Skip to content

fix(prod): reconcile master hotpatch drift into monorepo - #3

Open
alpha1122x wants to merge 12 commits into
feat/ac-attestation-provenance-multiaccountfrom
fix/prod-hotpatch-reconciliation
Open

fix(prod): reconcile master hotpatch drift into monorepo#3
alpha1122x wants to merge 12 commits into
feat/ac-attestation-provenance-multiaccountfrom
fix/prod-hotpatch-reconciliation

Conversation

@alpha1122x

Copy link
Copy Markdown
Owner

Summary

  • Reconcile production bind-mount hotpatches from master base-master-prod into the monorepo so the next image rebuild does not silently lose live truth.
  • Land only bucket B (prod-ahead) code that is real product logic; leave true bidirectional D conflicts unresolved; skip host-only policy/secrets.

Source of truth

  • Host: 86.38.238.235 (read-only)
  • Mounts: live docker inspect on base-master-prod-base-master-validator-1 (37 unique sources; dual mounts counted once)
  • Base branch: feat/ac-attestation-provenance-multiaccount @ 6d9c7565

Hotpatch classification table

Hotpatch Bucket Decision
ac_authorization.py D not-landed — bidirectional
ac_benchmarks.py D not-landed
ac_canonical_compose.py D not-landed
ac_compose.py B landed — site-packages REPO_ROOT
ac_container_builder.py A identical
ac_docker_review/phala_pre_launch.sh A identical
ac_eval_wire.py D not-landed
ac_evaluation_runner.py A identical
ac_key_release_endpoint.py A identical
ac_lifecycle.py B landed — gateway-free analyzer
ac_miner_env.py A identical
ac_review_deployment.py A identical (prod aligned with repo install)
ac_review_runtime.py D not-landed
ac_sdk_config.py D not-landed
ac_selfdeploy_bundle/cli.py D not-landed
ac_selfdeploy_bundle/eval.py D not-landed
ac_selfdeploy_bundle/lifecycle.py B* not-landed — D-coupled to shapes
ac_selfdeploy_bundle/phala.py B* not-landed — D-coupled
ac_selfdeploy_bundle/review.py D not-landed
ac_selfdeploy_bundle/shapes.py D not-landed
ac_tbench_integrity.py A identical
ac_urls.py A identical
lium_training/cli_app/main.py B landed
lium_training/compute/lium.py B landed — training GPU lock
lium_training/compute/lium_capacity.py B landed NEW
lium_training/compute/lium_orphan.py B landed NEW
lium_training/compute/lium_training_wiring.py B landed NEW
lium_training/config/settings.py B landed
lium_training/constation/allowlist_repository.py B landed
lium_training/master/orchestration.py B landed
lium_training/prism/config.py B landed
lium_training/prism/evaluator/checkpoint_intake.py B landed
lium_training/prism/evaluator/checkpoint_publisher.py host not-landed — upload-off + host repo id
lium_training/prism/evaluator/pod_boot.py B landed NEW
prism_cpu_reexec/base-master-entrypoint host/B partial — embed.env only; no host Prism defaults
prism_plagiarism_llm/plagiarism_adjudicator.py B landed NEW
prism_plagiarism_llm/queue.py D not-landed

Supporting image-resident constation modules

Landed from container (not bind-mounted) because hotpatched main imports them and they were missing on this branch tip: custody_keys, attestation_keys, orchestrator, pod_binding, bundle_seal, prod routes + __init__.

Dead weight (on disk, not mounted)

ac_api_routes, ac_execution_pool, ac_progress*, superseded ac_selfdeploy_*, measurements.py, app_proxy, root entrypoint, prism_cpu_reexec/* (except mounted entrypoint), prism_{db,models,repository,routes}, overridden prism_plagiarism_llm/config.py, service/settings/weights*, secrets/flags/baks.

Bucket D unresolved

Do not invent merges for: ac_eval_wire, ac_selfdeploy_bundle/{cli,eval,review,shapes}, ac_authorization, ac_sdk_config, ac_canonical_compose, ac_benchmarks, ac_review_runtime, prism_plagiarism_llm/queue.

Test plan

  • AC pytest gate from packages/challenges/agent-challenge
  • Blackbox 17 failed / 19 passed
  • embed.env unit tests 5 passed
  • ruff clean on changed files
  • Secret scan clean on full diff

Gate result

57 failed, 3262 passed, 9 skipped, 6 errors

failed≤57 ✓ errors≤6 ✓ skipped=9 ✓ · passed 3262 (this base tip collects 3334 tests; 57+3262+9+6=3334)

Commits

6b807c2e style: ruff-fix imports and line length on landed hotpatches
663c9a8a test(agent-challenge): align lifecycle tests with gateway-free analyzer
666f4b36 feat(prism): add plagiarism LLM settings and checkpoint intake status
be4e62bf feat(prism): add pod_boot helpers and plagiarism LLM adjudicator
976a4a40 fix(agent-challenge): resolve review assets under site-packages and gateway-free analyzer
9eac481b fix(master): load per-challenge embed.env into isolated child env
e00c1709 feat(cli): wire constation runtime and Lium capacity scheduler into master
0bda6db4 feat(master): stamp constation pins and tick Lium capacity in orchestration
f4518404 feat(constation): land prod allowlist pin API and custody runtime modules
4302db4a feat(config): add ConstationSettings and LiumTrainingSettings
1417e8dc feat(compute): add Lium capacity scheduler and orphan reconciliation
8ec278d9 fix(compute): lock Lium prism training to 1-GPU Blackwell offers

HEAD: 6b807c2eae9974535eb0f1090d8c29046e185356

Notes

  • Read-only against production.
  • Host checkpoint publisher + entrypoint Prism policy defaults omitted on purpose.

echobt added 12 commits July 28, 2026 06:29
Prod hotpatch refuses multi-GPU hosts that reject gpu_count=1 splits and
exposes LiumClient.for_prism_training for the capacity scheduler path.
Bring prod hotpatch modules for master-owned training admission
(LiumCapacityScheduler) and orphan pod cleanup, plus settings wiring helpers.
Prod hotpatch settings surface for fail-closed constation orchestration
and optional master-owned Lium training admission (default off).
…ules

Hotpatched allowlist_repository gains get_active_pin and identity payload
mapping. Also restore image-resident custody/orchestrator/pod_binding/
attestation_keys/bundle_seal/routes pieces required by the prod main
wiring but missing from this branch tip.
…ration

Prod hotpatch adds ConstationPinSource stamping on Prism dispatch and
optional LiumCapacityAdmission.tick during run_once.
…aster

Prod hotpatch builds custody runtime, pre-forward hook, pin source, and
try_build_lium_capacity_scheduler when constructing the orchestration driver.
Prod entrypoint hotpatch (and fix/master-embed-ac-env-file) restore the
operator extension point dropped by env -i isolation so attestation and
review switches from embed.env actually reach embedded challenges.
Host-only Prism policy defaults are intentionally not copied.
…ateway-free analyzer

Prod hotpatches: REPO_ROOT finds docker/review when installed in
site-packages; host analyzer skips missing Base LLM gateway parking and
completes AST+similarity under gateway-free attested product mode.
Prod hotpatch modules: safe pod boot env/install planning, and the
OpenRouter dual-gate plagiarism adjudicator (wired when queue lands).
Prod hotpatch: PrismSettings plagiarism_llm_* / openrouter fields, and
CheckpointIntakeService last_status observability with fail-closed publish.
Prod hotpatch no longer parks missing Base LLM gateway as llm_standby.
Update the two contracts to expect first-pass allow and no requeue loop.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants