Skip to main content

AEROSIM-TS-165

Project task

Keep required pull-request CI below ten minutes

Required exact-head pull-request CI reports one authoritative result within a measured p95 queue-to-result budget below ten minutes while superseded runs stop consuming shared runners.

AEROSIM-TS-165Canonical ID TASK-0165
Verified flow state
Unverified
Owner
AeroSim Architecture and Delivery
Component
Gitea Actions — pull-request validation, scheduling, and runner isolation
Repository
corp-v1-aerosim/corp-v1-aerosim

Delivery scope

Separate fast merge validation from full release qualification; run independent lint, type, contract, unit/integration, build, image-contract, deterministic simulation, and short production-browser smoke gates in parallel; aggregate them fail closed; cancel superseded pull-request heads; isolate latency-sensitive and release-qualification runners; and publish p50/p95 queue, execution, flake, and resource measurements.

Implementation contract

Implementation artifacts

  • .gitea/workflows/application-validation.yaml
  • .gitea/workflows/feature-branch-push.yaml
  • tests/ci/application-validation-workflow.test.mjs
  • tests/ci/superseded-run-cancellation.test.mjs
  • tests/ci/ci-latency-budget.test.mjs

Inputs

  • TASK-0164 exact-candidate qualification and publication evidence boundary.
  • Observed pull-request validation durations of 111–249 minutes and obsolete-run startup delays up to 65.5 minutes.
  • Deterministic simulation and short browser-smoke candidates proven against the existing full release gate during an overlap period.

Outputs

  • One required exact-head PR aggregate with p95 queue-to-result below ten minutes and explicit merge-only semantics.
  • Prompt cancellation of superseded branch and PR runs with retained diagnostic evidence.
  • Separate runner capacity and telemetry for fast PR feedback and full release qualification.

Failure boundaries

  • Fail when the p95 budget excludes queue time, uses retries to hide flakes, or lacks a representative measurement window.
  • Fail when a superseded run continues expensive jobs after a newer exact head becomes authoritative.
  • Fail when the fast lane is treated as release qualification or omits a required merge-quality gate.
  • Fail when added workers share ports, state, artifacts, credentials, or caches without proven isolation.

Excluded scope

  • The fast lane does not replace full real-time release UAT, release approval, publication verification, or human acceptance.
  • No runner-capacity or GPU change is accepted without measured p50/p95, resource, renderer, and flake evidence.

Verification steps

  • Measure at least the agreed representative run window and require p95 queue-to-result below ten minutes with no masked retries.
  • Push two revisions to one PR and verify the older run terminates promptly while the newer exact head completes one full aggregate.
  • Inject failures into every required fast-lane gate and verify the aggregate fails closed.
  • Compare deterministic/smoke failures with the full release gate during an overlap period before removing full UAT from the PR critical path.

Traceability

Requirements

Dependencies

UI/UX applicability

Unclassified

Acceptance evidence

Required future evidence: a representative exact-head CI sample must demonstrate p95 queue-to-result below ten minutes, prompt superseded-run cancellation, complete fail-closed aggregation, no shared-state leakage, and preserved separate release qualification.

Current evidence boundary

No current implementation, acceptance, release, or deployment evidence is claimed for this planned Task. Any prior implementation may be used only as prototype and discovery evidence.