Skip to content

feat(workers): pause intake and drain without charging shutdown deliveries - #2645

Merged
rickylabs merged 21 commits into
mainfrom
feat/worker-pause-drain
Nov 17, 2026
Merged

rickylabs merged 21 commits into
mainfrom
feat/worker-pause-drain

Conversation

@rickylabs

@rickylabs rickylabs commented Nov 16, 2026 •

Copy link
Copy Markdown
Owner

Summary

Add worker pause/resume, a bounded graceful drain and configurable abort cleanup grace. Jobs returned during a controlled shutdown retain their delivery-attempt budget; ordinary handler failures continue to charge attempts.

Scope

Closes #2633
Closes #2635

  • Configured total drain defaults25s, with default19s work/5s abort cleanup/1s settlement. Shorter totals derive a smaller default grace. Per-attempt intake pause barriers preserve admitted work, lease renewal and the long-lived worker start.
  • Opt-in NackOptions.preserveDeliveryCount defaults false; atomic/fenced handback for counting adapters, confirmed publish before AMQP acknowledgement. Full decisions in the plan.

Slices

  • S1 Fresh-main research, genuine red tests, backend controls and locked plan.
  • Independent PLAN-EVAL before product code: 8/8 PASS.
  • S2 Core/queue contracts and Redis/Garnet/Postgres receipt settlement.
  • S3a Native Deno KV, KV polling and memory handback.
  • Independent listener-owner PLAN-EVAL PASS.
  • Listener owner policy and original S3 AMQP handback.
  • S4 Worker pause/supervision/drain wiring and lifecycle tests — 1802d73.
  • S5 Config/standard runtime forwarding and real runner cleanup — fefc5e3.
  • S6 Public docs and generated consumers — 667ffa8, 764271b.
  • S7 Full local/runtime gates, independent IMPL-EVAL and pre-transition head checks/thread gate.

Validation

  • Real project configuration/native queue/state/subprocess runtime plus schema/runner12/12pass. Both worker-only and combined startup use authored grace and explicit overrides in both directions, without a static registry;1.25s cleanup awaited when allowed, terminal state remains timeout.
  • Core full112pass/0fail/1existingignore. Queue full215pass/0fail/3existingoptional-backend ignores. All new backend cases execute: eight clean Worker replacements on all seven counting adapters, stable identity/headers/count and final-depth controls. Explicit/actual thrown failures remain charged.
  • Original full Worker/core-cause113/0/0; final startup-pause/supervisor25/25, corrected adjacent fixtures13/13. Final full plugin266pass/0fail/0ignored after startup ownership and fixture repairs.
  • Scoped check/lint/fmt0; quality:gatePASS; doc-lint exact measured baseline unchanged, new config/shutdown0. All three native JSR publish dry-runs and audits pass (existing cardinality/dynamic-import warnings; no new slow types). Canonical prose/API/asset generation and corpus/publish freshness pass. release:preflight/docs:accuracyPASS.
  • Supported full workflow43984548275 SUCCESS on attempt2 at product source2d7aa10448700b46cb1c2dbb4665044eafb60dfa: Postgres124/124, SQLite116/116 +showcase8/8, static/native/cleanup intact. Initial Postgres documented-scaffold db-generate dashboard refresh failed; same symptom was independently reproduced on main earlier. A single unchanged-source failed-job retry passed with all assertions intact; no worker cause inferred or speculative product change.
  • Independent Muse feature IMPL-EVAL PASS at79f35829e covers unchanged lifecycle/drain/config behavior. Subsequent source delta restores only the existing three-line startup log; unchanged guard and adjacent lifecycle suite27/27 pass. Final ready head445002122406a9accc96495c0dc80c72d6f6cc9d has19SUCCESS/13SKIPPED/0failures and review threads0/0. All4owned containers/4volumes removed after maintained preview and exact identity/mount/reference proof; foreign resources retained.

Harness

  • Run dir: .llm/runs/feat-worker-pause-drain--issues-2633-2635/.
  • Locked plan.
  • Feature tier; independent PLAN-EVAL and IMPL-EVAL mandatory, separate session/vendor family.

Drift / Debt

Task callbacks are covered by listener-generation barriers rather than the job-only active map. Portable native nack and AMQP acknowledgement/cancel behavior are necessary for shutdown handback. Focused KV polling extraction reduces the existing811-line adapter. Existing public documentation baseline measured; new diagnostics prohibited. No saga transport/Redis Streams, dependency/lock change or unrelated migration.

Definition of Done

  • Pause/resume stop/restart intake without cancelling admitted work.
  • Configurable drain waits for work then aborts; within-budget completion acknowledged.
  • Shutdown requeue preserves attempts on every counting adapter; actual failures still charge.
  • Abort grace is configurable/forwarded/default documented and >1second cleanup awaited.
  • Public docs/consumers, backend/full runtime gates, independent evaluation and current checks/threads complete.
  • Taxonomy and milestone0.0.10.

Acceptance evidence

issue: 2633
entries:
  - box-index: 1
    evidence: "Schema and both real startup paths, authored/explicit precedence and no-static-registry runtime-budget_test.ts. https://github.com/rickylabs/netscript/pull/2645#issuecomment-6933353414"
  - box-index: 2
    evidence: "Shared core shutdown resolver25s total /5s grace with JSDoc and schema budget tests. https://github.com/rickylabs/netscript/pull/2645#issuecomment-6933353414"
  - box-index: 3
    evidence: "Real1.25s cleanup in standalone runner and both startup paths;12/12contract/runtime tests. https://github.com/rickylabs/netscript/pull/2645#issuecomment-6933353414"
  - box-index: 4
    evidence: "Worker/core READMEs and tune-worker-runtime.md list option/default/precedence; docs accuracy and generated consumers pass. https://github.com/rickylabs/netscript/pull/2645#issuecomment-6933353414"
issue: 2635
entries:
  - box-index: 1
    evidence: "Intake lifecycle tests cover all jobs/tasks/triggers, healthy pause, generation-safe resume, start lifetime and both initialization windows;25/25with supervisor tests. https://github.com/rickylabs/netscript/pull/2645#issuecomment-6933353414"
  - box-index: 2
    evidence: "Successful in-budget jobs/tasks/triggers complete and ack without abort; monotonic phase/total boundary and noncooperative timeout tests. https://github.com/rickylabs/netscript/pull/2645#issuecomment-6933353414"
  - box-index: 3
    evidence: "Eight actual Worker replacements on all seven counting adapters; stable count1/identity/headers and final queue/DLQ depth controls. https://github.com/rickylabs/netscript/pull/2645#issuecomment-6933353414"
  - box-index: 4
    evidence: "Seven-adapter explicit and actual thrown-handler failure controls plus earlier business/deadline/cancel/backoff first-cause controls. https://github.com/rickylabs/netscript/pull/2645#issuecomment-6933353414"
  - box-index: 5
    evidence: "Worker/core/queue READMEs and tune-worker-runtime.md document pause/drain/count rule and provider limits; docs accuracy/consumer freshness pass. https://github.com/rickylabs/netscript/pull/2645#issuecomment-6933353414"

🤖 Generated with Codex

@rickylabs rickylabs added this to the 0.0.10 milestone Nov 16, 2026
@rickylabs rickylabs added ci:skip-e2e area:plugins plugins/* and plugin-core packages type:feat status:research Harness research phase priority:p1 High ci:skip-scaffold Skip the scaffold-static (deno-only) scaffold gate area:queue packages/queue: queue backends, delivery, dead-letter area:workers labels Nov 16, 2026
@rickylabs

Copy link
Copy Markdown
Owner Author

[PHASE: RESEARCH]

Current main findings reproduced before product edits. Core config/default cleanup tests fail; Worker pause is absent and drain aborts immediately. Memory/KV polling/real Redis/Garnet/Postgres shutdown handback increments count1→2, ordinary explicit failure controls count correctly. Actual Redis/Garnet Worker replacement also fails on the first restart. Native Deno KV explicit nack is a no-op, and an actual thrown-handler retry retains count1 instead of2. AMQP baseline cannot reach accounting because its upstream non-native async path double-acks and the broker rejects the tag; native channel/confirmed-handback prerequisites are recorded.

Four new executable test files; product unchanged. Corrected Postgres schema-startup harness race excluded. All five run-created containers (including initial Garnet replacement) and five volumes proven by native mount events removed by exact identity; no foreign cleanup. Research and exact result interpretation.

@rickylabs rickylabs added status:plan Harness plan phase and removed status:research Harness research phase labels Nov 16, 2026
@rickylabs

Copy link
Copy Markdown
Owner Author

[PHASE: PLAN]

Locked feature plan: core-owned lifecycle/clock/budgets,25s total default with19s graceful work/5s cleanup/1s settlement; per-attempt pause barriers retain worker start lifetime, admitted jobs/tasks, lease renewal and concurrency. Explicit shutdown-only count preservation is fenced/idempotent and keeps ordinary failures charged. Shared schema/native project loader and both standard constructors forward the actual runner grace. Confirmed AMQP publication precedes old acknowledgement.

Seven ordered slices, each<30files, explicit risks/publication/debt baselines and backend/runtime/consumer gates. Four new test files are red on unchanged product source and pass lint/fmt. No product implementation before independent feature PLAN-EVAL PASS.

@rickylabs rickylabs added status:plan-eval Awaiting plan evaluation and removed status:plan Harness plan phase labels Nov 16, 2026
@rickylabs

Copy link
Copy Markdown
Owner Author

[PHASE: PLAN-EVAL] [VERDICT: APPROVED]

Independent feature-tier Plan-Gate cycle 1 (of max 2) — PASS at head 89effd610f04deba89f08149f5d36ed822c6eb27. Implementation may begin (S2 onward) after applying the one required plan correction below; no design change is needed.

Route attestation: requested pinned primary glm_5_3/provider_default via OpenCode, options{}; observed exactly that route in a fresh independent session (live matrix lookup confirms feature PLAN-EVAL primary; fallback unused). Generator is Codex gpt-6.1-sol/xhigh (OpenAI vendor family) — separate session and vendor family. No substitution or unasked fallback.

Plan-Gate checklist (8/8 PASS)

  1. Research present and current — research.md baselined at base b0eb079575968e843b781579673c9fe2ae250e07; issue evidence line-verified at head.
  2. Decisions locked — lifecycle/intake, two-phase total-budget arithmetic (25s = 19s work + 5s abort cleanup + 1s settlement; grace = min(5s, floor(T/5)); one monotonic deadline; safe-integer validation), clock port reusing the existing JobRunnerClock seam, config precedence chain, shutdown-first-cause preserved-count contract, per-adapter refund mechanics, stop terminality, PR2641 independence.
  3. Open-decision sweep — no must-resolve choice remains; evaluator-run sweep found no unflagged rework-forcing decision.
  4. Commit slices — S1–S7 ordered, all under 30 files, each naming proofs, proving gates, and files; all slice-named files verified to exist at main (one path correction below).
  5. Risk register — pause/stop barrier separation, task activeJobs gap (verified: task dispatch never enters activeJobs), concurrent start/stop/pause, attribution races with negative controls, refund abuse/stale/duplicate fencing, AMQP broken-baseline honesty, F-1/F-16 sizing (Worker 463, KV polling 811, Postgres 581 baselines confirmed), JSR slow types, resource hygiene.
  6. Gate set selected — Archetype 3 all F-1…F-19 plus queue-integration/workers-plugin overlays per the archetype gate matrix; scoped check (--unstable-kv)/lint/fmt/test wrappers, arch:check + quality:gate, doc-lint against the zero-new-diagnostics baseline, JSR audit + package dry-runs, release:preflight + docs:accuracy + corpus/publish freshness, full backend suites, end-of-run scaffold.runtime on a supported runner, then IMPL-EVAL and current-head CI/review-thread gates.
  7. Deferred scope explicit — PR2641 process lifecycle, SSE2637, all saga transports/Streams, transaction/outbox redesign, transport-wide changes, dependency bumps; divergence-record rule stated.
  8. jsr-audit surface scan — rubric applied to the planned public surface before slicing: explicit owned types, no exported Zod implementation, no new dependency/lock/root-barrel surface, new types only via existing subpaths (core ./shutdown, core config, queue ports), measured doc-lint baseline (combined core2/queue1/plugin2), publish dry-run/audit and consumer contracts in S7.

Findings

  1. R1 (required before S6, plan bookkeeping): S6 names docs/site/background-processing/how-to/tune-worker-runtime/index.md, which does not exist; the actual file is docs/site/background-processing/how-to/tune-worker-runtime.md. Correct the slice file name (keep its [workers] InProcessJobRunner abort grace is 1 s and the generated worker cannot change it #2633 box4 / [workers] Worker has no pause, and a clean stop consumes delivery attempts #2635 box5 mappings) and record the fix in drift.md.
  2. A2 (advisory): In S6, name the generated carriers exactly (agent-docs-prose carrier and export-surface corpus via gen:agent-docs-prose / gen:mcp-export-corpus, plus the publish assets carrier) so S7 freshness checks have unambiguous targets.
  3. A3 (advisory): Decision 1 hands back a committed claim racing pause/drain, while Decision 5 gates preserved-count refunds to the controlled-shutdown first cause. Add one sentence stating that a claim raced by pause is an ordinary charged requeue (only drain/stop preserves), and have S4 tests pin it.

Verification performed

Repository diff at head is exactly four red test files plus run-dir documents — no product source changed. The red tests are executable against unchanged main contracts (verified compile-safety and failure mechanisms: schema strips the budget fields; dispatch settles within 1s before 1250ms cleanup; pause/resume missing; drain aborts at entry; counted shutdown handback on memory/KV polling/Redis/Garnet/Postgres and real Worker first replacement; Deno KV explicit nack no-op and thrown-retry undercount; AMQP broker-406 recorded as a baseline prerequisite, not an accounting result). Ordinary failure controls read green where claimed. All nine acceptance boxes across #2633/#2635 are mapped to concrete, meetable planned evidence.

Acceptance coverage

9/9 meetable — #2633 boxes 1–4 (config + real standard runtime forwarding, documented default derived from the drain budget, >1s cleanup awaited, docs) and #2635 boxes 1–5 (pause/resume, bounded drain with in-budget ack, shutdown requeue preserving attempts on every counting adapter including 8 restarts beyond limit 5, failures still charging, docs).

Next

  • Generator applies R1 (for S6) and may begin S2; A2/A3 are recommended before S4/S6.
  • Verdict written to .llm/runs/feat-worker-pause-drain--issues-2633-2635/plan-eval.md (uncommitted local run artifact, committed with the next slice per protocol).
  • PR at evaluation: required CI policy-skipped (draft skip labels, command-postgres green); no reviews yet. IMPL-EVAL, S7 gates and current-head checks remain required before completion.

PR #2645, head 89effd6, checks and reviews: required CI policy-skipped (command-postgres green), no reviews yet.

@rickylabs rickylabs added status:impl and removed ci:skip-e2e status:plan-eval Awaiting plan evaluation ci:skip-scaffold Skip the scaffold-static (deno-only) scaffold gate labels Nov 16, 2026
@rickylabs

Copy link
Copy Markdown
Owner Author

[PHASE: IMPL]

S2: added core lifecycle/clock/budget contracts and the bounded work→abort→settlement primitive, plus the opt-in shutdown-only nack contract. Redis/Garnet refunds use the existing atomic receipt fence; Postgres atomically refunds/unlocks using per-claim identity and prevents stale ack/release or duplicate settlement. Postgres adapter shrank581→568lines.

Validation: real Redis/Garnet/Postgres28/28 (including repeated handback, ordinary failure/expiry charging, terminal nack, duplicate and stale replacement fences); core budget/Postgres8/8; scoped check/lint/fmt0, new shutdown doclint0, release preflight/docs accuracyPASS. Other adapter and Worker/config red tests remain intentionally pending their slices. Run evidence.

@rickylabs

Copy link
Copy Markdown
Owner Author

[PHASE: IMPL]

S3a implements memory/DenoKV/KV-polling handback and once-only settlement. Native DenoKV now enqueues the retained/counted envelope before native ack; ordinary thrown handlers charge. KV polling uses opaque claim identity + CAS move, guards automatic settlement, reports handback storage failure through its listener and retains the durable processing entry. Adapter811→727lines.

Validation: local retry9/9, local stale/duplicate/terminal/late-context9/9, adjacent lifecycle/DLQ/outage22/22; final combined filtered24/24 after the per-generation controller fix. Check/lint/fmt0, exact per-entrypoint doc-lint baseline unchanged (combined1), preflight/accuracyPASS. The outage test reproduced an uncaught settlement rejection before the fix. Three owned implementation backends retained with recorded identities/mounts; no foreign cleanup.

A separate new red test exposes native post-intake drain expiry:3owner cases fail on Redis/Garnet/KV polling;3default controls pass. This is an explicit contract checkpoint, not a green implementation claim. Proposed ownership addendum awaits independent plan review; added-option code is not implemented.

@rickylabs

Copy link
Copy Markdown
Owner Author

[PHASE: PLAN]

Implementation discovery requires a narrow addition to the approved pause plan: Redis and KV polling resolve their native listener after a visibility-based wait even with an admitted callback still running. Real/scaled50msdrain vs500msheld callback yields3owner reds and3default-control greens. This would end a pause barrier early.

The locked addendum proposes optional provider-specific ListenOptions.drainTimeoutMs: omission preserves existing finite behavior; explicit Infinity delegates callback joining to the worker's already bounded25s total stop. It changes no visibility/claim/retry semantics, adds no commands/dependencies and retains all original gates. Added-option implementation waits for independent feature PLAN-EVAL follow-up.

@rickylabs

Copy link
Copy Markdown
Owner Author

[PHASE: IMPL] [VERDICT: IN_PROGRESS]

At2d7aa1044, S4–S6 are implemented: pause/resume generation barriers for every listener family, one25s total stop with19/5/1 defaults, shutdown-first-cause portable handback, validated project config + explicit runtime overrides, standalone5s grace, public docs and generated consumers.

The actual Worker replacement oracle now covers all seven counting adapters beyond the delivery limit, with identity/header/count and final-depth controls. Queue full215/0/3existingoptional ignores; core112/0/1existingignore; original Worker/core113/0/0; final startup/supervisor25/25 and fixture controls13/13. Real startup/config/runner12/12 includes1.25s handler cleanup and both override directions with no static pool registry.

A held registry lookup exposed the startup/pause ownership gap: red before/after-allocation tests precede immediate ownership transfer. AMQP abort fixture now models the actual native channel and strengthens cleanup ordering; seed fixture requires live directory recreation rather than a stale notification. Full plugin rerun in progress.

Check/lint/fmt, quality:gate, preflight/accuracy, all three native dry-runs/audits and consumer freshness pass; exact public doc-lint baseline retained. Remaining S7: full supported one-pass scaffold runtime, independent feature IMPL-EVAL, cleanup, current required head checks and thread gate. No readiness/merge/publication action.

@rickylabs

Copy link
Copy Markdown
Owner Author

[PHASE: IMPL]

Implemented worker pause/resume, graceful work-before-abort stop with one monotonic total, shutdown-only delivery-count preservation on every counting backend, and authored/explicit grace forwarding through standard/generated startup. Default total25s=19s work+5s cleanup+1s settlement; standalone runner also uses5s grace. Stop is terminal for one Worker instance; parent budget needs at least5s headroom. Public docs and canonical consumers refreshed. Product source2d7aa10448700b46cb1c2dbb4665044eafb60dfa; follow-up33c034cb5 changes run artifacts only.

Completed evidence: structured full plugin266pass/0fail/0ignored; queue215pass/0fail/3unchanged optional-backend ignores; core112pass/0fail/1unchanged ignore. All new backend cases execute. worker-shutdown-deliveries_test.ts and worker-shutdown-adapters_test.ts exercise eight real Worker replacements on all seven adapters, with stable count/identity/header and final-depth controls. shutdown-requeue_test.ts also proves explicit and actual thrown failures charge; first-cause controls keep earlier failure/deadline/cancel charged. worker-intake-lifecycle_test.ts covers all listener families, successful admitted completion/ack, resume generation joins and startup pause before/after allocation; supervisor/intake25/25.

runtime-budget_test.ts + worker-budget_test.ts + abort-cleanup-budget_test.ts:12/12, real startup/no-static-registry, authored and explicit grace both directions, handler cleanup1.25s awaited, timeout never converted into success. New and adjacent fixtures retain their assertions; native AMQP fixture adds exact cleanup order, seed recreation fixture rejects stale notifications.

Check/lint/fmt0; quality:gate/architecture fitnessPASS; release:preflight/docs:accuracyPASS; all three native JSR dry-runs/auditsPASS, no new slow types; doc-lint exact baseline core2/queue1/plugin2 unchanged, new config/shutdown0. Canonical site/prose/corpus/asset generation and consumer freshnessPASS. Docs: worker/core/queue READMEs and tune-worker-runtime.md list options, defaults, lifecycle, count rule, parent headroom and provider uncertainty.

Pending mandatory gates: supported full one-pass scaffold runtime dispatched once at the product source, independent feature IMPL-EVAL, final head checks/thread verification and positively owned backend cleanup. This is still a draft; no merge, auto-merge or publication.

@rickylabs

Copy link
Copy Markdown
Owner Author

[PHASE: IMPL-EVAL] [VERDICT: PASS]

Independent feature IMPL-EVAL PASS at head 79f35829e679e5ff0d2c62d0a1a9a62d2f1ae03e (base b0eb079575968e843b781579673c9fe2ae250e07; product source 2d7aa10448700b46cb1c2dbb4665044eafb60dfa, two follow-up run-doc commits only; tree clean). Requested pinned muse_spark_1_3/xhigh observed exactly; fallback unused; generator Codex gpt-6.1-sol/xhigh is a separate session and vendor family. No helper agents. Prior PLAN-EVAL and listener-owner addendum PASS preserved.

Findings

  1. Implementation complete — pause/resume generation barriers for jobs/tasks/triggers, one monotonic 25s total with derived grace/settlement, healthy paused lifetime, terminal/idempotent stop, and startup ownership before/after allocation during held registry lookup.
  2. Shutdown-only preservation verified — first-cause ShutdownError gates preserveDeliveryCount; earlier business failure/deadline/cancel/backoff/lease loss and duplicate/stale/expired receipts remain charged or terminal. Eight Worker replacements on all seven adapters retain identity/headers/count with final-depth controls; ordinary explicit/thrown failures charge.
  3. All local gates independently green — structured check/lint/fmt 0; quality:gate and arch:check exit 0; doc-lint queue 1 / core 2 / plugin 2 with exact per-entrypoint baseline and new entries 0; native JSR dry-runs Success with no slow types; audits exit 0 with only existing cardinality plus disclosed banner miscount; release:preflight and docs:accuracy PASS; freshness PASS; review threads 0/0.
  4. Full suites independently green — packages/queue 215/0/3 existing ignores; plugins/workers 266/0/0; packages/plugin-workers-core full green with 0 failures; real subprocess/config, intake/supervisor, shutdown, and adjacent fixtures green. Genuine reds and excluded harness mistakes are recorded; no false green.
  5. Boundaries hold — no generic registry, new dependency or lock change, saga transport or Streams edits; worker and adapter sizes within caps; real standard runtime without static registry; explicit override precedence; shared schema/defaults.

External gate (not certified)

Supported one-pass scaffold.runtime remains not green: live read shows overall in progress, Postgres leg failure, sqlite leg in progress, logs unavailable while in progress. No additional smoke launched; no wait or poll. DoD and S7 remain unchecked by design; parent owns final runtime, required-head CI, and cleanup. No ready-merge.

Next

  • Parent triage of the external scaffold.runtime Postgres failure and sqlite completion.
  • Parent-owned final head checks, DoD, and exact-identity resource cleanup.
  • No implementation rework requested by this verdict.

Run dir: .llm/runs/feat-worker-pause-drain--issues-2633-2635/ (evaluate.md left uncommitted).

@rickylabs

Copy link
Copy Markdown
Owner Author

[PHASE: IMPL]

Supported full runtime workflow43984548275 SUCCESS, attempt2, product source2d7aa10448700b46cb1c2dbb4665044eafb60dfa: Postgres124/124, SQLite116/116 +showcase8/8, static/native and all cleanup assertions intact. The first Postgres attempt failed during documented-scaffold db generate dashboard refresh; unchanged-main crossed evidence already records this symptom. Exactly one failed-job retry at unchanged source passed. No guard bypass, assertion removal or inferred worker regression/fix.

Independent feature IMPL-EVAL PASS retained; only run docs changed since evaluated code. All nine acceptance mappings/issue boxes complete. Exact4owned containers and4native recorded volumes removed after maintained preview plus identity/mount/reference proof; zero survivors, foreign resources untouched. Pre-transition head4SUCCESS/21SKIPPED,0failures; thread gate0/0. Body/DoD updated truthfully. Transitioning ready so required non-draft head lanes run; their final results still must be confirmed before task completion.

@rickylabs
rickylabs marked this pull request as ready for review November 17, 2026 01:07
@rickylabs

Copy link
Copy Markdown
Owner Author

[PHASE: IMPL]

Required ready-head CI found one failure among1753pass/1fail/21existingignores: maintained startup-log documentation guard. Locally reproduced1pass/1fail, then restored the exact existing in-process runner startup message accidentally omitted during lifecycle refactoring. Guard unchanged; documentation/intake/supervisor27/27green and scoped check/lint/fmt0, quality:gate/preflight/accuracy PASS. Three logging lines only; no control-flow, budget, generated wiring or API change. Existing independent implementation PASS and supported runtime remain behavior evidence; new-head required CI still must finish green. No resources started.

@rickylabs

Copy link
Copy Markdown
Owner Author

[PHASE: IMPL]

Final ready-for-review head445002122406a9accc96495c0dc80c72d6f6cc9d has19SUCCESS/13SKIPPED/0failures, including the formerly failing test-4 shard; review-thread gate0threads/0unanswered. Independent implementation PASS retained for unchanged lifecycle/drain/config behavior; subsequent delta restores only the exact existing three-line startup log, proven by the unchanged guard and adjacent27/27. Supported full runtime43984548275 attempt2 SUCCESS (Postgres124, SQLite116+showcase8, static/native/cleanup), all4owned containers/4volumes removed. All9acceptance mappings and DoD complete. Ready for independent review; no ready-merge, merge or publication.

@rickylabs

Copy link
Copy Markdown
Owner Author

[PHASE: IMPL]

Old head: 445002122406a9accc96495c0dc80c72d6f6cc9d. New head: b5512b12d105ee86d6239372e4b8ceb3cbf89190. Merged main: 8fe1080614df8b3f88c45960bba5b98b1cc8d455. Delta: merge plus canonical regeneration only; the complete handwritten PR diff against main is byte-identical before and after. No handwritten conflicts or shared handwritten files with #2641 (or the other merged fixes). Only the MCP prose/corpus carriers auto-merged; these and the conflicted publish-assets carrier were regenerated.

Semantic review: worker stop remains 25 seconds within the child wrapper’s 30-second work window plus five-second cleanup reserve. Signal forwarding, health hostname, host disposal and singleton KV closure remain intact; the six composition-boundary files hash-match their respective parents. No behavior repair was made.

Local gates PASS: check/lint/fmt (381 files); full workers (266 passed), queue (215 passed, three pre-existing ignores), core plus #2641 drain (136 passed); zero failures; quality/architecture, carrier chain/freshness, release:preflight, docs:accuracy, README fences; zero unanswered threads. Initial AMQP cookie/Garnet Lua fixture errors were corrected only in owned containers. One 50 ms Garnet lease-extension assertion failed under concurrent suite load; all six isolated cases and the final serial full queue run passed unchanged. All 6 owned containers and 5 original unshared volumes removed, zero survivors. Labels, PR body and issue checkboxes unchanged.

@rickylabs
rickylabs merged commit 87965d8 into main Nov 17, 2026
32 checks passed
@github-actions github-actions Bot added the canary:0.0.10-canary.4 Published NetScript prerelease 0.0.10-canary.4 label Nov 17, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

area:plugins plugins/* and plugin-core packages area:queue packages/queue: queue backends, delivery, dead-letter area:workers canary:0.0.10-canary.4 Published NetScript prerelease 0.0.10-canary.4 priority:p1 High status:shipped type:feat

Projects

None yet

1 participant