dist-gcs-pdf-processing 0.1.0

Distributed, scalable GCS PDF processing pipeline with Gemini OCR, Redis, and API endpoints.

DCPR PDF Processing Pipeline

Overview

This service processes PDFs from a Google Cloud Storage (GCS) bucket, performs OCR using the Gemini API, and uploads clean, text-based PDFs to a destination GCS folder. It features robust error handling, logging, monitoring, and is designed for scalable, production use.

---

Setup & Environment

1. Clone the repository and navigate to the project root:

   git clone <repo-url>
   cd nest-starters
   

2. Install dependencies:

   pip install -r requirements.txt
   pip install -e .
   

3. Environment Variables:

   • Place your .env file in the secrets/ directory at the project root.
   • Example .env variables:

     GCS_BUCKET=your-bucket-name
     GCS_SOURCE_PREFIX=source-folder
     GCS_DEST_PREFIX=dest-folder
     GEMINI_API_KEY=your-gemini-api-key
     SUPABASE_URL=https://your-supabase-url
     SUPABASE_API_KEY=your-supabase-api-key
     MAX_RETRIES=3
     GEMINI_GLOBAL_CONCURRENCY=10
     MAX_CONCURRENT_FILES=3
     PAGE_MAX_WORKERS=5
     DOC_BATCH_SIZE=10
     MAX_QUEUE=100
     POLL_INTERVAL=30
     G_MESSAGES_DEBUG=none
     G_DEBUG=fatal-warnings
     

---

Running the Worker & API

You can run the background worker or the FastAPI API server:

Run the Worker (background processing only)

python src/worker.py

Run the FastAPI API (with all endpoints)

uvicorn src.main:app --reload

• The API exposes health, status, logs, metrics, config, and file processing endpoints.
• The worker will start automatically in the background when the API starts.

---

Logging & Monitoring

• Logs:
  • Human-readable logs: /logs/worker.log (daily rotation)
  • JSON logs: /logs/json/YYYY-MM-DD.json
  • Dead letter logs: /logs/dead_letter/dead_letter.log
• Supabase:
  • Persistent errors are logged to the Activity_Error_Log table for monitoring.
• Suppressing GTK/GLib output:
  • Set in .env and at the top of main.py and worker.py.

---

Error Handling

• Retries for transient errors (network, quota, etc.) with configurable limits.
• Per-page retries: Each page is retried up to MAX_RETRIES times before being skipped.
• Per-file retries: If a file fails (e.g., page count mismatch), the whole file is retried up to MAX_RETRIES times.
• All persistent errors are logged to file, JSON, dead letter, and Supabase.

---

Scalability, Concurrency & Throttling

• Rolling Concurrency Model:
  • The worker always keeps up to MAX_CONCURRENT_FILES files in progress.
  • As soon as a file finishes, the next available file is picked up, until all are processed.
  • This ensures maximum throughput and efficient resource usage.
• Per-Page Concurrency:
  • Each file's pages are OCRed in parallel, up to PAGE_MAX_WORKERS at a time.
• Global Gemini API Throttling:
  • All Gemini API requests (across all files and pages) are globally throttled by GEMINI_GLOBAL_CONCURRENCY.
  • This ensures you never exceed your API quota or rate limits.
• Backpressure:
  • If too many files are queued (MAX_QUEUE), the worker will pause and log a warning.
• Horizontal scaling:
  • Run multiple stateless worker instances on different machines/VMs for even more throughput.

---

Temp/Log Cleanup

• Files in logs, logs/json, logs/dead_letter, staging, and processed older than 200 days are deleted before the worker starts.

---

Tests

• Unit and integration tests are located in /tests.
• Tests cover:
  • PDF splitting/merging
  • Per-page and per-file retry logic
  • File-level rolling concurrency (ensuring the concurrency window is always full)
  • Global Gemini API throttling
  • Trace ID propagation in logs
• To run tests:

  pytest
  

---

CI/CD

• GitHub Actions workflow runs linting and tests on every push.
• Example workflow file: .github/workflows/ci.yml.

---

Additional Notes

• All print/log statements are also written to log files.
• Trace/request IDs are used for end-to-end traceability.
• For any persistent errors, check Supabase and the dead letter log for details.

Project Structure

project-root/
├── src/                # All main code (import as src.module)
├── tests/              # All tests (import as from src.module import ...)
├── logs/               # Log output
├── secrets/            # Secrets and credentials
│   └── your-service-account.json
│   └── .env
├── requirements.txt    # Python dependencies
├── setup.py            # For pip install -e .
├── Dockerfile
├── README.md

Local Development

1. Install dependencies:

   pip install -r requirements.txt
   pip install -e .
   

2. Run the worker:

   python -m src.worker
   # or, if you have an entrypoint script, use that
   

3. Run tests:

   pytest --import-mode=importlib tests/
   # or, if you have trouble with imports:
   PYTHONPATH=. pytest tests/
   

Docker Usage

1. Build the Docker image:

   docker build -t nest-starters .
   

2. Run the container (worker only):

   docker run --rm -it -v $PWD/logs:/app/logs nest-starters python src/worker.py
   

3. Run the container (API server):

   docker run --rm -it -v $PWD/logs:/app/logs -p 8000:8000 nest-starters uvicorn src.main:app --host 0.0.0.0 --port 8000
   

• The Dockerfile can be overridden to run either the worker or the API server.
• The .dockerignore file ensures your build context is clean and fast.
• You can override the CMD to run tests or other scripts as needed:

  docker run --rm -it nest-starters python -m pytest --import-mode=importlib tests/
  

Continuous Integration (CI)

• Use GitHub Actions or similar CI to run:

  - name: Install deps
    run: |
      pip install -r requirements.txt
      pip install -e .
  - name: Run tests
    run: pytest --import-mode=importlib tests/
  

Secrets and Environment Variables

• Place your GCP credentials JSON file in a secrets/ directory at the project root (not tracked by git).
• In your .env file (in the secrets/ directory), set:

  GOOGLE_APPLICATION_CREDENTIALS=secrets/your-service-account.json
  

• The worker will automatically load .env from secrets/.
• For Docker/CI, mount the secrets/ directory and ensure the .env file and credentials are present.
• Never commit secrets or credentials to version control!

---

Installation

You can install the package from a GitHub Release:

pip install https://github.com/youruser/dist-gcs-pdf-processing/releases/download/v0.1.0/dist_gcs_pdf_processing-0.1.0-py3-none-any.whl

---

CLI Usage

After installation, you can run:

dist-gcs-worker  # Start the background worker

dist-gcs-api     # Start the FastAPI API server (with all endpoints)

Or, for advanced usage:

python -m dist_gcs_pdf_processing.worker
python -m dist_gcs_pdf_processing.main