Why is it not recommended to mix the Async API with the Sync API in the same application?

[mkmoisen]

The Concurrent Programming with asyncio documentation states:

You can choose whether to use the synchronous API or the asynchronous API in your code. It is recommended to not use both at the
same time in your application.

Would anyone be able to elaborate on this a bit?

Why is it not recommended? Is there specific cases where it would be bad but otherwise would be fine?

In my particular case, I have a sync application and want to use the sync API for the vast majority of the time. However, there are a few cases where inside of a sync thread, I need to make multiple independent queries to the database, and would like to be able to run these three concurrently in an event loop, instead of launching three threads.

Off the top of my head I can not think of any reason why it would be bad to use both sync and async APIs in the same code base.

Thank you.

[cjbj]

We didn't artificially prevent use of the two, but we don't test the scenario.

The general feeling was that async can be tricky for developers (you) to use the correct coding style eg. await appropriately. Having sync & async in the same code is a recipe for coding slips. @anthony-tuininga may have more tech comments when his week starts.

[anthony-tuininga]

It is generally important to ensure that only async-aware APIs are used with asyncio; otherwise, the event loop is blocked and won't allow other coroutines to be processed while the blocking API is executing (or waiting).

If you want to use both sync and async APIs concurrently, then you have more work cut out for you and additional complexities to navigate! You can create an event loop, process a bunch of coroutines and then shut down the event loop -- this is likely the simplest approach and works fairly well. There is some overhead involved in the setup and teardown of the event loop but I believe it is manageable. You can try to create the event loop in a separate thread and leave it there permanently -- but I'm not sure myself how feasible that is and how much overhead you would save if it was feasible! Other possibilities may exist, of course -- but they are not commonly used, which means you are generally on your own.

If you have converted most of your code to use asyncio, you can use the to_thread() function to force blocking code to run in a separate thread. As noted there, that only helps with IO-bound functions such as the existing sync code for python-oracledb -- but it can be useful if you have a section of code that hasn't been converted yet or simply cannot be converted to use asyncio.