Dask Gateway

2026.3.0 verified Fri May 01 auth: no python

A client library for interacting with a Dask Gateway server, enabling scalable multi-tenant HPC and cloud deployments. Version 2026.3.0 supports Python ≥3.10. Released ~2x per year, stable API.

pip install dask-gateway

Common errors

error ModuleNotFoundError: No module named 'dask_gateway' ↓

cause dask-gateway not installed in the current Python environment.

fix

Run pip install dask-gateway.

error dask_gateway.exceptions.GatewayError: 401 Unauthorized - authentication failed ↓

cause Invalid or missing authentication credentials.

fix

Pass the correct auth parameter (e.g., auth='jupyterhub') or set DASK_GATEWAY_AUTH environment variable.

error TypeError: __init__() got an unexpected keyword argument 'proxy_address' ↓

cause In older versions (pre-2023.1.0), the `proxy_address` parameter was not available.

fix

Upgrade dask-gateway: pip install --upgrade dask-gateway.

Warnings

breaking In v2024.1.0, the default auth no longer falls back to 'jupyterhub'. You must explicitly pass `auth='jupyterhub'` or set environment variables. ↓

fix Pass `auth='jupyterhub'` when using JupyterHub, or configure `DASK_GATEWAY_AUTH` env var.

deprecated The `GatewayCluster.adapt()` method is deprecated. Use `auto_scale()` instead. ↓

fix Replace `cluster.adapt(...)` with `cluster.auto_scale(...)`.

gotcha When using Kerberos auth, the server must have the same Kerberos principal configured. Otherwise, authentication fails with a vague HTTP 401. ↓

fix Ensure Dask Gateway server has `kerberos.principal` set correctly in its config.

Install

pip install dask-gateway[kerberos]

pip install dask-gateway[jupyter]

Imports

GatewayCluster
wrong
```
from dask_gateway.client import GatewayCluster
```
correct
```
from dask_gateway import GatewayCluster
```
GatewayCluster is a top-level alias for the client class; direct import from .client is internal and may break.
Gateway
wrong
```
from dask_gateway.gateway import Gateway
```
correct
```
from dask_gateway import Gateway
```
Gateway is available at the top-level package; importing from .gateway is an internal path.

Quickstart

Creates a Dask Gateway cluster, scales to 2 workers, and gets a client.

import os
from dask_gateway import GatewayCluster

# Create a cluster with token auth (or env vars)
cluster = GatewayCluster(
    address=os.environ.get('DASK_GATEWAY_ADDRESS', 'http://gateway:8080'),
    proxy_address=os.environ.get('DASK_GATEWAY_PROXY_ADDRESS', 'gateway:8786'),
    auth='jupyterhub' if os.environ.get('JUPYTERHUB_USER') else None
)
cluster.scale(2)
client = cluster.get_client()
# Submit work
client.close()
cluster.close()