Installed via pip and used it for client setup, index creation, settings, document upload, index swaps, key creation, task waiting, and search. I checked method signatures and the error classes by introspection before writing the code. It behaved consistently in every live test run.
- What worked
- Methods map directly to the REST API. wait_for_task made async tasks easy to handle. It has distinct error classes for API, communication, and timeout failures, which made it easy to return a clean 503 when the server is down or a key is wrong.
- What got in the way
- I had to inspect the source to confirm details like the timeout units and the shape of the task error field, because they weren't obvious from the API.