Skip to content

Finalize deployment consistency and clean-user E2E validation - #9

Open
taha328 wants to merge 2 commits into
mainfrom
codex/final-deployment-consistency-e2e
Open

Finalize deployment consistency and clean-user E2E validation#9
taha328 wants to merge 2 commits into
mainfrom
codex/final-deployment-consistency-e2e

Conversation

@taha328

@taha328 taha328 commented Jun 11, 2026

Copy link
Copy Markdown
Owner

Summary

Align final deployment configuration, consolidate runtime threshold source of truth, and validate the authorized clean-client CPU Docker flow.

Fixes

  • Default root and client PostGIS to imresamu/postgis:16-3.4.
  • Expose and inject canonical threshold env variables.
  • Remove active legacy threshold settings and stale model artifact references.
  • Align deployment backend image reference to cpu-v0.1.2.

Threshold source of truth

APP_CHANGE_THRESHOLD and APP_SEMANTIC_THRESHOLD are the only runtime sources. Request fields remain accepted but cannot override settings. Clean-client inference proved manifest and run response values of 0.35 and 0.50 with source backend_settings_env.

Release/tag strategy

Use immutable v0.1.2 after merge because tracked backend/deployment files changed after v0.1.1. Continue using the authorized v0.1.1 model release asset. No tag or release was created before merge.

Clean-user Docker validation

  • Clean clone branch checkout completed.
  • Local fallback model artifact installed with checksum verification.
  • PostGIS, Redis, backend API, Celery worker, and frontend became healthy on native linux/arm64.
  • MODEL_DEVICE=auto resolved to CPU with PyTorch 2.12.0+cpu.

Real pipeline validation

Real BANDON Docker inference completed. Final rerun: job job-e659a64fe6d748e0ac56347d8359aada, request hash 5ca577a5ad01eb19f8bc97ea, 14 artifacts, and PNG retrieval passed. Browser validation confirmed the deployed frontend and map render.

Tests

  • Focused backend: 26 passed.
  • Full backend: 474 passed, 5 skipped, 7 unrelated existing failures.
  • Frontend original worktree: 69 tests passed and build passed.
  • Clean clone frontend: 55 tests passed; build unavailable because clean client intentionally had no Node dependencies.
  • Root/deploy compose resolution, runtime validation, health checks, and real smoke inference passed.

Known limitations

  • GHCR publish of cpu-v0.1.2 is blocked because the available GitHub token lacks package write scope; E2E used the exact locally loaded arm64 image. The remote tag does not exist yet.
  • A repeat start.sh after backend environment changes can leave the existing frontend proxy pointing at the old backend container until frontend recreation.
  • CUDA remains optional and pending production validation; no CUDA readiness claim is made.
  • Phase 4G remains incomplete until cpu-v0.1.2 is published and a clean pull-only validation passes.

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.

1 participant