fix(dash): share one ABR availabilityStart + stop non-audio shards waiting for audio

[ntt0601zcoder]

Problem (audit B-4, MEDIUM)

In an ABR DASH ladder each shard anchored its own availabilityStartTime at its first flush. The audio-packing shard flushes at ~segDur, but a non-audio shard never has audioReady true, so it only flushed at the pairing deadline — ~1 s later. That per-shard AST skew (Δ≈1 s) caused:

• (a) the single published AST is Δ-wrong for every rendition except the slug-sorted winner;
• (b) behindPrevSegEnd paces each shard against its own later AST, so non-audio reps publish every segment Δ late forever (availability over-promise / edge-switch stall risk);
• (c) if the audio shard isn't slug-first, the published AST can change mid-stream — a dynamic-MPD AST-constancy violation.

(On default config Δ≈1 s is absorbed by suggestedPresentationDelay=6 s, hence Medium.)

Fix

Two changes — no StartTicks rebase (timelines are already media-PTS-aligned):

1. Non-audio shards pair on video alone. tryCut now computes audioReady := !p.cfg.PackAudio || (haveAudio && queue.AudioLen()>0), so an ABR non-primary shard cuts at the first IDR instead of waiting out the pairing window for audio it will never pack — collapsing the per-shard availStart skew to inter-shard IDR jitter (≤~150 ms).
2. One ladder-wide AST. New ABRMaster.SharedAST(candidate) returns a single AST, set once by whichever shard flushes first; at its first flush each shard adopts it (p.availStart = ABRMaster.SharedAST(now)). Single-rendition streams have no master and keep their own now. All shards now anchor tfdt + behindPrevSegEnd pacing to the same origin → AST correct for every rendition (a, b) and constant for the session regardless of flush order (c).

Test

TestABRMaster_SharedAST (set-once semantics + zero-candidate guard), TestPackager_NonAudioShardEmitsBeforePairingDeadline (a PackAudio=false shard emits its first segment well within a 30 s pairing window — proving it paired on video alone).

go test -race ./internal/publisher/dash/ green incl. all existing table tests (no timeline regression), golangci-lint 0 issues, full go build ./... green.

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!