fix: robust pdfjs-dist worker resolution + Docker deployment

[Chocksy]

Summary

• Fix pdfjs-dist worker resolution using createRequire() instead of relative imports that break in bundled/Docker environments. This replaces the less robust fix in 3356286.
• Add Docker deployment support: web app Dockerfile (multi-stage, standalone output), improved ingestion-worker Dockerfile, .dockerignore, and docker-compose.prod.yml for self-hosted deployment.

Changes

• services/ingestion-worker/src/lib/pdf.ts — Use createRequire to resolve worker path reliably + explicit pdfjs options
• apps/web/Dockerfile — Multi-stage Node 20 Alpine build with standalone output
• services/ingestion-worker/Dockerfile — Improved build
• .dockerignore — Standard exclusions
• docker-compose.prod.yml — Production compose with web, worker, postgres, redis

Test plan

• Verify PDF parsing works in Docker container (both digital and scanned PDFs)
• Verify docker compose -f docker-compose.prod.yml up starts all services
• Verify web app builds and serves correctly from standalone output

[vercel]

@Chocksy is attempting to deploy a commit to the Zach's Projects Team on Vercel.

A member of the Team first needs to authorize it.

[greptile-apps]

Greptile Summary

This PR fixes PDF.js worker resolution in Docker/bundled environments using createRequire and adds a full Docker deployment stack (multi-stage web Dockerfile, improved worker Dockerfile, .dockerignore, docker-compose.prod.yml).

• The web Dockerfile depends on Next.js standalone output (copies .next/standalone, runs server.js), but apps/web/next.config.ts does not set output: 'standalone' — the standalone directory will never be generated, breaking the container on startup.
• web and worker in docker-compose.prod.yml depend only on postgres:service_healthy, not on the one-shot migrate container completing, creating a race condition where the app starts against an empty schema.

Confidence Score: 4/5

Not safe to merge as-is: the web container will fail to start due to missing standalone config, and a race condition exists between migrations and app startup.

Two P1 defects block the primary deployment path — the standalone output misconfiguration makes the web Dockerfile non-functional, and the missing migrate dependency causes unpredictable startup failures.

apps/web/next.config.ts (needs output: standalone) and docker-compose.prod.yml (web/worker depends_on migrate)

Important Files Changed

Filename	Overview
apps/web/Dockerfile	Multi-stage standalone build — broken because next.config.ts lacks output: 'standalone', making the standalone directory and server.js unavailable at runtime
docker-compose.prod.yml	Production compose with postgres, web, worker, and one-shot migrate — web and worker miss a dependency on migrate completing, risking startup against an empty schema
services/ingestion-worker/Dockerfile	Improved multi-stage build; copies pdf.worker.mjs into dist/ but that file is never referenced at runtime (dead copy)
services/ingestion-worker/src/lib/pdf.ts	Robust worker path resolution via createRequire; additional pdfjs options (useWorkerFetch, isEvalSupported, useSystemFonts) are reasonable for a Node.js environment
.dockerignore	Standard exclusions including node_modules, .env*, .next, .git — no issues

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[docker compose up] --> B[postgres]
    B -->|service_healthy| C[migrate]
    B -->|service_healthy| D[web]
    B -->|service_healthy| E[worker]
    C -->|one-shot: runs db:migrate| F[migrations complete]
    D -->|starts immediately after postgres| G[web app running]
    E -->|starts immediately after postgres| H[worker running]
    F -.->|race condition: may finish AFTER web/worker| G
    F -.->|race condition: may finish AFTER web/worker| H
    style F fill:#f9f,stroke:#f00
    style G fill:#faa,stroke:#f00
    style H fill:#faa,stroke:#f00

Loading

_{Reviews (1): Last reviewed commit: "fix: robust pdfjs-dist worker resolution..." | Re-trigger Greptile}

[greptile-apps]

output: 'standalone' not set in next.config.ts

The Dockerfile copies from .next/standalone and runs node apps/web/server.js, but apps/web/next.config.ts does not include output: 'standalone'. Without this option Next.js will not generate the standalone directory or the server.js entrypoint — the COPY --from=base /app/apps/web/.next/standalone ./ step copies nothing, and the container will crash on startup.

Add to apps/web/next.config.ts:

const nextConfig: NextConfig = {
  output: 'standalone',
  transpilePackages: [
    '@openvitals/common',
    // ...
  ],
};

[greptile-apps]

web and worker can start before migrations complete

Both services depend only on postgres:service_healthy, not on the one-shot migrate container finishing. If either starts while migrations are still running the application will hit an empty schema and crash. Add migrate: condition: service_completed_successfully to both depends_on blocks.

Suggested change

	depends_on:
	postgres:
	condition: service_healthy
	depends_on:
	postgres:
	condition: service_healthy
	migrate:
	condition: service_completed_successfully

[greptile-apps]

Copied worker file is never used

The cp $(find ...) step copies pdf.worker.mjs into dist/, but the production stage does not include that file, and pdf.ts resolves the worker path at runtime via createRequire pointing to node_modules/pdfjs-dist/ — not dist/. This copy is a no-op in the final image and can be removed.