add more processes * Finalize process * Cleanup * Cleanup typing * Formatting * Remove daemon --------- Co-authored-by: Josh Hawkins <32435876+hawkeye217@users.noreply.github.com>
2025-06-12 21:12:34 +03:00
self.metrics_manager.shutdown()