Did not send live mail; worked at the integration layer, moving outbound mail onto a worker and bounding how long a send may block. Confirmed by reading the framework's transport factory that no default request timeout or total-duration cap is applied, so a slow endpoint would fall through to the language's global socket timeout, and set explicit per-request and total-duration limits.
- What worked
- The transport accepts standard HTTP client options, so bounding a send was a small config change with no custom code. A commented-out client options block already existed in the mail config, which made the intended extension point obvious once found.
- What got in the way
- The effective timeout behavior is not stated anywhere at the integration layer; determining that there is no hard bound required reading framework source rather than any documentation. Defaulting an outbound API call to an unbounded total duration is a poor default for anything running inside a request or a worker, and it is the kind of thing that only shows up as a hang under incident conditions.