| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
|
|
Now that the prebuffered stream paints in ~0.7s, the parallel snapshot usually
loses the race and, if slow (a Unifi takePicture cache-miss can be several
seconds), would land after live video and overpaint it with a staler still.
pushSnapshot takes a shouldPaint predicate; startStream passes
() => !firstStreamFrameSeen (flipped by the ffmpeg first-frame handler), so a
snapshot that finishes after the stream is already painting is dropped before
the POST. Preserves the snapshot as a gap-filler for the slow paths (live-edge
fallback / cold prebuffer) where the stream hasn't painted yet.
|
|
|
|
Remove the temporary instrumentation used to diagnose the event-wake and
cold-start work: the evtscan system-wide listener, the attach/trace/streamopt/
src/inputArgs logs, the StartStop entry logs, and ffmpeg -loglevel info/-nostats
(back to error). Keep the lightweight one-shot cold-start stamps (spawned →
first byte → first frame → first socket.write) and usingPrebuffer in the config
line — cheap regression signals — plus all the actual fixes and the prebuffered-
stream selection logic.
|
|
|
|
Event/lifecycle fixes:
- attachListener now self-heals: it's idempotent (tracks the bound camera
per viewport) and re-runs on registerViewport success, so the camera
subscription survives the reload/add storage-attach race that previously
left a bound viewport with no listener until a manual re-save.
- stop() always drains the global cleaner array (never gates on this.running)
and also cancels bindingDebounce timers + clears the stream idle timeout on
abort, so StartStop.stop()+start() == a fresh load with no orphaned ffmpeg,
sockets, listeners, or intervals.
- ignore all camera events while a stream is live or starting (guard moved to
the top of handleCameraEvent) — the wake window is anchored to the first
event; later events don't queue, relaunch, or extend it.
- prune stale childIds whose storage container is gone (deleted device).
Triggers:
- default to person + doorbell (motion opt-in; doorbell cameras are chatty);
doorbell only offered for doorbell-capable cameras. Reconcile stored
triggers + re-render settings live when the camera binding changes.
Cold-start (wake -> first live frame): ~6s -> ~0.7s.
- Root cause: Unifi GOP is ~5s and the rebroadcast RTSP serves live-edge, so
ffmpeg waited for the next keyframe.
- Fix: select the prebuffered substream by id (smallest that covers the panel,
not the 5MP High), request prebuffer >= GOP, and drop -avioflags direct /
-fflags nobuffer on the prebuffer path so ffmpeg gulps the buffered-keyframe
burst instead of trickling it. Live-edge fallback keeps the low-latency flags.
Steady-state g2g unchanged (~50-100ms).
Includes temporary diagnostics (evtscan, attach/trace/streamopt/src logs,
ffmpeg -loglevel info) to be stripped in the follow-up commit.
|
|
|
|
|
|
|
|
broken)
ScryptedDevice.listen() takes a single ScryptedInterface | string |
EventListenerOptions, not an array. We were passing
[BinarySensor, MotionSensor, ObjectDetector] which stringifies to a
comma-joined garbage filter, leaking unrelated events (Settings, ...)
through and apparently dropping BinarySensor entirely — bell rings
never surfaced. Also drops the child-device traversal: confirmed via
unifi-protect/src/main.ts that the bell BinarySensor lives on the
camera device itself for doorbells.
|
|
|
|
|
|
|
|
|
|
|
|
diagnostic
|
|
|
|
Scripts plugin)
The 'Unknown' text the user saw above the Status and Controls panel
is the device TYPE, not the lifecycle state. Scripts plugin
(plugins/core/src/script.ts:65) hardcodes type=ScryptedDeviceType.Unknown
when it registers any script device, regardless of what interfaces
the loaded class actually implements.
Override it from inside the script: call deviceManager.onDeviceDiscovered
ourselves with type=Bridge, which semantically fits our DeviceProvider
that bridges multiple child viewport devices. The override runs after
Scripts plugin's postRunScript-driven discovery so the later call wins.
Pass the full interface set explicitly (auto-detection found from
method names: Settings/DeviceProvider/DeviceCreator/HttpRequestHandler/
StartStop, plus Scripts base Scriptable+Program) — partial lists drop
interfaces.
The override is wrapped in try/catch — if the device record's
provider mapping changes in a future Scrypted version, we degrade
to the old Unknown label rather than the script failing to boot.
|
|
|
|
In v101fb3e we shipped StartStop + OnOff side-by-side to see which
the 'Status and Controls' panel actually wired to. The v44c7a63
diagnostic session confirmed: clicking STOP fires stop() (StartStop),
not turnOff() (OnOff). User's console showed:
lifecycle: start() called (running=false)
...
lifecycle: stop() called (running=true)
stop: tearing down 1 resources
So StartStop alone is the right interface. OnOff just produced a
duplicate 'Status and Controls' panel for the same lifecycle. Drop it.
Also dropping the verbose lifecycle: ... logging — we know the
binding now. The cleanup mechanism is the load-bearing observable
('stop: tearing down N resources' from drainShutdownCleaners).
The status text still rendering 'Unknown' (rather than Running/Stopped)
is a separate cosmetic concern — possibly the duplicate panels were
confusing the UI; with a single StartStop panel left, the displayed
status may now reflect this.running correctly. To be verified.
|
|
|
|
start() previously logged only at entry, so a silent throw between
'Scrypted Viewport up' and 'this.running = true' would leave us
guessing. Add explicit end-of-method logs with the post-write state,
plus a try/catch around bootstrap that surfaces the exception
message before re-throwing.
This will tell us whether start() reaches its set this.running=true
on script load (i.e. whether the prototype state-proxy write fires)
or whether bootstrap is silently failing somewhere mid-way.
|
|
|
|
panel binding
The 'Status and Controls' panel in @scrypted/core 0.3.147 was rendering
Unknown despite the previous commit adding StartStop. Reading the
SDK (sdk/src/index.ts:195 + plugins/core/src/script.ts:47) confirmed
how it should work:
- ScryptedDeviceBase installs Object.defineProperty getters/setters
for every interface property at runtime, so this.running = true
propagates through _lazyLoadDeviceState → getDeviceState proxy →
system state.
- Scripts plugin's mergeHandler auto-detects interfaces by mapping
method names: start/stop → StartStop, turnOn/turnOff → OnOff,
putSetting/getSettings → Settings, etc.
In theory the previous commit was sufficient. To pin down whether the
panel is calling something different (older Scrypted UIs lean toward
OnOff), this commit:
- Implements OnOff alongside StartStop. Both pairs are bound to the
same drain/bootstrap logic; whichever the panel calls, the user's
intent succeeds.
- Initialises this.running = false and this.on = false synchronously
in the constructor so the device state record carries a defined
value at registration time, instead of Scrypted defaulting to
Unknown on undefined.
- Logs 'lifecycle: X() called (...)' on every method entry. After
the user reloads + clicks STOP / START, the console will reveal
exactly which method Scrypted is invoking — or whether neither is.
If one of the methods fires, the other is dead code we can drop. If
neither fires on the panel's STOP/START button, the panel is a
Scripts-runtime control unrelated to device interfaces and we need
to look for a different lifecycle hook.
|
|
|
|
Silent early-return at 'if (!v.cameraId) return;' makes a brand-new
viewport with no camera selected look identical (in the console) to
one that subscribed successfully — there's no positive or negative
signal until you try to fire a camera event. After observing a fresh
viewport produce zero output on a doorbell press, switching the
early-return to a warning that says 'no camera assigned — open
Settings and pick a camera; subscription skipped' so the missing
configuration becomes self-evident.
|
|
|
|
The Scripts 'Status and Controls' panel previously rendered Unknown
because the Provider exposed no lifecycle interface. STOP / START in
that panel didn't do anything useful — Scrypted's only option was
to unload the script entirely, which left the user's re-paste flow
relying on the constructor-time globalThis cleanup hack (and even
that didn't cover in-flight streams until 1ce49d3).
This commit makes the panel real:
class ScryptedViewportProvider ... implements ..., StartStop
- async start() : drain shutdown cleaners, bootstrap, running=true
- async stop() : drain shutdown cleaners, clear viewport+listener+
stream maps, running=false
Both methods are idempotent (no-op when already in target state).
The constructor calls start() automatically so script load still
bootstraps without user action.
The private start() method that did the actual provisioning is
renamed to bootstrap() to free up the public name for the StartStop
contract. registerShutdownCleaners is renamed to drainShutdownCleaners
and now takes a 'reason' label so the log line distinguishes 'start:
tearing down N resources' (previous load) from 'stop: tearing down N
resources' (user-driven).
Effect on the workflow the user described:
1. Click STOP in the Scripts UI → stop() drains every resource
(streams, listeners, intervals) the provider currently holds.
2. Re-paste the new script.
3. New constructor runs → start() drains anything left behind →
bootstrap re-discovers children + re-attaches listeners → ready.
No more accumulating ghosts of prior loads.
|
|
|
|
Scrypted's Scripts sandbox doesn't release a previous load's resources
before constructing a new Provider instance, so anything long-lived
survives a re-paste. Previously we hand-rolled cleanup via two
separate globalThis hooks: __viewportListenerCleaners (camera event
listeners) and __viewportRegisterInterval (the 5-minute re-register
timer). Anything else — in-flight ffmpeg children, TCP sockets to
the firmware, the per-stream stats interval — leaked across reloads.
Visible symptom: re-pasting the script during an active stream left
the old stream's ffmpeg + socket running against a dead Provider
instance, accumulating across reloads until the host was restarted.
Unify all cleanup into one globalThis array, __viewportShutdownCleaners,
that every long-lived resource pushes onto. The next start() snapshots
and drains it before doing anything else. Resources covered:
- 5-minute re-register interval (clearInterval)
- Every camera + child-device event listener (reg.removeListener)
- Every in-flight stream's AbortController (abort.abort, which the
existing abort listeners chain into socket.destroy() + ffmpeg
SIGTERM + clearInterval on the stats logger)
Each cleanup is removed from the array when the resource naturally
ends (stream timeout / stopStream), so the list stays compact across
many stream cycles rather than growing forever.
Snapshot-before-walk avoids the iteration-skip bug where a stream
abort fires its abort listener, which splices the closure out of the
same array we're iterating. Snapshot the array, clear the original,
then walk the snapshot.
Doesn't address the 'Unknown' status panel in Scrypted Scripts UI —
that's a separate cosmetic issue (we don't implement a lifecycle
interface that reports running state).
|
|
|
|
The Settings 'Status (live)' section reaches the device via two
sequential GETs. A transient socket-level failure (httpd worker
pool briefly saturated by the active stream connection, mid-reboot
window, network jitter) used to leave the page showing
'device: offline / unreachable (fetch failed)' even though the
device was fine — a refresh would clear it.
One automatic 250 ms-backoff retry on either GET removes the
sporadic false positive. The 3 s per-request timeout stays the
same, so the worst-case page latency on a genuinely offline device
is ~6.5 s (3 + 0.25 + 3) instead of 3 s — acceptable for a
deliberate Settings open.
|
|
|
|
Unifi doorbell cameras expose the bell button as a child device of
the camera (separate nativeId, its own BinarySensor interface), not
as a property of the camera itself. The previous code did cam.listen()
on the parent camera only, so motion + person events arrived (those
fire on the camera itself) but bell-press events never reached
handleCameraEvent.
HomeKit kept working because Scrypted's HomeKit bridge auto-syncs all
child devices; our plugin was silently the only consumer missing the
event. Confirmed by the trace log in 3b0ab73 staying silent across a
bell press while motion events still printed.
Fix: in attachListener, walk systemManager.getDeviceIds() and pick
any device with providerId === cam.id as an additional listen target.
Subscribe on all of them with the same iface list — listen() no-ops
on unsupported ifaces, so we don't have to introspect each child.
Tracking changed from Map<nativeId, EventListenerRegister> to
Map<nativeId, EventListenerRegister[]> so detach/script-reload cleanup
removes every listener, not just the camera's.
|
|
|
|
painted<sent gap
Document the architecture pivot. The per-frame budget table now
reflects the long-lived TCP stream socket on port 81 (replaced the
HTTP /frame pattern) and the new recv/decode/paint timings:
recv 14-18ms on the wire (pure body), dec ~5ms, paint <35us, with
decode-task idle ~30ms per frame waiting on recv.
The critical change wasn't a TCP tuning knob — it was splitting
recv off its own FreeRTOS task with a 3-buffer PSRAM ring and a
1-deep latest-frame slot to the decode/paint task (d1c8d45). The
single-task loop blocked the socket for 6ms during decode+paint,
which against the IDF-default 5760-byte window forced a stop-go
cycle that capped painted at ~17fps. Raising the window to 65535
made it worse (g2g grew to 17s) because lwIP RX couldn't drain
the bursts and there's no way to skip-oldest on the kernel queue.
The task split eliminated the coupling: recv-task drains
continuously regardless of decode timing, the kernel buffer stays
near-empty, and painted now matches sent at the source rate.
Backlog entries for body shrink and OTA marked done.
|
|
handle_client previously ran recv → decode → paint serially on one
FreeRTOS task. The kernel TCP buffer filled during decode+paint
(~6ms), and against the IDF-default 5760-byte window the sender
naturally stop-go-rate-limited to ~consumption. Raising the window
to 65535 (previous experiment) regressed g2g from ~100ms to 17s
growing unbounded — the sender pumped 45+ segments per round into
a kernel buffer the app couldn't drain in time, and there was no
way to skip-oldest on the kernel queue.
This commit decouples recv from decode+paint:
recv-task: owns the socket. Reads header + body into one of three
preallocated PSRAM body buffers. On body complete, swaps
the just-filled buffer into a 1-deep pending slot and
picks a free buffer for the next recv. If the slot
already held a frame (decode is slow), drops oldest in
place — mirror of the Scrypted-side skip-oldest from
e5acf93.
decode-task: waits on a binary semaphore. On signal, claims pending,
then decodes + paints without holding any shared lock.
Frees its prior buffer implicitly by overwriting
s_decode_idx on the next claim.
3 PSRAM body buffers (~3MB of 28MB free) ensure the invariant
{recv_idx, pending_idx, decode_idx} are pairwise distinct without
ever blocking recv. jpeg_decoder.c grew an alloc_input_buffer helper
+ jpeg_decoder_decode now takes an explicit input pointer so the
stream and http_api snapshot paths don't share scratch.
New stats:
- recv_dropped_oldest: per-window count of pending-slot overwrites
- decode_idle_min/avg/max_us: time decode-task spent waiting on signal
Measurement at IDF-default 5760 window, Unifi medium substream:
before split: recv_avg=32ms recv_max~44ms fps=22-26 (recv blocked
during 6ms decode+paint; chunk_max capped at 5760)
after split: recv_avg=17ms recv_max=18-37ms fps=21-29 steady,
decode_idle_avg=27-40ms (decode mostly waiting),
drop_oldest=0, painted at source rate
The bottleneck moved from 'decode+paint serializes recv' to the
wire's own send rate. Bigger windows are now safe (recv-task drains
continuously, can't bury us), but won't add fps until source rate
goes up — that's a separate conversation.
|
|
regressed g2g
Step-2 experiment from the iterative tuning plan: bumped
CONFIG_LWIP_TCP_WND_DEFAULT 5760 → 65535 (and matching SND_BUF +
RECVMBOX 6 → 16) under the hypothesis that the 4×MSS window
throttled receive throughput.
Instrumentation showed the window WAS the per-call throttle
(recv_chunk_max was exactly 5760 before, grew to 33-60KB after).
But end-to-end got dramatically worse:
- g2g exploded from ~100ms steady to 17 SECONDS growing unbounded
(1964 → 6010 → 10632 → 14725 → 17889ms over consecutive windows).
- painted dropped from ~24fps to 15-20fps.
- recv_avg rose from 32ms steady to 50-90ms with frequent
230-500ms outliers (the retransmit / RTO signature).
- chunk_min fell to 61 bytes, chunk_avg dropped from 3415 to ~2200
— fragmentation as lwIP RX struggled with the bigger bursts.
Root cause: the single recv→decode→paint task serializes; the
kernel buffer fills against ANY window during the ~6ms
decode+paint, but with 65535 bytes available the sender pumps
~45 segments before stopping vs ~4 before. That overruns the
lwIP RX path (RECVMBOX=16 mailboxes, default pbuf pool) → drops
→ retransmits → multi-frame stalls. And Scrypted has no way to
skip-oldest on the kernel queue, so frames just pile up on the
wire / in the ESP32 kernel buffer → g2g grows linearly with time.
Lesson: window size isn't the right knob until the receiver can
drain at line rate independent of decode+paint. Next iteration:
split recv into its own FreeRTOS task with a 1-deep latest-frame
slot to the decode/paint task — only then revisit window.
|
|
handleCameraEvent now logs iface/typeof/data/details for every event
the camera fires on BinarySensor/MotionSensor/ObjectDetector. Helps
diagnose why a Unifi doorbell press isn't waking the viewport while
the same event reaches HomeKit. Temporary — remove once root cause is
confirmed.
|
|
call/chunk stats, SO_RCVBUF probe)
Per-frame samples aggregated over the existing 30-frame window:
- queued_at_body_start (FIONREAD just before body recv loop): how much
of the frame the kernel already absorbed during the previous
decode+paint. Close to jpeg_len → wire delivered the full frame
while we were busy (we're decode/paint-bound). Much smaller →
wire is throttled (window or buffer too small to absorb a frame
in our paint window).
- recv_calls: number of recv() syscalls the body read needed per
frame. High → small chunks → window-throttled sender.
- recv_chunk min/avg/max: bytes returned per recv() return in the
window. Avg = window body bytes / total syscalls.
- SO_RCVBUF: one-shot getsockopt at accept, logged and stashed in
stats. Confirms whether sdkconfig values reached the build —
TCP_WND_DEFAULT discrepancies are otherwise invisible.
All surfaced in the windowed log and in /state JSON alongside the
existing recv/dec/paint/idle stats. No behavior change yet.
|
|
Hold at most one pending frame; new ffmpeg frames arriving while the
socket is backpressured replace the held one (drop oldest) instead of
queueing. On 'drain', flush the held frame. The in-flight write stays
out of our hands. Steady-state Node-side buffer drops from up to 20MB
to ~1 frame, eliminating the cap-and-reconnect loop as the primary
shedding mechanism — it's now an emergency safety net for stuck
connections that never fire 'drain'.
Capture-time timestamp instead of write-time so a held frame keeps its
true age and g2g reflects any pending-slot hold latency.
New log fields: drop-oldest=N (frames shed from the pending slot this
window) and bp=N% (share of ffmpeg frames in the window that found the
link backpressured at decision time).
|
|
Streams the raw .bin to the inactive ota_0/ota_1 slot via esp_ota_*, flips otadata, replies 200, reboots after 500 ms. Single-shot guarded by atomic_flag (409 on concurrent). CONFIG_BOOTLOADER_APP_ROLLBACK_ENABLE armed: new images boot pending-verify and ota_arm_healthy_timer marks them valid after 30 s of healthy uptime; otherwise the bootloader reverts on next reset. /state gains ota_state.
|
|
|
|
Two reasons the Status (live) section was showing
"offline / unreachable (fetch failed)" intermittently:
1. /state and /config were fetched in parallel, eating both available
httpd sockets simultaneously (Phase 2 dropped max_open_sockets
from 4 to 2). Any other inbound HTTP at the same instant would
either queue or error.
2. The 1.5s timeout was tight given the firmware now juggles a live
stream socket on port 81 (with occasional cap-flush reconnects)
alongside the httpd workers on port 80.
Sequence the two fetches and bump the timeout to 3s. Total worst-case
6s if both are slow; that's fine for a Settings page, far from
"feels offline."
|
|
|
|
backlog
User's intent: "we want to see what's going on right now, not all the
shit we throttled and backlogged in the buffer." Dropping new arrivals
when the buffer is full doesn't help — the already-queued frames still
get processed in order, painting 5+ seconds of stale content before
catching up to live.
Better: when the buffer exceeds the configured cap, blow away the
whole queue by destroying the socket. The firmware's accept_task
picks up the next reconnect within ~500ms and the pipeline restarts
with the freshest ffmpeg frame as seq=1. Trade a sub-second visual
gap for clearing the entire backlog instantly.
Per-viewport setting "Max Scrypted-side buffer (MB)" under Display.
Default 20 MB (≈5s buffering ceiling at our measured ~3.5 MB/s
ffmpeg → firmware throughput). Range 1-200. Lower for tighter g2g
at cost of more frequent reconnect gaps under sustained backpressure;
higher to tolerate longer backlogs before flushing.
flushCount tracked per-stream and surfaced in the unified log line
alongside drops, plus node_buf shown vs the cap so the user can see
how close they're running to the threshold.
Demux loop now: if !socketReady → drop. Else if writableLength >
cap*MB → log + sock.destroy() + increment flushes. Else normal write.
|
|
|
|
Adds two values to the unified stream log line:
node_buf=NKB≈Mms
node_buf bytes is sock.writableLength — bytes that socket.write()
accepted from us but the kernel send buffer hasn't pulled yet, so
they sit in Node's internal queue. These have NOT been sent over the
wire; the firmware can't have them.
Dividing by the current send rate (bytes/sec computed from the same
window's bytesSent) gives the buffer depth in milliseconds: how many
seconds of already-extracted-from-ffmpeg frames are stalled at the
script. When g2g shows 7s steady state and node_buf shows e.g.
5000-6000KB ≈ 6500ms, the conclusion is obvious — the buffering
isn't in the firmware or on the wire, it's in Scrypted's Node
process.
No behavior change; pure diagnostic.
|
|
|
|
sent-vs-painted log
#1 — Camera listener leak
Same Scripts-sandbox lifecycle gap that bit us with setInterval (fixed
in 521de7e): cam.listen() returns an EventListenerRegister with a
removeListener() method. The listener is held on the camera plugin's
side, not the script's, so when the old Provider instance gets GC'd
on script reload its registrations stay live with a dead callback.
Field symptom: after re-pasting the script N times, every camera
event arrives N times in handleCameraEvent. Today the streamStarting
guard catches sequential dupes, but two callbacks firing on the same
event within the JS microtask window both check streamStarting
BEFORE either has added their nativeId, both proceed, two
pushSnapshots launch in parallel. They race for the firmware
decoder lock — one paints, one gets 503 — and the user perceives
sharp-fighting-with-itself as "snapshot quality is bad again."
Fix: same globalThis trick as the setInterval. Push a remover
closure onto G.__viewportListenerCleaners every time attachListener
fires; at start() of the new Provider instance, drain the array
calling each remover before creating fresh listeners. Idempotent
across reloads.
#2 — Unified sent-vs-painted log
The user asked: "do we have an understanding in the logging/
instrumentation what the FPS of the stream output is from scrypted
out is vs the actual rendered FPS once the stream is loaded?"
Both numbers existed but in separate log lines (skipLogger every 10s
+ fwPoller every 5s, interleaved in the console). Folded the /state
poll into the 10s skipLogger so there's ONE line per window with
sent fps + painted fps side-by-side plus the gap explicitly labeled:
stream "kitchen": sent=24.2fps painted=22.8fps (fw-skipped=1.4fps,
drops=0) 4.58MB/s sent / 4.29MB/s painted |
socket.write p50=0ms p95=1ms max=5ms backpressured=true |
recv=27776/37237/44470us dec=5788/5991/6626us
paint=30/36/46us idle=164/588/11383us | g2g=142ms
fw-skipped = (sent − painted), how many frames per second the
firmware's FIONREAD skip dropped to keep the panel on the freshest
frame. The g2g value is now meaningful per-frame thanks to the
firmware-side live-update we landed last commit.
Drops the separate fwPoller; one comprehensive log per 10s window.
|
