SnapshotKeeper class — single long-running ffmpeg subprocess на каждый
instance, читает output_rtsp_url непрерывно и dumps latest frame в файл
(/tmp/snapshot-keeper/<instance>.png) каждые 500 ms через ffmpeg fps=2
-update 1. /snapshot endpoint просто serves файл — disk read ~4 ms (было
~5 sec на cold ffmpeg start + RTSP negotiate + keyframe wait).
Auto-restart с exponential backoff при exit (RTSP source перезапустился,
network glitch). Cold ffmpeg fallback остаётся в endpoint — если keeper
ещё не успел dump первый PNG (первые ~1-2 sec после controller start).
UI: snapshot poll interval 700 → 250 ms (4 fps preview, было ~1.4 fps).
Keeper dump rate сейчас 2 fps — practical limit. При желании поднять
до 4 fps — fps=4 в snapshot_keeper.py (~5% доп CPU на ffmpeg PNG encode).
Применение: основной user-facing path = visual overlay editor http://controller:8083/.
Раньше polling 700 ms показывал тот же frame несколько раз пока ffmpeg
запускался. Сейчас preview почти real-time, drag-and-drop reference точный.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
При startup controller iterate'т FrigateBridge mappings и рендерит:
- offline_<cell>.png с camera label из placeholder_label (или frigate_camera
как fallback) — для каждого pad
- offline.png generic — для cells без mapping
Filter cuda_grid looks up "offline_<pad>.png" first per-cell, fallback к
"offline.png". User меняет yaml placeholder_label → restart controller →
PNGs регенерируются. Никаких ручных PNG manipulations.
Russian text supported (PIL + freetype, vs single-byte path в filter).
Pattern:
frigate.mappings:
- frigate_camera: parking_overview
placeholder_label: Парковка
cell: 0
...
При stale input на cell 0 → filter blits offline_0.png с "Парковка — НЕТ
СИГНАЛА" (текст желтый, transparent bg — filter сам fill'ит black за иконкой).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Resilience improvement — раньше pipeline mог hung без exit (NVENC stuck,
output broken pipe), Docker restart policy не triggered. Никакой alert.
Now: poll mediamtx /v3/rtspsessions/list каждые N sec, track publish session
inboundBytes. Не растёт 3 polls (~9 sec) → emit MQTT 'pipeline_stalled' event
(через dispatcher.on_event = mqtt.publish_event). User / Home Assistant
automation решает что делать (restart container, notify).
Wired:
pipeline_monitor.on_event = mqtt.publish_event # __main__.py
Bytes started growing again → emit 'pipeline_unstalled'.
Alert single-shot: пока stalled flag set, no dup alerts. Reset когда
bytes counter растёт.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Pipeline filter state (overlays, layout, cell_map, audio) живёт в RAM
ffmpeg process. При recreate container (compose up, OOM, NVENC crash,
config change) state lost — controller'у нужно re-push.
Раньше user'у приходилось вручную:
curl POST /layout/.../set
docker restart cuda-grid-controller # для browser/dynamic re-register
Теперь автоматизировано:
PipelineMonitor polls ZMQ каждые 3 sec (no-op set_layout).
On timeout/error → mark instance lost.
First success after lost → trigger restore:
1. set_layout к state.active_layout
2. set_audio_output_enabled к state.audio_output_enabled
3. re-push все overlays из state.overlays
4. browser/dynamic/frigate hooks: mark_all_unregistered() —
их loops автоматически re-add на next iteration
Verified test: docker restart cuda-grid-pipeline → 10 sec downtime →
monitor logs lost+restored+restore_done с count=6 overlays.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
FrigateCameraMapping +priority +main_cam_index — для auto-layout decision.
FrigateBridgeCfg.auto_layout flag — toggle через REST.
Logic (FrigateBridge._update_auto_layout):
0 active cameras → quad (default overview)
1+ active → single, main_cam = highest priority active
Equal priority → first active wins (deterministic)
Dispatcher.set_main_cam — ZMQ streamselect@main_cam map <index>
Config.main_cam_filter_target = "streamselect@main_cam"
REST:
GET /auto-layout/{instance} — current toggle state
POST /auto-layout/{instance} — { enabled: bool }
при включении сразу применяет
UI:
+ checkbox "auto" в Layout card — toggleAuto() hits POST /auto-layout
Live verified: enable → immediately picks layout=single, main=gate_lpr
(priority=10, highest active). Visual confirms gate_lpr full screen.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
snapshot_history.py — async periodic capture per instance:
interval_sec / keep_last (FIFO eviction) / dir configurable
GET /snapshots/{instance}?limit=N → list metadata
GET /snapshots/{instance}/{filename} → image bytes
Persisted в /var/lib/cuda-grid/snapshots/{instance}/<ts>.png
layout_map / layout_filter_target в InstanceCfg — для будущей runtime switch
архитектуры (через streamselect либо filter rework — выбор за Phase 7).
Текущий _set_layout dispatches к layout_filter_target c index из map.
UI:
+ Layout buttons (quad/single/main_plus_preview placeholder)
+ Snapshot history grid с thumbnails (loaded /snapshots/{inst}?limit=24)
+ "Reload history" button
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
gx