Amundsen OIDC Coverage Starts With a Service-by-Service Inventory

Amundsen is a metadata discovery platform built from a frontend, a search service, and a metadata service. Its architecture documentation describes the frontend as the web portal, the search service as the query layer, and the metadata service as the API for reading and updating metadata. A security review that protects only the familiar frontend port misses the services behind it.

The Amundsen Helm values documentation shows that OIDC configuration is defined separately for the frontend, metadata, and search services. It also warns against placing client secrets directly in values and points to environment variables for those values. That design is useful because each service can receive its own client registration, but it creates a coverage problem: an operator must verify all three paths.

The metadata service documentation describes a REST API for metadata such as descriptions and tags. It can use Neo4j or Apache Atlas as a backing store. The documentation also notes that Amundsen does not yet support all of Atlas's fine-grained access controls. That does not mean every deployment lacks authorization. It means the integration's actual policy must be verified, not inferred from the backend's capabilities.

A ZoomEye measurement on 2026-11-21 at 05:41 CST used page 1, page size 1, and sub_type=all:

Query Index matches
app="Amundsen" 0
title="Amundsen" 427

The title result is broad and may include unrelated uses of the name. The zero app result means the selected app fingerprint was not observed in this probe. Neither count confirms an Amundsen portal, and the title query should be treated only as a candidate-discovery signal.

Begin with a port and service map. Identify the frontend, search, metadata, Neo4j, Elasticsearch, and any proxy or ingress path. Confirm which services are reachable from the user network, the application network, and the management network. The metadata API may be intended for internal service traffic, but an ingress rule can expose it inadvertently.

Review OIDC configuration per service. Record client ID, issuer, redirect URI, scopes, secret storage, and allowed endpoints. The Helm example includes a whitelisted endpoint setting for status and health checks. Verify that the whitelist contains only the intended paths and that health endpoints do not reveal metadata.

Test the frontend and the service API separately. Sign in through the frontend and confirm the expected page access. Then call the metadata API directly with no session and with a session that lacks the required role. Confirm the result matches the intended policy. A frontend redirect is not an API authorization boundary.

Review data editing authority. A user who can edit table descriptions, tags, or ownership metadata can affect data discovery and governance decisions. Separate read-only consumers from metadata editors and platform administrators where possible. Log changes to high-value tags or ownership fields.

The backing store has its own credentials. The metadata service may hold a Neo4j username and password or an Atlas proxy credential. Protect those values in a secret store, restrict who can change the service configuration, and rotate them with the same ownership discipline as the OIDC clients. A frontend login does not protect the service's database credential.

Use ZoomEye to find candidate portals and document the query fields and observation time. Follow up against owned assets with direct service checks and configuration evidence. Amundsen's OIDC story is strongest when all three services have explicit coverage and the metadata API has an independent authorization test.

References