Skip to content

managed runner, deterministic benchmark, and causal evidence (PR3 of #377 split) - #822

Open
BColsey wants to merge 12 commits into
JustVugg:devfrom
BColsey:successor/pr3-runner
Open

managed runner, deterministic benchmark, and causal evidence (PR3 of #377 split)#822
BColsey wants to merge 12 commits into
JustVugg:devfrom
BColsey:successor/pr3-runner

Conversation

@BColsey

@BColsey BColsey commented Aug 4, 2026

Copy link
Copy Markdown

Draft successor — supersedes #377. Reconstructed from the frozen green donor 5f6f31a. Do not merge #377.

Scope

  • c/ramdisk_support/benchmark.py, c/ramdisk_support/runtime_monitor.py
  • Tests: test_ramdisk_benchmark.py, test_ramdisk_runtime_monitor.py, test_ramdisk_benchmark_module.py
  • (Reconstruction adds: supervision hunks of lifecycle.py/processes.py; live tmpfs cells of test_rammap_e2e.py — split from PR1/PR2)

Reconstruction work

  • Kernel-level safe process supervision (cgroup-v2 / pidfd per-descendant) — closes the setsid/env-shedding descendant escape the donor leaves at the cooperative UID+nonce attribution boundary.
  • Causal four-cell PIN/RAMMAP matrix (design L99-123): ≥7 randomized fresh processes/cell, DRAM + file/anonymous/shmem accounting, NUMA placement, paired confidence intervals. The donor's benchmark does not implement this — PR3 supplies a new experiment runner + evidence schema and must not claim evidence it does not produce.
  • runtime_monitor identity binding/revalidation. Managed start/stop JSON (deferred from PR2).

Dependency

Depends on PR2 (#821): benchmark.py imports .common/.accelerator; runtime_monitor imports ramdisk_support.processes (PR2). Started as DRAFT because the causal matrix and kernel-containment supervision are new reconstruction work.

Build verification (subagent worktree)

  • PR3 own modules import cleanly (from ramdisk_support import benchmark, runtime_monitor OK — PR2 .common/.accelerator/.processes deps present on this stacked branch). PR3 own tests GREEN: test_ramdisk_benchmark + test_ramdisk_runtime_monitor + test_ramdisk_benchmark_module = 47 passed / 0 failed. Full discovery hard-aborts only on PR2’s carried bundle self-check; no defect in benchmark/runtime_monitor.

Reconstruction update

PR3 scaffold landed (commit 1be8751): NEW ramdisk_support/supervision.py (cgroup-v2 membership absence check), causal four-cell matrix structure in benchmark.py (neutral, claims nothing yet), runtime_monitor identity binding helpers. Green: 70 passed / 0 failed. Still TODO (large/new): wire supervision into recovery, real causal-matrix runner + thresholds, managed start/stop JSON.


Update — 2026-08-07: forward-merged upstream/dev (two forward-only merge commits; reconstruct history preserved). Conflicts resolved against the green donor 3658a5d; merged dev @ aa46e7b then @ 8d06bd1. Compiles clean, targeted C tests pass, test_openai_server 134 OK, and the merge introduces zero new failures vs the pre-merge tip. GitHub reports this branch MERGEABLE into dev.

⚠️ Pre-existing (not merge-induced): this draft's coli launcher still requires frontend modules (ramdisk_ui.py/ramdisk_textual.py/curses_ui.py) it does not ship, so import coli raises SystemExit and ~13 test modules fail — identical on the pre-merge tip. The headless-launcher fix belongs to #821; porting it here is follow-up work. This does not affect mergeability.

@JustVugg JustVugg added discussion Proposta / discussione aperta, non un task enhancement New feature or request performance Velocità / tok-s / ottimizzazioni labels Aug 7, 2026
@BColsey
BColsey force-pushed the successor/pr3-runner branch 2 times, most recently from f9bb5f8 to 68d9999 Compare August 9, 2026 12:49
@BColsey

BColsey commented Aug 9, 2026

Copy link
Copy Markdown
Author

Merge order: 3 of 3

Merge this PR only after #820 and #821 have merged, in that order.

Dependency chain: #820#821#822

@BColsey

BColsey commented Aug 9, 2026

Copy link
Copy Markdown
Author

@JustVugg, this PR is ready for review.

The current CI workflow and check workflow are both action_required with no jobs started. Because this is a fork PR, they require approval from a base-repository maintainer before CI can run.

Merge position: 3 of 3 — merge after #820 and #821, in that order.

@BColsey BColsey changed the title managed runner, deterministic benchmark, and causal evidence (PR3 draft of #377 split) managed runner, deterministic benchmark, and causal evidence (PR3 of #377 split) Aug 9, 2026
@BColsey
BColsey marked this pull request as ready for review August 9, 2026 13:10
@BColsey
BColsey force-pushed the successor/pr3-runner branch from 68d9999 to f234d28 Compare August 14, 2026 01:14
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

discussion Proposta / discussione aperta, non un task enhancement New feature or request performance Velocità / tok-s / ottimizzazioni

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants