🛡️ Dagster AuthKit

Community authentication wrapper for self-hosted Dagster OSS.

Authentication, RBAC, and Audit logs for Dagster without touching internal code.

🎯 What is this?

Dagster OSS has no auth. If you run it in a VPC or locally, anyone with the URL has full admin access.

AuthKit solves this by wrapping the dagster-webserver command to add:

• ✅ Login Interface: Simple username/password flow.
• ✅ RBAC (4 Levels): Granular control over who can do what.
• ✅ Audit Logs: JSON logs for monitoring who is doing what.
• ✅ Multi-Backend: Works with SQLite, Postgres, MySQL (via Peewee ORM) and Redis.

No code changes required. You don't touch your repository.py or dagster.yaml.

✨ What's New in v1.0.0

☸️ Helm Chart

• Production-ready Kubernetes chart under helm/dagster-authkit/, synced with the application's config (env vars, secrets, image tag) and versioned automatically by semantic-release. Requires an explicit image.tag.

🔄 Automated CHANGELOG

• CHANGELOG.md is now auto-generated by semantic-release on each release, alongside the version bump. Versions are inserted above the <!-- version list --> marker.
• RBAC deny-by-default — unknown GraphQL mutations require ADMIN by default.

🔐 Hardening & Fixes

• Rebuild of the fix backlog from v0.4.2 across CSRF (per-client double-submit), fail-closed GraphQL batches, token-gated /auth/metrics, Redis 6-compatible atomic rate limiting, LDAP connection cleanup, session revoked-token caps, and detection-layer Starlette Middleware.cls compatibility.
• Test coverage boosted 44% → 52% (+98 tests: detection layer, rate limiter, metrics gate, CookieBackend, CLI, LDAP, middleware).

⚠️ Upgrading from v0.3.x

1. Set DAGSTER_AUTH_SECRET_KEY in your environment. Generate one with:

   python -c 'import secrets; print(secrets.token_urlsafe(32))'
   

2. If using proxy mode, set DAGSTER_AUTH_PROXY_TRUSTED_IPS to your proxy's IP address.
3. Database migration happens automatically on first boot — no manual steps needed for SQLite/Postgres/MySQL. A session_version column is added to the users table.
4. Role serialization changed from string ("ADMIN") to int (40) in session cookies. Existing sessions continue to work (backward-compatible from_dict).

📂 Ready-to-Run Examples

We provide ready-to-use stacks for different scenarios in the examples/ directory:

examples
├── authelia              # NEW! Authelia + Caddy + LDAP SSO (Docker)
│   ├── Makefile
│   ├── docker-compose.yml
│   ├── Caddyfile
│   └── authelia/
├── kubernetes            # NEW! Minikube deployment
│   ├── Makefile
│   └── k8s/
├── ldap                  # Active Directory integration (**Experimental**)
│   ├── Makefile
│   ├── docker-compose.yml
│   └── ldap-bootstrap.ldif
├── postgresql_redis      # Recommended production setup
│   ├── Makefile
│   └── docker-compose.yml
└── quickstart-sqlite     # Simple local testing
    ├── Makefile
    └── docker-compose.yml

How to run

Pick a scenario, go into the folder, and check the Makefile.

1. Authelia SSO (Docker) Complete SSO with Authelia, Caddy, and OpenLDAP:

cd examples/authelia
make up
# Access: https://auth.company.com (admin/password123)
# Then:   https://dagster.company.com

2. Kubernetes (Minikube) Same stack running on Kubernetes:

cd examples/kubernetes
make build  # Build the Docker image inside Minikube
make up     # Deploy everything
# In another terminal: make connect (runs minikube tunnel)
# Add to /etc/hosts: $(minikube ip) auth.company.com dagster.company.com

3. Standard Setup (Postgres + Redis)

cd examples/postgresql_redis
make up

4. Local Quickstart (SQLite)

cd examples/quickstart-sqlite
make up

5. LDAP/AD Testing ⚠️ EXPERIMENTAL

cd examples/ldap
make up

🚀 Manual Installation (Python)

If you aren't using Docker, you can install via pip.

# For local testing (SQLite)
pip install dagster-authkit[sqlite]

# For server usage (Postgres + Redis recommended)
pip install dagster-authkit[postgresql,redis]

# For LDAP/Active Directory integration (**Experimental**)
pip install dagster-authkit[ldap]

Usage:

# Initialize the database and create the first admin
dagster-authkit init-db --with-admin

# Run Dagster (replaces the standard 'dagster-webserver' command)
dagster-authkit -f your_pipeline.py -h 0.0.0.0 -p 3000

# For proxy mode (Authelia/OAuth2 Proxy)
export DAGSTER_AUTH_BACKEND=proxy
export DAGSTER_AUTH_PROXY_LOGIN_URL=https://auth.yourcompany.com
dagster-authkit -f your_pipeline.py -h 0.0.0.0 -p 3000

☸️ Helm (Kubernetes)

Deploy on Kubernetes via the Helm chart in helm/dagster-authkit/:

helm upgrade --install dagster-authkit ./helm/dagster-authkit \
  --set image.tag="$(git describe --tags --abbrev=0)" \
  --set authkit.secretKey="$(python -c 'import secrets; print(secrets.token_urlsafe(32))')" \
  --set authkit.adminPassword="your-admin-password"

See values.yaml for all configuration options.

🔐 Roles (RBAC)

We provide 4 levels of access. Permissions are enforced via GraphQL query analysis.

How it works: AuthKit analyzes GraphQL queries using the official GraphQL parser to accurately identify mutations and block unauthorized actions.

📦 Backends

🤖 Service Token (API clients)

Non-interactive clients that poll /graphql (Prometheus exporters, CI scripts) cannot go through the login form. Set a static service token to let them authenticate with a header instead:

export DAGSTER_AUTH_SERVICE_TOKEN=$(python -c 'import secrets; print(secrets.token_urlsafe(32))')

curl -H "Dagster-Cloud-Api-Token: $DAGSTER_AUTH_SERVICE_TOKEN" \
     -H "Content-Type: application/json" \
     -d '{"query": "{ version }"}' https://dagster.yourcompany.com/graphql

The token goes through the same RBAC as any user, so a VIEWER token can run queries but every mutation is denied and audit-logged. It is a static, non-expiring credential: keep it at VIEWER unless the client really needs more. Session backends only (ignored in proxy mode); WebSocket subscriptions remain cookie-only.

🛠️ CLI Management

Manage users directly from the shell. Useful for CI/CD or admin tasks.

# Create a new launcher
dagster-authkit add-user bob --launcher

# Reset password
dagster-authkit change-password bob

# List everyone
dagster-authkit list-users

# View RBAC permissions matrix
dagster-authkit list-permissions

🔮 Roadmap

Current (v1.0.0)

• ✅ Username/password auth (bcrypt)
• ✅ 4-level RBAC (ADMIN/EDITOR/LAUNCHER/VIEWER)
• ✅ SQLite, PostgreSQL, MySQL, Redis support
• ✅ GraphQL mutation blocking with official AST parser
• ✅ LDAP backend (experimental)
• ✅ Proxy authentication (Authelia, Caddy, Traefik)
• ✅ Kubernetes example with full SSO stack
• ✅ Helm chart for Kubernetes deployments
• ✅ Redis session revocation and rate limiting
• ✅ Centralized UI templates
• ✅ CSRF protection
• ✅ Cross-pod session revocation (DB-backed session_version)
• ✅ WebSocket authentication (GraphQL subscriptions)
• ✅ Dual rate-limiting (username + IP)
• ✅ Proxy trusted IP allowlist
• ✅ Token-protected /auth/metrics
• ✅ Service token for non-interactive API clients
• ✅ Automated CHANGELOG.md generation

Next

• 🔄 OIDC backend (beyond proxy mode)

What we will NOT do:

• ❌ Inject React code into Dagster UI (too brittle)
• ❌ Complex enterprise features (that's what Dagster+ is for)

🤝 Contributing

Found a bug? Want to add a feature? Open a PR. If it works and keeps things simple, we'll merge it.

Especially needed:

• People with Active Directory experience to validate the LDAP backend
• Testing on different Dagster versions
• Helm chart contributions

📄 License

Apache 2.0 - see LICENSE

🙏 Credits

Built by Demetrius Albuquerque because self-hosting Dagster shouldn't mean no auth.

Inspired by the community's need for a middle ground between "no auth" and "pay for Dagster+".