[afonsobspinto]

CH-302

Problem

AsyncOperation.execute() waits for Argo to acknowledge workflow creation. A stalled Argo response can leave the calling API request waiting indefinitely and keep any database row lock held by the caller. A configurable timeout lets callers limit that wait through the Cloud Harness API.

Change

Add an optional request_timeout to argo_service.submit_workflow(). AsyncOperation.execute() and ContainerizedOperation.submit() forward submission options to that service.

workflow = operation.execute(request_timeout=(2, 10))

The Argo SDK accepts a timeout in seconds or a (connect, read) tuple for the HTTP create request. Cloud Harness closes the HTTP client on success or failure, propagates submission errors, and makes one submission attempt per call. Argo may accept a workflow before the response times out, so callers should check whether it was created before retrying.

Validation

• The original commits passed common-library, model, deployment CLI, and Argo component CI, plus lint and CodeQL.
• 54 SDK and workflow tests passed, covering legacy calls, numeric and tuple timeouts, the returned Workflow wrapper, one POST per call, and client cleanup.
• A regression test with a stalled local HTTP server verified that execute() times out after one CREATE attempt and allows the caller to reconcile the outcome.

Superseded by #899.