`asyncio.TimeoutError` raised on network timeout and CPU timeout

[Dave Hall]

A key difference between async-timeout and asyncio.wait_for is that wait_for performs a final test that the task is completed before raising a TimeoutError, whereas async-timeout immediately raises the error when the timeout expires.

This difference is really important when a system is at 100% CPU usage.

Let's say you're making a HTTP request with a timeout of 15 seconds. The request takes 14.5 seconds to complete, but then the system hangs for a full second due to CPU contention.

• async-timeout will report a timeout error, since the timeout will be triggered immediately after the program un-freezes, but before the task completes.
• asyncio.wait_for will not report a timeout error, since it will perform a final check to see if the network request completed before raising the error.

In other words, async-timeout reports timeouts due to CPU and network, whereas wait_for, only reports timeouts due to network.

The result being that, in a system that experiences a brief CPU spike, a great many network requests apear to "time out". The reality is that the python interpreter hung for a second, and the nework is fine.

The affects me when using aiohttp. I've had to wrap the library with a shim that disables all built-in timeouts and wraps network calls with asyncio.wait_for.

[Andrew Svetlov]

I don't think that your description is correct. The system can hang at any place, e.g. before finishing a request.
async-timeout doesn't spawn a task but adds a timer that cancels the current task. If it takes longer than delay -- TimeoutError is raised. That's it.