Starport makes one provider request for each executor attempt. The Vertex AI connector therefore uses one explicit Google Cloud location and does not retry through hidden backup regions.
Set the project and one location with the conventional names declared by the Starmap provider record:
export GOOGLE_CLOUD_PROJECT=your-project-id
export GOOGLE_CLOUD_LOCATION=us-central1Starport checks the ordered conventional names before the derived
STARPORT_GOOGLE_VERTEX_PROJECT and STARPORT_GOOGLE_VERTEX_LOCATION names.
It binds the selected values to the endpoint template from Starmap.
Configure Google Application Default Credentials for a renewable token. For example, use one of the standard Google credential mechanisms:
gcloud auth application-default loginDefault mode supports local Application Default Credentials, Google managed identity, and workload identity federation. Starport requests the Google Cloud platform scope. It preserves the quota project from the detected credentials and sends it with each request. It refreshes the bearer token two minutes before expiry.
The current Starmap Vertex AI inference profile uses the compiled
google-default authentication primitive. It does not declare a static token
profile or an AUTH_MODE field. If the catalog adds a new typed profile,
Starport can select it without a provider-specific branch when the primitive
is already compiled.
Starmap resolves its catalog-acquisition credentials independently. Starport does not copy a Starmap credential into an inference request.
The connector returns the first location failure to the central attempt executor. The executor applies one total attempt and elapsed-time budget across retries and route fallbacks. A streaming request can change routes only before Starport returns its first event.
To use regional redundancy, represent each region as an explicit route in a future catalog extension. Do not put a region loop inside the provider adapter. This keeps attempt counts, time budgets, and availability evidence complete.