Fluke API

Fluke API is the standalone HTTP service for Fluke's public whale catalog, sightings, predictions, photo workflows, and administrative tools. It owns the runtime API contracts and PostgreSQL schema used by the web and iOS clients.

It is one repository in the Fluke project — a non-commercial, single-author field guide for the orcas of the Pacific Northwest. This service is the data spine the web and iOS clients read from.

Architecture

• Fastify serves versioned routes under /api/v1.
• Prisma owns the PostgreSQL schema and migrations in prisma/.
• Zod schemas in src/contracts/ validate data at service boundaries.
• Generated JSON Schema and fixture artifacts in contracts/ give clients deterministic compatibility inputs.
• Photo storage uses the local filesystem by default. The R2 configuration is reserved for the storage adapter and requires all R2_* values.

This repository preserves the API and shared-contract history extracted from the original calelamb/fluke monorepo. New API behavior, migrations, and canonical contracts belong here.

Requirements

• Node.js 22.17.0
• pnpm 10.33.0
• PostgreSQL compatible with the committed Prisma migrations

Corepack can install the pinned package manager:

corepack enable
corepack prepare pnpm@10.33.0 --activate

Local development

Copy .env.example to .env, replace its placeholders, then install and prepare the client:

pnpm install --frozen-lockfile
pnpm db:generate
pnpm db:migrate:deploy
pnpm dev

The development server listens on http://localhost:4000 by default. pnpm db:migrate:deploy applies committed migrations only; create and review new migrations in a development database before committing their SQL.

Useful commands:

Command	Purpose
pnpm layout:check	Reject legacy monorepo paths and verify the standalone layout.
pnpm contracts:generate	Regenerate deterministic schemas and fixtures from API-owned contracts.
pnpm contracts:check	Fail when generated contract artifacts have drifted.
pnpm typecheck	Type-check without emitting JavaScript.
pnpm lint	Run ESLint across runtime code, scripts, and the seed.
pnpm test	Run the Vitest unit and integration suite.
pnpm test:coverage	Run tests with the enforced 80% line threshold.
pnpm build	Compile production JavaScript to dist/.
pnpm audit	Reject high or critical production dependency advisories.

Environment

All environment input is validated at startup. The process exits with field-specific guidance when required configuration is missing or malformed.

Variable	Required	Description
DATABASE_URL	Yes	Pooled PostgreSQL connection used by the running API.
DIRECT_URL	Yes	Direct PostgreSQL connection used by Prisma migrations.
JWT_SECRET	Yes	Random secret of at least 32 characters. Generate it outside source control.
NODE_ENV	No	development, production, or test; defaults to development.
PORT	No	HTTP port; defaults to 4000.
ADMIN_COOKIE_NAME	No	Admin authentication cookie; defaults to fluke_admin.
WEB_ORIGIN	No	Comma-separated browser origins allowed by CORS.
API_PUBLIC_ORIGIN	No	Public API origin used to create absolute photo URLs.
IDENTIFIER_SERVICE_URL	No	Base URL for the separate identifier service.
STORAGE_BACKEND	No	local or s3; defaults to local. Production requires s3.
UPLOADS_DIR	No	Local upload directory; defaults to uploads in source. The container sets /app/uploads.
OBJECT_STORAGE_BUCKET	For S3	Private S3-compatible bucket name.
OBJECT_STORAGE_REGION	For S3	S3-compatible region.
OBJECT_STORAGE_ENDPOINT	For S3	Public HTTPS S3-compatible endpoint without a path, query, or credentials.
OBJECT_STORAGE_ACCESS_KEY_ID	For S3	Least-privilege access key ID.
OBJECT_STORAGE_SECRET_ACCESS_KEY	For S3	Least-privilege secret access key.
OBJECT_STORAGE_FORCE_PATH_STYLE	For S3	Explicit true or false path-style selection.

Never commit .env or production credentials. CI uses isolated, synthetic test-only values.

Health checks

• GET /api/v1/health is a liveness check. It returns 200 without touching the database, so an orchestrator only restarts a process that cannot serve HTTP.
• GET /api/v1/ready is a readiness check. It runs a bounded database probe and requires this image's schema migration with no unresolved migrations. A database migrated beyond the image remains ready, which preserves application rollback. Probe failures return 503 {"status":"unready"} without exposing database errors.

Use readiness for release health checks and traffic admission. Do not use liveness to decide whether a database migration succeeded.

Contract artifacts

The API is the source of truth for runtime contracts. After changing a contract:

pnpm contracts:generate
pnpm contracts:check
pnpm test

Commit the matching files in contracts/schemas/ and contracts/fixtures/. Client repositories consume those artifacts instead of importing API source.

Container

The multi-stage image pins Node and pnpm, prunes development dependencies, and runs as the unprivileged node user:

docker build --tag fluke-api:local .
docker run --rm --publish 4000:4000 \
  --env-file .env \
  --env NODE_ENV=production \
  --env UPLOADS_DIR=/app/uploads \
  --volume fluke-api-uploads:/app/uploads \
  fluke-api:local
curl --fail http://localhost:4000/api/v1/ready

The runtime image already sets NODE_ENV=production and UPLOADS_DIR=/app/uploads; the explicit flags above document the effective values and the volume's writable mount point. Its entrypoint runs prisma migrate deploy before the API process and exits without starting Node if migration deployment fails. CI proves this path against a blank PostgreSQL database before it accepts the image.

Production release

Release A currently runs the repository Docker image on a Render Free web service at https://fluke-api.onrender.com, backed by external Neon Postgres. The image applies committed migrations before starting the API; verify that startup and then require public 200 responses from /api/v1/health and /api/v1/ready, validate the catalog and CORS contract, and confirm Release B routes remain fail-closed. The checked-in railway*.json files remain an alternative paid topology, not the active host.

Scheduled ingestion and prediction run through .github/workflows/scheduled-jobs.yml using bounded standard GitHub-hosted jobs and repository secrets. See the operational documents below for the exact no-charge topology, probes, timeouts, and retry rules.

Operational procedures:

• Deployment
• Application rollback
• Database restore
• Scheduled jobs

CI release gates

GitHub Actions installs the pinned toolchain, starts PostgreSQL, applies migrations, checks standalone layout and contract drift, type-checks, lints, enforces coverage, builds, audits production dependencies, and installs Gitleaks 8.30.1. The action's event scan runs without PR comments under read-only permissions, followed by an explicit --all full-history scan. CI then builds the non-root container and verifies that its production entrypoint migrates a blank PostgreSQL database before readiness succeeds.

Run the locally available equivalents before pushing:

pnpm layout:check
pnpm contracts:check
pnpm typecheck
pnpm lint
pnpm test:coverage
pnpm build
pnpm audit
gitleaks git --log-opts=--all --redact --no-banner .
git diff --check
git fsck --full --strict

Contributions and use

The source is public so people can see how the service is built — the fail-closed configuration, the contract-artifact discipline, the bounded read routes. That transparency is the point.

It is not open to outside contributions. This is a single-author personal project; pull requests and feature issues aren't being accepted. Read it, learn from it — you're not expected or invited to contribute. No open-source license is attached, so all rights are reserved; the source is available to understand, not a grant of reuse.