Removing wallclock made A/V length drift far worse (audio 11.8% long). The
known-clean config used wallclock + master aresample=async=1; the leading
silence is a standby backlog artifact addressed by the bridge live-edge flush +
record-start audio FIFO drain, not by changing the timestamp source.
The persistent ~2.5s of leading silence was the master aresample=async=1 PADDING
the audio to reconcile a PTS-origin mismatch: video PTS starts at frame 0
(-framerate), but -use_wallclock_as_timestamps stamped the first audio chunk at
its wall-clock arrival time (~2.5s after the ffmpeg graph opened). aresample
filled the gap with silence.
Drop wallclock: audio PTS now comes from the 48kHz sample count starting at 0 —
the same origin as video frame 0 — so the streams align with no pad. The bridge
already hands live audio (backlog flushed on attach), so no rate reference is
needed from wallclock.
The ~2.5s of leading silence at record start was the VHD audio slot QUEUE: while
the recorder is idle (no FIFO reader), the bridge blocks on open(O_WRONLY) but the
board keeps buffering audio slots. When the record ffmpeg attaches, the bridge
streamed that stale backlog first — heard as leading silence and pushing audio
out of alignment with the live video.
On each reader attach, drain slots that lock FAST (already-queued backlog) and
stop at the first lock that takes ~a frame period (= waiting on a live slot), so
the reader is handed the live edge, A/V aligned.
Root cause of 'silent first ~1s then clean' + ~0.5% audio-too-long: in standby
the bridge keeps filling the audio FIFO while the idle-preview consumes only
video, so when recording starts ffmpeg reads a ~0.5s backlog of stale audio,
AND the video-only pre-roll discards video frames the audio never had.
Fix: (1) skip the video-only pre-roll in standby (warm slot = no unstable
frames), (2) drain the audio FIFO non-blocking immediately before ffmpeg opens
it, so audio starts at the live edge aligned with the first real video frame.
The 16ch interleave in the deltacast bridge produced audio at HALF the correct
sample rate (measured 24224 vs 48000 samples/s/ch), which broke A/V sync and
pitch. Per the working baseline (audio was clean before the channel selector),
revert the bridge audio thread to the original single-group 2ch extraction and
the capture-manager audio input to -ac 2 + wallclock + aresample.
KEPT the good fixes: long-GOP HEVC for non-growing (NVENC realtime, no frame
drops) and GPU-only codec list. 16ch/channel-select is shelved for a separate,
properly-validated change.
The pre-roll drained only the video pipe (fc_pipe) while the audio FIFO kept
buffering, so ffmpeg read ~PRE_ROLL_SECONDS of surplus pre-roll audio — making
audio longer than video, which when synced compresses audio ~0.5% (pitch-up,
measured: 2591573 audio samples vs 2579395 expected for the video duration).
In standby the framecache slot is already warm (no unstable startup frames), so
the drain is unnecessary; skipping it lets ffmpeg open video and audio together
from the same instant. Cold on-demand spawns keep the brief drain.
- restore -use_wallclock_as_timestamps on audio input: without it ffmpeg's raw
s16le reader stalled the graph (NVENC idle at 9%, ~half frames dropped). With
it + long-GOP HEVC the encoder runs realtime and A/V length stays locked.
- remove all CPU codec options (prores*, dnxh*, libx264/265) from recorder UI;
GPU NVENC only (hevc_nvenc / h264_nvenc). 3x L4 cluster, no reason for CPU.
- GPU codec defaults in env builders + proxy default h264_nvenc.
The hevc_nvenc codec was hardcoded to all-intra (-force_key_frames expr:1), which
is ~4x the NVENC load. Applied to every recording it exceeded the L4's realtime
budget at 1080p59.94 10-bit -> fc_pipe dropped ~half the frames -> video came out
shorter than the (correct) audio -> A/V drift + pitch-up on playback.
Now all-intra is used ONLY when growing-files is on (where it's required for the
editable head). Normal recordings use efficient long-GOP HEVC (2s GOP, 2 B-frames)
which NVENC sustains in realtime with zero drops.
Removing -use_wallclock_as_timestamps on the SDI audio input. The bridge writes
SDI-clock-paced samples, so PTS from the 48kHz sample count shares the video's
clock domain and the audio length tracks the video length exactly. Wall-clock
timestamps made audio length = real elapsed time, which drifted ~1% longer than
the frame-count video when the encoder dipped under realtime (pitch-up).
Taking the MAX sample count across the 4 audio groups could emit more audio
frames per slot than group 0 (the SDI-clock reference), drifting the audio
stream slightly longer than video — heard as a ~1% pitch-up. Group 0 paces the
timeline exactly as the original 2ch path did; shorter groups are silence-padded
to its length, never extending it.
Recorders are now physical capture ports, not user-created rows:
- migration 036: label, enabled, auto_provisioned + UNIQUE(node_id,device_index)
(the structural fix that makes two recorders sharing a port impossible)
- mam-api: auto-provision one recorder row per port from heartbeat capabilities
(reconcileRecordersForNode); create-once, never overwrites operator config
- mam-api: POST /:id/enable + /:id/disable (provision/teardown standby sidecar);
PATCH accepts label; config persists across enable/disable
- node-agent: freeCapturePort() force-removes any container on a capture port
before standby/start — eliminates the EADDRINUSE collisions
- web-ui: recorder menu grouped by node (online/offline), Enable/Disable toggle,
per-recorder config modal (codec/bitrate/growing/label/project), friendly
label over hardware name, no destructive delete
Fixes the delete/recreate churn that orphaned standby sidecars and collided on
capture ports during this session's outage.
Re-provisions the persistent standby sidecar for SDI/deltacast recorders that
lost theirs (manual cleanup, node redeploy, wiped /dev/shm). Without this the
recorder falls back to slow on-demand spawn on /start, which can collide on the
capture port (EADDRINUSE). Idempotent; { force:true } recreates even when a
container_id is already set.
- capture-manager: remove dead legacy deltacast FIFO video path (FC_SLOT_ID
is now always set by node-agent, framecache mandatory on all SDI nodes)
- node-agent: correct stale comment about legacy FIFO fallback
- onboard-node.sh: harden detect_sdi (device-node checks, not just lspci) and
persist COMPOSE_PROFILES so framecache survives every redeploy on SDI nodes
- remove committed capture.js.bak
Root cause of this session's outage: zampp3 came up without the capture
compose profile, so framecache never started; the bridge published to shm
with no consumer and recorders showed 'receiving' with no real capture.
The framecache ring delivers frame-accurate frames at exactly the SDI clock
rate. -use_wallclock_as_timestamps was wrong for this source — it stamped
frames by ffmpeg arrival time rather than capture time, causing the recorded
file to report wrong framerates (e.g. 56.06 instead of 59.94) and a
glitchy first second at startup (NVENC cold-start backlog bunched timestamps).
Fix: remove -use_wallclock_as_timestamps from the rawvideo (pipe:0) input
and rely on -framerate for correct CFR timestamps from frame 0.
Audio keeps its FIFO wallclock; aresample=async=1 on the master output
resamples audio to align with the CFR video PTS.
Sidecars now spawn at recorder CREATE time instead of /start time.
The container boots in STANDBY=1 mode (idle preview only, no ffmpeg master).
On /start, mam-api sends per-session params (CLIP_NAME, ASSET_ID, PROJECT_ID)
to the running sidecar via HTTP POST /capture/start — ffmpeg starts in <1s.
On /stop, mam-api calls HTTP POST /capture/stop — container stays alive in
standby, ready for the next take immediately.
Container is only killed on recorder DELETE.
This eliminates: Docker create/start overhead (~1-2s), bridge startup (~2-5s),
and pre-roll wait (~5s). Latency from 'record' click to first encoded frame
drops from ~10s to ~1s.
Changes:
- capture/src/index.js: boot in standby when STANDBY=1 env is set; still
start idle preview (live thumbnail visible before recording)
- capture/src/routes/capture.js: POST /start accepts full codec params and
asset_id in body (skips mam-api asset creation when asset_id provided)
- node-agent/index.js: handleSidecarStandby() + POST /sidecar/standby route;
warms bridge at recorder create time
- recorders.js POST /: spawn standby sidecar after DB insert (non-fatal)
- recorders.js POST /:id/start: HTTP fast-path to standby sidecar; falls
back to on-demand spawn if standby not available
- recorders.js POST /:id/stop: HTTP /capture/stop, keep container in standby
- recorders.js GET /:id/status: use port-based URL for local capture status
Reverts the local-temp+faststart approach from 549ca6c. Masters now stream
ffmpeg stdout directly to S3 via multipart upload — no local disk consumed
on the worker. Uses +frag_keyframe+empty_moov+default_base_moof which
Premiere Pro 25.x handles natively (to be confirmed separately).
Zero /tmp/capture files. Worker disk stays flat during recording.
fc_slot_create, fc_slot_destroy, fc_slot_open, fc_slot_close, and
fc_slot_write_frame were defined in slot.c but never declared in slot.h.
Any translation unit calling them without seeing a proper prototype
would fall back to implicit int return (32 bits), truncating 64-bit
pointers and causing SIGSEGV on dereference.
This affected framecache.c (POST /slots → fc_slot_create, DELETE
→ fc_slot_destroy) and other callers.
The struct fc_slot was defined only in slot.c, making it an incomplete type
in slot.h. The inline accessor functions (fc_slot_id, fc_slot_header, etc.)
in slot.h could not compile because they referenced incomplete struct
members. The compiler fell back to implicit int return type, truncating
64-bit pointers to 32 bits, causing SIGSEGV in registry_add() when
strncpy received a truncated slot_id pointer.
Fix: move the struct definition to slot.h and add proper function
declarations for the accessors (definitions stay in slot.c).
