fix(producer): withhold de parallel router below a machine ram floor

[vanceingalls]

What

Adds a machine-RAM floor to the DE parallel router. shouldPreferParallelDrawElement now also requires totalMemoryMb >= minMemoryMb (default 24576 MB, override via HF_DE_PARALLEL_MIN_MEM_MB, 0 disables the guard). Single-worker DE (the inversion) is unaffected — only the parallel bet is withheld below the floor.

Why

Wild report on CLI 0.7.52: a 16 GB darwin machine hit vertical black slabs on the second render with HF_DE_PARALLEL_ROUTER on; disabling the flag fixed it. Parallel DE runs 3 concurrent hardware-GPU Chrome instances — on a 16 GB machine that's plausible compositor-tile eviction under memory pressure. Every clean routed render seen in telemetry so far has been on machines ≥24 GB.

Validation

4 new unit tests: routes at the floor, withholds below it, minMemoryMb<=0 disables the guard entirely. Full producer test suite (129 tests) green.

🤖 Generated with Claude Code

[vanceingalls]

• fix(cli): surface video frame-injection failures in snapshot output #2255
• fix(shader-transitions): pin cloned scene roots to composition dimensions #2254
• fix(cli): fold canvas pixel hashes into the frozen-sweep fingerprint #2253
• chore(producer): resync stale committed puppeteer-core symlinks to lockfile version #2259
• fix(producer): disk-headroom gate before hdr raw frame pre-extraction #2256
• feat(cli): preview --browser-no-gpu for gpu-unstable hosts #2252
• fix(cli): settle post-seek font subset loads before audit screenshots #2251
• fix(engine): densify drawelement self-verify with parallel worker count #2250
• fix(producer): withhold de parallel router below a machine ram floor #2249 👈 (View in Graphite)
• fix(producer,engine): gate drawelement on ancestor background-image + tail verify sample #2247
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[miga-heygen]

#2249 — withhold de parallel router below a machine ram floor

Verdict: LGTM 🟢

RAM floor for parallel drawElement routing: 3 concurrent hardware-GPU Chrome instances on a 16 GB machine produced vertical black slabs (wild 0.7.52 report). Default floor 24 GB — telemetry-backed (every clean routed cohort ≥24 GB). Env override HF_DE_PARALLEL_MIN_MEM_MB (0 disables). Single-worker DE stays available below the floor; only the parallel bet is withheld.

SSOT: the guard is a single boolean term in the existing shouldPreferParallelDrawElement predicate — no duplicated decision. Tests cover below-floor, at-floor, and guard-disabled cases. Clean.

Review by Miga

[miguel-heygen]

Reviewed current head 4c02c8f. RAM units/default/override are explicit and the guard is limited to the parallel DE router; tests cover below/at floor and disabled guard. Required functional checks are green; Graphite mergeability is still pending only. APPROVE.

[james-russo-rames-d-jusso]

Reviewed at 4c02c8f.

Small clean predicate change with matching env override, sane default (24 GB matches the wild-report boundary + your telemetry cohort), and the disable path (minMemoryMb <= 0) is tested explicitly. Only the parallel bet is withheld; single-worker DE stays available below the floor — that's the right inversion.

One clarifying question, not a blocker:

Route-decline reason surfacing. shouldPreferParallelDrawElement still returns a plain bool. perfSummary.drawElement.parallelRouter captures the routed/not-routed outcome, but not WHICH gate declined — RAM floor vs short comp vs CSS-effect-risk vs experimental opt-in all collapse to false. For FF-flip observability on this floor specifically (is 24 GB the right knob?), threading a parallelRouterDeclineReason through — same shape as the existing deClampReason — would let you see the RAM-rejected cohort in aggregate. Pre-existing gap the whole predicate carries, not this PR's job — but this PR adds a decline reason that would benefit from being observable.

LGTM from my side.

— Review by Rames D Jusso

[jrusso1020]

Approving per the channel stamp-flow: Rames-D and Magi both cleared, CI clean at 4c02c8f.

The base branch was changed.