Add ocr-doc-api scaffold (Rust API + GCP Terraform/Terragrunt)

[pollenjp]

Summary

Scaffolds a document OCR REST API under ocr-doc-api-2026-05-07/.

• API (api/) — Rust + axum on Cloud Run. Endpoints:
  • POST /v1/jobs (multipart: file, engine, formats)
  • GET /v1/jobs/{id} (status JSON)
  • GET /v1/jobs/{id}/{format} (json|csv|html|md|pdf, streamed from GCS)
  • GET /healthz
  • On upload, the API stores the source in the input GCS bucket, writes an
    initial status.json to the output bucket, and publishes a JobMessage
    to Pub/Sub.
• Workers (workers/) — Cloud Run Job containers per OCR engine:
  • yomitoku/ shells out to the yomitoku CLI.
  • tesseract/ runs tesseract and renders json/csv/md/html(hOCR)/searchable-pdf.
  • The engine is selected per request, so adding a new backend = new image + Terragrunt env.
• Infra (infra/) — Terraform modules + Terragrunt envs (GCP):
  • Modules: storage, pubsub, artifact_registry, iam, api_service, worker_job.
  • live/dev/ wires storage → pubsub → IAM → AR → API/workers with explicit dependency blocks.

Architecture

client → Cloud Run (Rust API) → GCS input + Pub/Sub
                                           ↓ Eventarc
                                Cloud Run Job (yomitoku|tesseract)
                                           ↓
                                   GCS output (json/csv/html/md/pdf)
client ← Cloud Run (streams from GCS) ← GET /v1/jobs/{id}/{format}

Test plan

• cargo check / cargo build --release in api/
• terragrunt run-all init && plan in infra/live/dev/ against a real project
• Build and push api, worker-yomitoku, worker-tesseract images to Artifact Registry
• End-to-end: POST /v1/jobs with a sample PDF, poll status, GET each format
• Confirm Eventarc trigger destination — see infra/README.md "Notes / TODO" (Eventarc-to-Job needs a shim or the GA cloud_run_job destination)

https://claude.ai/code/session_01EioxbNVnmj9q3se6vUT1Z5

---

Generated by Claude Code