Miscellaneous fixes (#23032)
Some checks are pending
CI / AMD64 Build (push) Waiting to run
CI / ARM Build (push) Waiting to run
CI / Jetson Jetpack 6 (push) Waiting to run
CI / AMD64 Extra Build (push) Blocked by required conditions
CI / ARM Extra Build (push) Blocked by required conditions
CI / Synaptics Build (push) Blocked by required conditions
CI / Assemble and push default build (push) Blocked by required conditions

* ensure embeddings process restarts after maintainer thread crash

* add docs link to media sync settings

* fix color

Co-authored-by: Copilot <copilot@github.com>

* match link color with other sections

* ensure recording staleness threshold scales with segment_time

* docs tweak

* Fix llama.cpp media marker

* Fix gemini tools call

---------

Co-authored-by: Copilot <copilot@github.com>
Co-authored-by: Nicolas Mowen <nickmowen213@gmail.com>
This commit is contained in:
Josh Hawkins 2026-04-29 17:20:19 -05:00 committed by GitHub
parent a182385618
commit 95b5b89ed9
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
7 changed files with 87 additions and 22 deletions

View File

@ -171,7 +171,7 @@ When choosing images to include in the face training set it is recommended to al
- If it is difficult to make out details in a persons face it will not be helpful in training. - If it is difficult to make out details in a persons face it will not be helpful in training.
- Avoid images with extreme under/over-exposure. - Avoid images with extreme under/over-exposure.
- Avoid blurry / pixelated images. - Avoid blurry / pixelated images.
- Avoid training on infrared (gray-scale). The models are trained on color images and will be able to extract features from gray-scale images. - Avoid training on infrared (gray-scale). The models are trained on color images and will not be able to extract features from gray-scale images.
- Using images of people wearing hats / sunglasses may confuse the model. - Using images of people wearing hats / sunglasses may confuse the model.
- Do not upload too many similar images at the same time, it is recommended to train no more than 4-6 similar images for each person to avoid over-fitting. - Do not upload too many similar images at the same time, it is recommended to train no more than 4-6 similar images for each person to avoid over-fitting.

View File

@ -4,6 +4,7 @@ import base64
import json import json
import logging import logging
import os import os
import sys
import threading import threading
from json.decoder import JSONDecodeError from json.decoder import JSONDecodeError
from multiprocessing.synchronize import Event as MpEvent from multiprocessing.synchronize import Event as MpEvent
@ -52,6 +53,14 @@ class EmbeddingProcess(FrigateProcess):
self.stop_event, self.stop_event,
) )
maintainer.start() maintainer.start()
maintainer.join()
# If the maintainer thread exited but no shutdown was requested, it
# crashed. Surface as a non-zero exit so the watchdog restarts us
# instead of treating the silent thread death as a clean shutdown.
if not self.stop_event.is_set():
logger.error("Embeddings maintainer thread exited unexpectedly")
sys.exit(1)
class EmbeddingsContext: class EmbeddingsContext:

View File

@ -143,15 +143,17 @@ class GeminiClient(GenAIClient):
) )
elif role == "tool": elif role == "tool":
# Handle tool response # Handle tool response
function_response = { response_payload = (
"name": msg.get("name", ""), content if isinstance(content, dict) else {"result": content}
"response": content, )
}
gemini_messages.append( gemini_messages.append(
types.Content( types.Content(
role="function", role="function",
parts=[ parts=[
types.Part.from_function_response(function_response) # type: ignore[misc,call-arg,arg-type] types.Part.from_function_response(
name=msg.get("name", ""),
response=response_payload,
)
], ],
) )
) )
@ -350,15 +352,17 @@ class GeminiClient(GenAIClient):
) )
elif role == "tool": elif role == "tool":
# Handle tool response # Handle tool response
function_response = { response_payload = (
"name": msg.get("name", ""), content if isinstance(content, dict) else {"result": content}
"response": content, )
}
gemini_messages.append( gemini_messages.append(
types.Content( types.Content(
role="function", role="function",
parts=[ parts=[
types.Part.from_function_response(function_response) # type: ignore[misc,call-arg,arg-type] types.Part.from_function_response(
name=msg.get("name", ""),
response=response_payload,
)
], ],
) )
) )

View File

@ -44,6 +44,7 @@ class LlamaCppClient(GenAIClient):
_supports_tools: bool _supports_tools: bool
_image_token_cache: dict[tuple[int, int], int] _image_token_cache: dict[tuple[int, int], int]
_text_baseline_tokens: int | None _text_baseline_tokens: int | None
_media_marker: str
def _init_provider(self) -> str | None: def _init_provider(self) -> str | None:
"""Initialize the client and query model metadata from the server.""" """Initialize the client and query model metadata from the server."""
@ -56,6 +57,7 @@ class LlamaCppClient(GenAIClient):
self._supports_tools = False self._supports_tools = False
self._image_token_cache = {} self._image_token_cache = {}
self._text_baseline_tokens = None self._text_baseline_tokens = None
self._media_marker = "<__media__>"
base_url = ( base_url = (
self.genai_config.base_url.rstrip("/") self.genai_config.base_url.rstrip("/")
@ -141,6 +143,13 @@ class LlamaCppClient(GenAIClient):
chat_caps = props.get("chat_template_caps", {}) chat_caps = props.get("chat_template_caps", {})
self._supports_tools = chat_caps.get("supports_tools", False) self._supports_tools = chat_caps.get("supports_tools", False)
# Media marker for multimodal embeddings; the server randomizes this
# per startup unless LLAMA_MEDIA_MARKER is set, so we must read it
# from /props rather than hardcoding "<__media__>".
media_marker = props.get("media_marker")
if isinstance(media_marker, str) and media_marker:
self._media_marker = media_marker
logger.info( logger.info(
"llama.cpp model '%s' initialized — context: %s, vision: %s, audio: %s, tools: %s", "llama.cpp model '%s' initialized — context: %s, vision: %s, audio: %s, tools: %s",
configured_model, configured_model,
@ -465,10 +474,11 @@ class LlamaCppClient(GenAIClient):
jpeg_bytes = _to_jpeg(img) jpeg_bytes = _to_jpeg(img)
to_encode = jpeg_bytes if jpeg_bytes is not None else img to_encode = jpeg_bytes if jpeg_bytes is not None else img
encoded = base64.b64encode(to_encode).decode("utf-8") encoded = base64.b64encode(to_encode).decode("utf-8")
# prompt_string must contain <__media__> placeholder for image tokenization # prompt_string must contain the server's media marker placeholder.
# The marker is randomized per server startup (read from /props).
content.append( content.append(
{ {
"prompt_string": "<__media__>\n", "prompt_string": f"{self._media_marker}\n",
"multimodal_data": [encoded], # type: ignore[dict-item] "multimodal_data": [encoded], # type: ignore[dict-item]
} }
) )

View File

@ -24,7 +24,7 @@ from frigate.config.camera.updater import (
) )
from frigate.const import PROCESS_PRIORITY_HIGH from frigate.const import PROCESS_PRIORITY_HIGH
from frigate.log import LogPipe from frigate.log import LogPipe
from frigate.util.builtin import EventsPerSecond from frigate.util.builtin import EventsPerSecond, get_ffmpeg_arg_list
from frigate.util.ffmpeg import start_or_restart_ffmpeg, stop_ffmpeg from frigate.util.ffmpeg import start_or_restart_ffmpeg, stop_ffmpeg
from frigate.util.image import ( from frigate.util.image import (
FrameManager, FrameManager,
@ -34,6 +34,23 @@ from frigate.util.process import FrigateProcess
logger = logging.getLogger(__name__) logger = logging.getLogger(__name__)
# all built-in record presets use this segment_time
DEFAULT_RECORD_SEGMENT_TIME = 10
def _get_record_segment_time(config: CameraConfig) -> int:
"""Extract -segment_time from the camera's record output args."""
record_args = get_ffmpeg_arg_list(config.ffmpeg.output_args.record)
if record_args and record_args[0].startswith("preset"):
return DEFAULT_RECORD_SEGMENT_TIME
try:
idx = record_args.index("-segment_time")
return int(record_args[idx + 1])
except (ValueError, IndexError):
return DEFAULT_RECORD_SEGMENT_TIME
def capture_frames( def capture_frames(
ffmpeg_process: sp.Popen[Any], ffmpeg_process: sp.Popen[Any],
@ -164,6 +181,12 @@ class CameraWatchdog(threading.Thread):
self.latest_cache_segment_time: float = 0 self.latest_cache_segment_time: float = 0
self.record_enable_time: datetime | None = None self.record_enable_time: datetime | None = None
# `valid` segments are published with the segment's start time, so the
# gap between consecutive publishes can reach 2 * segment_time. Pad the
# staleness threshold so it's never tighter than that worst case.
segment_time = _get_record_segment_time(self.config)
self.record_stale_threshold = max(120, 2 * segment_time + 30)
# Stall tracking (based on last processed frame) # Stall tracking (based on last processed frame)
self._stall_timestamps: deque[float] = deque() self._stall_timestamps: deque[float] = deque()
self._stall_active: bool = False self._stall_active: bool = False
@ -413,16 +436,17 @@ class CameraWatchdog(threading.Thread):
# ensure segments are still being created and that they have valid video data # ensure segments are still being created and that they have valid video data
# Skip checks during grace period to allow segments to start being created # Skip checks during grace period to allow segments to start being created
stale_window = timedelta(seconds=self.record_stale_threshold)
cache_stale = not in_grace_period and now_utc > ( cache_stale = not in_grace_period and now_utc > (
latest_cache_dt + timedelta(seconds=120) latest_cache_dt + stale_window
) )
valid_stale = not in_grace_period and now_utc > ( valid_stale = not in_grace_period and now_utc > (
latest_valid_dt + timedelta(seconds=120) latest_valid_dt + stale_window
) )
invalid_stale_condition = ( invalid_stale_condition = (
self.latest_invalid_segment_time > 0 self.latest_invalid_segment_time > 0
and not in_grace_period and not in_grace_period
and now_utc > (latest_invalid_dt + timedelta(seconds=120)) and now_utc > (latest_invalid_dt + stale_window)
and self.latest_valid_segment_time and self.latest_valid_segment_time
<= self.latest_invalid_segment_time <= self.latest_invalid_segment_time
) )
@ -439,7 +463,7 @@ class CameraWatchdog(threading.Thread):
) )
self.logger.error( self.logger.error(
f"{reason} for {self.config.name} in the last 120s. Restarting the ffmpeg record process..." f"{reason} for {self.config.name} in the last {self.record_stale_threshold}s. Restarting the ffmpeg record process..."
) )
p["process"] = start_or_restart_ffmpeg( p["process"] = start_or_restart_ffmpeg(
p["cmd"], p["cmd"],

View File

@ -28,6 +28,7 @@ class MonitoredProcess:
restart_timestamps: deque[float] = field( restart_timestamps: deque[float] = field(
default_factory=lambda: deque(maxlen=MAX_RESTARTS) default_factory=lambda: deque(maxlen=MAX_RESTARTS)
) )
clean_exit_logged: bool = False
def is_restarting_too_fast(self, now: float) -> bool: def is_restarting_too_fast(self, now: float) -> bool:
while ( while (
@ -72,7 +73,9 @@ class FrigateWatchdog(threading.Thread):
exitcode = entry.process.exitcode exitcode = entry.process.exitcode
if exitcode == 0: if exitcode == 0:
if not entry.clean_exit_logged:
logger.info("Process %s exited cleanly, not restarting", entry.name) logger.info("Process %s exited cleanly, not restarting", entry.name)
entry.clean_exit_logged = True
return return
logger.warning( logger.warning(

View File

@ -10,13 +10,16 @@ import axios from "axios";
import { toast } from "sonner"; import { toast } from "sonner";
import { useJobStatus } from "@/api/ws"; import { useJobStatus } from "@/api/ws";
import { Switch } from "@/components/ui/switch"; import { Switch } from "@/components/ui/switch";
import { LuCheck, LuX } from "react-icons/lu"; import { LuCheck, LuExternalLink, LuX } from "react-icons/lu";
import { cn } from "@/lib/utils"; import { cn } from "@/lib/utils";
import { formatUnixTimestampToDateTime } from "@/utils/dateUtil"; import { formatUnixTimestampToDateTime } from "@/utils/dateUtil";
import { MediaSyncResults, MediaSyncStats } from "@/types/ws"; import { MediaSyncResults, MediaSyncStats } from "@/types/ws";
import { useDocDomain } from "@/hooks/use-doc-domain";
import { Link } from "react-router-dom";
export default function MediaSyncSettingsView() { export default function MediaSyncSettingsView() {
const { t } = useTranslation("views/settings"); const { t } = useTranslation("views/settings");
const { getLocaleDocUrl } = useDocDomain();
const [selectedMediaTypes, setSelectedMediaTypes] = useState<string[]>([ const [selectedMediaTypes, setSelectedMediaTypes] = useState<string[]>([
"all", "all",
]); ]);
@ -109,13 +112,25 @@ export default function MediaSyncSettingsView() {
<Heading as="h4" className="mb-2 hidden md:block"> <Heading as="h4" className="mb-2 hidden md:block">
{t("maintenance.sync.title")} {t("maintenance.sync.title")}
</Heading> </Heading>
<div className="max-w-6xl"> <div className="max-w-6xl">
<div className="mb-5 mt-2 flex max-w-5xl flex-col gap-2 text-sm text-muted-foreground"> <div className="mb-5 mt-2 flex max-w-5xl flex-col gap-2 text-sm text-muted-foreground">
<p>{t("maintenance.sync.desc")}</p> <p>{t("maintenance.sync.desc")}</p>
</div>
</div>
<div className="flex items-center text-primary-variant">
<Link
to={getLocaleDocUrl(
"configuration/record#syncing-media-files-with-disk",
)}
target="_blank"
rel="noopener noreferrer"
className="inline"
>
{t("readTheDocumentation", { ns: "common" })}
<LuExternalLink className="ml-2 inline-flex size-3" />
</Link>
</div>
</div>
</div>
<div className="space-y-6"> <div className="space-y-6">
{/* Media Types Selection */} {/* Media Types Selection */}
<div> <div>