googleapis / googleapis/google-cloud-node
google-auth-library: support client-side Credential Access Boundaries (Java ClientSideCredentialAccessBoundaryFactory equivalent)
- Dominant language
- TypeScript
- Stars
- 3.2k
- Forks
- 712
- Avg merge
- 2d 3h
- Merged PRs (30d)
- 99
Description
### Feature request
Support client-side Credential Access Boundaries in `google-auth-library-nodejs` — the equivalent of Java's [`ClientSideCredentialAccessBoundaryFactory`](https://github.com/googleapis/google-auth-library-java/tree/main/cab-token-generator).
### Background
The Node library ships the server-side `DownscopedClient`, which performs one STS token exchange per downscoped token. Google's IAM documentation ([Create a downscoped credential](https://cloud.google.com/iam/docs/create-downscoped-short-lived-credentials)) now describes the client-side approach — exchange once for an access boundary intermediary token, then mint downscoped tokens locally with varying boundary rules, no STS round trip per token — and calls it "the recommended approach because it provides better performance, scalability, and future feature compatibility."
Today that implementation exists only in Java (`cab-token-generator` module, `ClientSideCredentialAccessBoundaryFactory`). Searching the googleapis org, there is no Node, Python, or Go implementation and no open tracking issue, so filing this one for Node.
### Use case
A token broker minting many unique prefix-scoped Cloud Storage tokens at high frequency: per-tenant object-name prefixes, one short-lived token per tenant session. With the server-side `DownscopedClient`, every unique boundary costs an STS exchange against the `sts.googleapis.com` token-exchange quota (6,000/project/minute). Caching per tenant helps, but fresh-mint demand scales with distinct active tenants; client-side minting removes the ceiling entirely, as the docs note.
### Questions
1. Is client-side CAB support planned for `google-auth-library-nodejs`? Any timeline?
2. If it is not planned soon: is the intermediary-token wire format (the protobuf-encoded boundary + encryption scheme visible in Java's `cab-token-generator`) considered stable and supported for third-party client implementations, or is it internal and subject to change?
### Environment
`google-auth-library` 10.x on Node 24 / Bun, tokens consumed by Cloud Storage JSON/XML APIs.
Contributor guide
Research direction
Start by reviewing the Java `cab-token-generator` module and `ClientSideCredentialAccessBoundaryFactory`, then compare them with the Node library's existing `DownscopedClient`. Determine the supported intermediary-token format and the Node API needed for local minting; done means client-side CAB support is implemented with the requested interoperability and usage behavior resolved.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- google-cloud, nodejs, typescript
- Domain
- authentication, backend, security
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Quiet
- Clarity
- Needs clarification
- Newbie friendliness
- 30/100