Access Control

Kaapana contains strong authorization features that allow to control access to the following resources:

  • Data

    • DICOM Data stored in PACS

    • Files stored in MinIO

    • Metadata stored in OpenSearch

  • Objects

    • Workflows

    • Jobs

    • Datasets

    • Projects

  • Services
    • APIs

      • Workflow Management System

      • Extensions API

      • Project API

    • Web Interfaces

      • Airflow

      • Kubernetes Dashboard

      • Keykloak

      • Traefik

      • Prometheus and Grafana

      • Extensions page

  • Access to active applications

  • Workflow execution

How Access Control Works

Every web request to Kaapana is subject to authorization checks that determine whether access to a resource is permitted. The authorization decision is based on:

  • The access token included with the request

  • Configurable policies that define the access rules

Authorization decisions are evaluated at the Policy Decision Point (PDP), which is an instance of an Open Policy Agent (OPA). Access policies are written using the Rego policy language.

Global System Groups

System-wide permissions in Kaapana are managed centrally through groups in Keycloak, the identity and access management system. For more information about the available system groups and detailed instructions on how to add users to groups, refer to the corresponding section in the User Guide.

Project Rights and Claims

Most resources in Kaapana are organized within the scope of a project, which serves as the central unit for managing access to these resources.

In addition to configurable access policies, Kaapana allows the definition of custom rights, providing fine-grained control over project-specific resources. These rights are reflected as claims in the user’s access token, enabling external services to enforce access restrictions based on project membership and assigned permissions.

Custom rights can be used to control access to services such as:

  • MinIO

  • OpenSearch

  • DICOMWeb API

This mechanism ensures consistent, project-aware access control.

Custom rights can be created in the confimap of the Access Information Interface (AII) before the platform is build.

Access Information Interface (AII)

The AII provides a REST API for managing:

  • Rights: Fine-grained permissions associated with specific actions or resources

  • Roles: Collections of rights that define permission levels

  • Projects: Bundle of resources and services.

  • User-Project-Role Mappings: Assignments that link users to specific roles within projects

During authentication, Keycloak queries the AII to determine a user’s rights. For each right granted to the user (based on their project-specific roles), Keycloak populates the user’s access token with the corresponding claim. User-Project-Role mappings can be managed via the Project Management Interface.

This mechanism enables precise, project-specific authorization. It allows you to define different permission levels for users on resources associated with projects, such as:

  • Read-only access to datasets

  • Permission to submit workflows for specific DAGs

  • Administrative control over project resources

Authorization flow

The following sequence diagram gives a simplified overview of the authorization flow for web requests.

sequenceDiagram participant C as Client participant R as reverse-proxy participant K as Keycloak participant P as PDP participant I as AII C->>R: Web request K->>I: Request rights I-->>K: Return rights K-->>R: Return access token R->>P: Forward request for authorization P-->>R: Return authorization decision alt Authorized Note right of R: If authorized, forward<br/> request to Kubernetes R-->>C: Relay response <br/> from Kubernetes service else Unauthorized R-->>C: Relay 403 end

Access Control Within Processing-Containers

Processing-containers are tightly coupled to a task-run in Airflow, and thus, to a DAG-run. Since every DAG-run belongs to a specific project, processing-containers always operate within a well-defined project context.

Processes running inside these containers often require access to storage services such as:

  • DICOMWeb

  • MinIO

  • OpenSearch

To ensure strict project-based isolation, processes must only be able to access storage resources belonging to the project associated with their DAG-run.

This is enforced through the following mechanisms:

  • Processing-containers are executed within dedicated Kubernetes namespaces, one per project.

  • For each project, a system user is created with permissions strictly limited to that project’s resources.

  • Processing-containers use the credentials of the corresponding project system user to authenticate against the storage services.

This design guarantees that processes inside containers can only interact with storage resources belonging to their project, preventing cross-project data access.

Workflow Execution and Active Applications

The Project Management UI provides fine-grained control over which DAGs can be executed as workflows within specific projects. This is achieved by managing project-software mappings, which define the association between a project and the workflows available to it. These mappings can be created or removed as needed to control workflow availability. Additionally, a default set of project-software mappings can be defined prior to building your custom Kaapana platform in the correpsonding configuration file

Similarly, active applications are always tied to a specific project context. When launching an application via the Extension Page, the currently selected project determines the project within which the application will be deployed. For applications started automatically by workflows, such as those initiated by MITK-Flow, the project association is inherited from the project of the corresponding workflow execution.

This mechanism ensures that both workflow execution and application deployment in Kaapana remain strictly project-bound, maintaining clear separation and secure access to project resources.

Note

Only users within dedicated global system groups are able to manage project-software-mappings and to start applications. Check out the Keycloak user guide for more information.

Client access to Kaapana APIs

You can access all APIs in Kaapana from an external client using the KaapanaApiService class, which is provided by the kaapana_client Python package. Authentication is handled automatically via the OAuth 2.0 Device Authorization Grant. This flow is designed for clients that cannot open a browser themselves — the user approves access once in a browser, after which the service holds a long-lived refresh token and renews access tokens silently.

Prerequisites

  • The kaapana_client package is installed (pip install kaapana_client).

Initialization

KaapanaApiService requires four constructor arguments:

Parameter

Description

root_url

Base URL of the Kaapana instance (e.g. the Traefik gateway URL). All endpoint paths are appended to this value.

project_id

UUID of the project you want to operate in.

client_id

OAuth2 client ID registered in Keycloak.

client_secret

OAuth2 client secret for the given client, or None for public clients.

When you create a KaapanaApiService instance, it immediately requests a device code from Keycloak and prints a verification URL to the log:

from kaapana_client.services.ApiService import KaapanaApiService

api = KaapanaApiService(
    root_url="https://<host>",
    project_id="d7e991b3-9463-48e7-98c2-661da8b83018",
    client_id="kaapana",
    client_secret=None,
)
# INFO - Open the following URL in a browser to grant the ApiService
#        access to Kaapana: https://<host>/auth/realms/kaapana/device?user_code=XXXX-YYYY

As a convenience, if the required values are available as environment variables, you can use the get_api_service_from_env factory function instead:

from kaapana_client.services.ApiService import get_api_service_from_env

api = get_api_service_from_env()

Open the printed URL in a browser and confirm access with a Kaapana account. You do not need to do this before calling a method — if the token has not been obtained yet, the first HTTP call will poll for approval automatically (up to 10 attempts, 5 seconds apart) and log the URL again on each retry.

Making requests

All five HTTP methods — get, post, put, delete, and head — accept an endpoint path relative to root_url, followed by any keyword arguments accepted by the underlying requests library (e.g. json, params, data, timeout). Authentication headers and the project cookie are injected automatically.

# GET  /aii/projects
response = api.get("aii/projects")
projects = response.json()

# POST /workflow-api/v1/workflow-runs  with a JSON body
response = api.post("workflow-api/v1/workflow-runs", json={"workflow": "..."})

# PUT  /aii/projects/<id>
response = api.put(f"aii/projects/{project_id}", json={"name": "my-project"})

# DELETE a resource
response = api.delete(f"aii/projects/{project_id}")

Token lifecycle

The service manages tokens transparently:

  • Access token absent — triggers the device-code polling loop described above.

  • Access token expired — a silent refresh-token grant is performed before the request is sent. No user interaction is required.