fix(host/send): a stall-resume frame paces at the proven rate instead of blasting

Stall program T2 (amplification kill), the resume-burst half. The native
pace budget was min(0.9 × time-to-deadline, overflow at ~3× stream rate) —
for steady-state frames the rate term is smaller and decides, but for an
OVERSIZED frame (a capture-stall resume carrying seconds of scene delta, a
cold IDR) the deadline term clamped a multi-interval overflow into the
remainder of ONE: an instantaneous many-×-stream-rate blast that overruns
the socket tx-buffer and loses the very frame that would have ended the
freeze. Field fingerprint across three RDNA4 standby-sink cases:
WSAENOBUFS(10055) + loss_ppm spikes at stall edges, then a recovery-IDR
round trip per retry while the client shows 'current bitrate 0.1'.

The budget is now the overflow's wire time at the pace rate itself
(send_pacing::native_budget, pure + unit-tested), bounded by an absolute
100 ms ceiling so a pathological frame can't park the send thread; the
deadline stays a target, never a license to blast. Steady-state frames
produce byte-identical schedules (the rate term already decided);
PUNKTFUNK_PACE_FACTOR=0 keeps the legacy deadline-only spread; the
GameStream plane's Moonlight-pinned schedule is untouched.

Gates: host clippy --all-targets -D warnings + 9 send_pacing tests green
(linux/amd64 container), fmt clean.

crates/punktfunk-host/src/native/stream.rs

@@ -441,26 +441,27 @@ fn idd_adaptive_enabled() -> bool {
 /// Seal one access unit and send it with MICROBURST pacing (the shared
 /// [`send_pacing`](crate::send_pacing) policy, native parameterization): the first `burst_cap`
 /// bytes go out immediately (one absorbed burst the NIC / socket tx-buffer can swallow), and
-/// only the OVERFLOW beyond that is spread across `min(~90% of the time to deadline, the time
-/// the overflow needs at pace_rate_bps)` in ADAPTIVE chunks — 16 packets at today's rates,
-/// coarsening to at most 64 (the GSO-segment cap) once the rate would otherwise skip every
-/// sub-floor sleep, so ≥1 Gbps frames still pace instead of collapsing into an unpaced blast
-/// (plan Phase 1.2). `burst_cap` `None` = auto: `max(128 KB, this AU's wire bytes / 4)`, so
-/// the burst stays a bounded fraction of a high-rate frame instead of swallowing it whole
-/// (plan Phase 1.3); `Some` = PUNKTFUNK_PACE_BURST_KB pinned an absolute cap. So a
-/// normal-bitrate frame (≤ cap) leaves in one immediate burst at ~0 added latency, while a
-/// genuine IDR / sustained-high-bitrate frame (≫ cap) still spreads — keeping the freeze fix
-/// exactly where it's needed (an unpaced line-rate burst overruns the kernel tx buffer →
-/// EAGAIN drop → under infinite GOP, a freeze until the next keyframe). With no slack
-/// (encode ≈ interval) the budget collapses to 0 and even the overflow goes out immediately,
-/// so this is never slower than unpaced.
+/// only the OVERFLOW beyond that is spread across the time it needs at `pace_rate_bps` in
+/// ADAPTIVE chunks — 16 packets at today's rates, coarsening to at most 64 (the GSO-segment
+/// cap) once the rate would otherwise skip every sub-floor sleep, so ≥1 Gbps frames still pace
+/// instead of collapsing into an unpaced blast (plan Phase 1.2). `burst_cap` `None` = auto:
+/// `max(128 KB, this AU's wire bytes / 4)`, so the burst stays a bounded fraction of a
+/// high-rate frame instead of swallowing it whole (plan Phase 1.3); `Some` =
+/// PUNKTFUNK_PACE_BURST_KB pinned an absolute cap. So a normal-bitrate frame (≤ cap) leaves in
+/// one immediate burst at ~0 added latency, while a genuine IDR / sustained-high-bitrate frame
+/// (≫ cap) still spreads — keeping the freeze fix exactly where it's needed (an unpaced
+/// line-rate burst overruns the kernel tx buffer → EAGAIN drop → under infinite GOP, a freeze
+/// until the next keyframe).
 ///
-/// `pace_rate_bps` (latency plan T1.2) bounds the spread from above: the deadline term alone
-/// smears a big frame's tail across the whole remaining interval (~15 ms at 60 fps) even when
-/// the link could drain it in 2–3 ms. The caller passes ~3× the live encoder bitrate — a rate
-/// the link is proven to carry sustained, so the bounded excursion keeps the anti-freeze
-/// property while the tail leaves as soon as the link plausibly allows. `0` = uncapped
-/// (legacy smoothness-only spread, and the fallback when the bitrate isn't known yet).
+/// `pace_rate_bps` (latency plan T1.2; resume-safe form, stall program T2): the caller passes
+/// ~3× the live encoder bitrate — a rate the link is proven to carry sustained — and the
+/// overflow's wire time at that rate IS the pace budget ([`crate::send_pacing::native_budget`],
+/// [`crate::send_pacing::MAX_PACE_SPREAD`]-bounded). The frame deadline no longer under-cuts
+/// the spread: for a steady-state frame the rate term was the smaller one anyway (tail gone in
+/// a fraction of the interval), and for an oversized frame (stall-resume scene delta, cold
+/// IDR) the old deadline clamp was exactly the line-rate blast → tx-overrun → freeze path this
+/// module exists to prevent. `0` = uncapped legacy deadline-only spread
+/// (PUNKTFUNK_PACE_FACTOR=0, and the fallback when the bitrate isn't known yet).
 #[allow(clippy::too_many_arguments)]
 fn paced_submit(
     session: &mut Session,
@@ -498,34 +499,22 @@ fn pace_sealed(
         chunk: crate::send_pacing::ChunkPolicy::Adaptive { base: 16, max: 64 },
         sleep_floor: std::time::Duration::from_micros(500),
     };
-    // T1.2 rate cap: the overflow's wire time at `pace_rate_bps`. Only the bytes past the
-    // burst pace at all, so only they bound the budget.
+    // T1.2 rate cap, resume-safe form (stall program T2): the overflow's wire time at
+    // `pace_rate_bps` IS the budget — the deadline no longer under-cuts it, so an oversized
+    // frame (a stall-resume scene delta, a cold IDR) paces at the proven 3× rate instead of
+    // collapsing into a line-rate blast that overruns the socket buffer and loses the very
+    // frame that ends a freeze. See `send_pacing::native_budget` for the full argument.
     let overflow_bytes = wire_bytes.saturating_sub(burst_bytes) as u64;
-    let cap = if pace_rate_bps > 0 && overflow_bytes > 0 {
-        std::time::Duration::from_nanos(
-            (overflow_bytes * 8).saturating_mul(1_000_000_000) / pace_rate_bps,
-        )
-    } else {
-        std::time::Duration::MAX
-    };
+    let budget = crate::send_pacing::native_budget(deadline, pace_rate_bps, overflow_bytes);
     // Time the socket handoff per chunk and fold it into the session's SealPerf split — the
     // sleeps between chunks stay excluded, so sock_ns is pure send_gso/sendmmsg time.
     let mut sock_ns = 0u64;
-    let result = crate::send_pacing::pace_frame(
-        &refs,
-        crate::send_pacing::PaceBudget::UntilDeadline {
-            deadline,
-            fraction: 0.9,
-            cap,
-        },
-        &cfg,
-        |chunk| {
-            let t0 = std::time::Instant::now();
-            let r = session.send_sealed(chunk).map(|_| ());
-            sock_ns += t0.elapsed().as_nanos() as u64;
-            r
-        },
-    );
+    let result = crate::send_pacing::pace_frame(&refs, budget, &cfg, |chunk| {
+        let t0 = std::time::Instant::now();
+        let r = session.send_sealed(chunk).map(|_| ());
+        sock_ns += t0.elapsed().as_nanos() as u64;
+        r
+    });
     drop(refs); // release the borrow of `wires` so it can return to the seal pool
     session.reclaim_wires(wires);
     session.note_sock_ns(sock_ns);

crates/punktfunk-host/src/send_pacing.rs

@@ -55,7 +55,7 @@ pub(crate) enum ChunkPolicy {
 }
 
 /// The time the paced (post-burst) packets spread across.
-#[derive(Clone, Copy, Debug)]
+#[derive(Clone, Copy, Debug, PartialEq)]
 pub(crate) enum PaceBudget {
     /// `min((deadline − now-after-burst) × fraction, cap)`, collapsing to 0 with no slack
     /// (native: fraction 0.9). `cap` bounds the spread to the time the overflow actually needs
@@ -68,10 +68,53 @@ pub(crate) enum PaceBudget {
         fraction: f32,
         cap: Duration,
     },
-    /// A precomputed fixed budget (GameStream: ¾ of the frame interval).
+    /// A precomputed fixed budget (GameStream: ¾ of the frame interval; native: the rate-cap
+    /// spread from [`native_budget`]).
     Fixed(Duration),
 }
 
+/// Absolute ceiling on one frame's paced spread (native plane): a pathological frame must not
+/// park the send thread for longer than this, whatever the rate math says. At the ceiling the
+/// tail is late but delivered whole — still strictly better than the blast-loss → freeze →
+/// recovery-IDR round trip it replaces.
+pub(crate) const MAX_PACE_SPREAD: Duration = Duration::from_millis(100);
+
+/// The native plane's pace budget for one frame (pure — unit-tested): with the T1.2 rate cap
+/// active, the paced overflow spreads across exactly the time it needs at the pace rate
+/// (`cap`, bounded by [`MAX_PACE_SPREAD`]) and is NEVER under-cut by the frame deadline.
+///
+/// The old schedule took `min(0.9 × time-to-deadline, cap)`. For a steady-state frame the cap
+/// is the smaller term and nothing changes. But for an OVERSIZED frame — a stall-resume scene
+/// delta after seconds of frozen composition, a cold IDR — the overflow needs SEVERAL frame
+/// intervals at the pace rate, and the deadline term clamped that into the remainder of ONE:
+/// an instantaneous many-×-stream-rate blast that overruns the socket tx-buffer and loses the
+/// very frame that would have ended the freeze (field fingerprint: WSAENOBUFS 10055 +
+/// `loss_ppm` spikes at capture-stall edges, then a recovery-IDR round trip per retry). The
+/// pace rate is ~3× a rate the link demonstrably carries, so holding it past the deadline is
+/// safe by the same argument that introduced the cap — the deadline stays a *target*, not a
+/// license to blast.
+///
+/// `pace_rate_bps == 0` (PUNKTFUNK_PACE_FACTOR=0) or an overflow-free frame keeps the legacy
+/// deadline-only spread.
+pub(crate) fn native_budget(
+    deadline: Instant,
+    pace_rate_bps: u64,
+    overflow_bytes: u64,
+) -> PaceBudget {
+    if pace_rate_bps > 0 && overflow_bytes > 0 {
+        let cap = Duration::from_nanos(
+            (overflow_bytes * 8).saturating_mul(1_000_000_000) / pace_rate_bps,
+        );
+        PaceBudget::Fixed(cap.min(MAX_PACE_SPREAD))
+    } else {
+        PaceBudget::UntilDeadline {
+            deadline,
+            fraction: 0.9,
+            cap: Duration::MAX,
+        }
+    }
+}
+
 /// Per-plane pacing parameters. See the module doc for the two canonical values.
 #[derive(Clone, Copy, Debug)]
 pub(crate) struct PaceCfg {
@@ -598,6 +641,43 @@ mod tests {
         );
     }
 
+    /// [`native_budget`]: with the rate cap active the budget is the overflow's wire time at
+    /// the pace rate — a FIXED spread the deadline can no longer under-cut — bounded by
+    /// [`MAX_PACE_SPREAD`]; rate 0 / no overflow keep the legacy deadline-only schedule.
+    #[test]
+    fn native_budget_is_rate_bound_never_deadline_cut() {
+        // The stall-resume case the fix exists for: a 3 MB overflow at 3×240 Mbps needs
+        // ~33 ms — an IMMINENT deadline (the old min() made this a blast) must not shrink it.
+        let deadline = Instant::now() + Duration::from_millis(4); // 240 fps interval
+        let b = native_budget(deadline, 720_000_000, 3_000_000);
+        assert_eq!(b, PaceBudget::Fixed(Duration::from_nanos(33_333_333)));
+
+        // A steady-state frame: overflow 90 KB at 3×240 Mbps = 1 ms — identical to what the
+        // old min(slack, cap) chose (cap was the smaller term), so nothing regresses.
+        let b = native_budget(deadline, 720_000_000, 90_000);
+        assert_eq!(b, PaceBudget::Fixed(Duration::from_micros(1_000)));
+
+        // A crater-rate resume (ABR backed off to 20 Mbps, pace 60 Mbps): the raw rate math
+        // says 400 ms for 3 MB — the absolute ceiling bounds the send thread's stall.
+        let b = native_budget(deadline, 60_000_000, 3_000_000);
+        assert_eq!(b, PaceBudget::Fixed(MAX_PACE_SPREAD));
+
+        // Rate cap off (PUNKTFUNK_PACE_FACTOR=0): the legacy deadline-only spread, uncapped.
+        let b = native_budget(deadline, 0, 3_000_000);
+        assert!(matches!(
+            b,
+            PaceBudget::UntilDeadline {
+                fraction,
+                cap: Duration::MAX,
+                ..
+            } if fraction == 0.9
+        ));
+
+        // No overflow (the whole frame bursts): budget is never consulted — legacy shape.
+        let b = native_budget(deadline, 720_000_000, 0);
+        assert!(matches!(b, PaceBudget::UntilDeadline { .. }));
+    }
+
     /// `inject_video_drop` is a no-op when the knob is off (the default test env).
     #[test]
     fn drop_injection_off_by_default() {