Skip to content

Rename GCP CLI flags to provider-agnostic inference flags and auto-provision WIF #884

@waynesun09

Description

@waynesun09

Problem

The current --gcp-project, --gcp-region, and --gcp-wif-provider flags are only used for inference (Vertex AI), but their names suggest general GCP usage, which is confusing alongside --mint-project. Additionally, --gcp-wif-provider requires the user to manually look up and pass the full WIF provider resource name, and --gcp-region has no default despite global being the standard choice.

Proposed changes

1. Rename flags to provider-agnostic names

Current Proposed
--gcp-project --inference-project
--gcp-region --inference-region
--gcp-wif-provider --inference-wif-provider

2. Default --inference-region to global

Only --inference-project should be required to enable inference. Region defaults to global.

3. Auto-discover or create WIF provider

When --inference-wif-provider is not provided but --inference-project is:

  1. List WIF pools/providers in the project for one with an assertion.repository == '{org}/.fullsend' attribute condition
  2. If found, use that provider's resource name
  3. If not found, create the pool + provider (reuse logic from the mint provisioner)

--inference-wif-provider remains as an optional override for custom WIF setups.

4. Minimal inference setup

After these changes, enabling inference requires only:

fullsend admin install "$ORG" \
  --inference-project "$GCP_PROJECT" \
  --mint-project "$GCP_PROJECT"

Metadata

Metadata

Assignees

No one assigned

    Labels

    Type

    No type

    Projects

    Status

    Done

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions