We deployed a change to simplify the HTTP client that we used. Due to a difference in configuration in production vs. development and test, we did not catch that this would break in production.

We've now updated the tests to run the code in a more production-like way to avoid this problem in future.