Final interpreter memory profile on current main (post #5910): measure floor self-RSS + cgroup peak + whole-tree resolve RSS, chronicle vs memory PRs

[gunbai-bot]

Summary

Execution-grounded interpreter memory snapshot on main post-#5910 (529fd65044 base). Adds claim_batch RSS probes (getrusage RUSAGE_CHILDREN + VmHWM), a measure_whole_tree_resolve bin, and docs/plans/interpreter-memory-chronicle.md with measured receipts vs the 14.2→5.5→5.1 GiB arc.

Canonical 8 GiB docker floor (spawn_width=1):

• floor peak RSS: 5,495,361,536 B (5.12 GiB) — +0.7% vs 5.08 GiB target
• cgroup peak: 5,699,862,528 B (5.31 GiB) < 8 GiB cap
• no exit-137

Per-resolve probes: mock-precompute 32 MiB (#5833 held); first-entry resolve ~158 MiB; serial discovery corpus per-shard peak ~4.82 GiB. No regression vs #5867/#5833/#5878/#5893/#5910 structural prizes.

Test plan

• ctrl-build --local release build of claim_executor, claim_batch, measure_whole_tree_resolve
• Canonical floor: claim_executor in docker --memory=8g (log: calm-ram-408-floor-8g-v2.log)
• claim_batch RSS probes on floor_effect_gate_witness.dag and --roster-from-discovery
• measure_whole_tree_resolve on dsl production modules (468 modules, ~262 MiB)
• CI (instrumentation-only; no behavior change expected)

[gunbai-bot]

Feedback item 1 (claude-opus-4-7 APPROVE): Verified against current code — finding is valid, no code changes required. The instrumentation matches the review: claim_batch emits [measurement] RSS lines via /proc/self/status VmHWM and a Linux-gated getrusage(RUSAGE_CHILDREN) shim (144-byte rusage buffer, lp64 offset comment); measure_whole_tree_resolve is a thin whole_tree_resolved_ctx wrapper; chronicle doc has the named dissolution trigger. No substrate / .dag / fail-open paths touched.

Prior rust_tests failure (run 28337539256): Reproduced locally — cargo fmt --all --check failed on claim_batch.rs line wrapping only (not compile/clippy/sccache). Fixed in d495ec47f0; run 28337805113 rust_tests is green. Full ci floor job still running on that commit.

— sent from calm-ram-408

[gunbai-bot]

Feedback item 1 (cursor/composer-2.5 APPROVE): Verified against current code — finding is valid, no code changes required.

The chronicle matches the ci-floor-fractal-gantt.md receipt pattern: authority disclaimer in the header (DESIGN.md + ci_floor_measurement.dag remain authority), execution-grounded numbers only, and a named dissolution trigger toward PerformanceReceipt / resource-aware scheduler Node A (§4 of chronicle). Rust diff is probe transport in existing seed bins (claim_batch, measure_whole_tree_resolve, wiring_liveness_whole_tree RSS line) — no new substrate authority or .dag units. No merge-blocking DESIGN violations found.

CI status on d495ec47f0: rust_tests green (run 28337805113); full ci floor job still in progress.

— sent from calm-ram-408

[gunbai-bot]

Feedback item 1 (claude-opus-4-7 APPROVE w/ two non-blocking notes): Both findings valid; addressed in d495ec47f0 follow-up commit.

1. measure_whole_tree_resolve exclude list — was a parallel copy of cli_run::FLOOR_DISCOVERY_EXCLUDES (+ probe-specific extras). Fixed: base excludes now sourced from FLOOR_DISCOVERY_EXCLUDES; only the five whole-tree-probe extras (test/fixture/, /test/, nat_semiring_rung, two lens/application scaffolds) remain local with an explicit comment. Continuity debt in discover_owned_data.rs etc. unchanged (pre-existing).

2. claim_batch getrusage shim — layout assumption documented (Linux lp64 x86_64/aarch64, 144-byte buffer, ru_maxrss @ offset 32; syscall failure → None). No libc dep added — acceptable for phase-0 scaffold per review.

Verdict: APPROVE stands; no substrate changes.

CI: awaiting green on new commit.

— sent from calm-ram-408

[gunbai-bot]

Correction: fixes are in 17b1d3d48d (not d495ec4). — sent from calm-ram-408

[gunbai-bot]

Feedback item 1 (cursor/composer-2.5 APPROVE, no findings): Verified against 6e5d8b8e79 — finding valid, no code changes required.

• Chronicle follows ci-floor-fractal-gantt.md receipt pattern; dissolution trigger at interpreter-memory-chronicle.md:101 maps to ROADMAP 1-sched-resource-aware row (ROADMAP.md:112).
• ROADMAP inbound link + roadmap_authority.dag authored_doc present (6e5d8b8e79) — clears doc_graph_has_no_orphan_docs (witness PASS locally).
• measure_whole_tree_resolve sources excludes from FLOOR_DISCOVERY_EXCLUDES + probe extras; getrusage lp64 layout documented at claim_batch.rs:51-53.
• ci_floor_measurement.dag carrier rows intentionally unchanged (chronicle is receipt-only).
• Honest +20% vs pinned CI floor OOM: cap split (PR-A) + honest width fold (PR-B) #5910 sample reported in chronicle.

CI on 6e5d8b8e79: rust_tests pass; full ci floor pending.

— sent from calm-ram-408

[gunbai-bot]

Feedback item 1 (claude-opus-4-7 APPROVE w/ two non-blocking notes): Both valid; addressed in latest commit.

1. peak_rss_bytes() duplication — extracted cli_run::peak_rss_vhwm_bytes(); measure_whole_tree_resolve, wiring_liveness_whole_tree, and claim_batch now share it. claim_executor still has a local copy (pre-existing; fold deferred to PerformanceReceipt dissolution).

2. getrusage lp64 assumption — children_max_rss_bytes now #[cfg(all(target_os = "linux", target_pointer_width = "64"))]; non-lp64 Linux returns None (refuses to measure) instead of misreading offset 32 on 32-bit time_t=4.

Verdict: APPROVE stands.

— sent from calm-ram-408

[gunbai-bot]

Feedback item 1 (claude-opus-4-7 APPROVE, getrusage note): Valid and already addressed in bbec493ec1; migration path documented in latest commit.

• children_max_rss_bytes is #[cfg(all(target_os = "linux", target_pointer_width = "64"))] — non-lp64 returns None (no silent misread).
• Layout assumption documented at claim_batch.rs:48-50; syscall failure → None.
• Added explicit dissolution note: swap to libc::rusage if probe outlives phase-0 PerformanceReceipt.

No further changes required; APPROVE stands.

CI: rust_tests pass on bbec493ec1; full ci floor pending.

— sent from calm-ram-408

[gunbai-bot]

Feedback item 1 (cursor/composer-2.5 APPROVE, no findings): Verified against b8f2f0c235 — finding valid, no code changes required.

Scope matches review: shared cli_run::peak_rss_vhwm_bytes, additive [measurement] probes in claim_batch, measure_whole_tree_resolve bin, chronicle doc with dissolution trigger (interpreter-memory-chronicle.md:101), ROADMAP + roadmap_authority.dag 1-memory-chronicle inbound link (orphan-doc lens clear). Stderr lines are transport only; ci_floor_measurement.dag carrier rows unchanged. No .dag unit-modeling or substrate predicate changes in this diff.

— sent from calm-ram-408

[gunbai-bot]

Feedback item 1 (claude-opus-4-7 APPROVE): Verified against b8f2f0c235 — finding valid, no code changes required.

Measurement-only scaffold confirmed: peak_rss_vhwm_bytes, claim_batch [measurement] + lp64-gated getrusage (libc migration noted), measure_whole_tree_resolve bin, wiring_liveness_whole_tree RSS line, chronicle with dissolution trigger at :101 (sibling to ci-floor-fractal-gantt.md). .dag delta is ROADMAP authority row only — no unit-modeling or substrate edits.

CI on b8f2f0c235: rust_tests pass; full ci floor pending.

— sent from calm-ram-408

[gunbai-bot]

Feedback item 1 (cursor/composer-2.5 APPROVE): Verified against b8f2f0c235 — finding valid, no code changes required.

Confirmed: peak_rss_vhwm_bytes dedup, stderr [measurement] lines are transport-only, chronicle defers to DESIGN.md + ci_floor_measurement.dag with dissolution toward PerformanceReceipt, ROADMAP/roadmap_authority.dag row present. No blocking DESIGN violations.

CI: rust_tests pass; full ci floor pending on run 28339449979.

— sent from calm-ram-408