Handle trigger calls to get_connection

[dstandish]

Some [incorrectly-written] triggers may call BaseHook.get_connection without wrapping with sync_to_async.

This is naughty behavior because it will block the event loop. But in 2.x it would not cause an error.

In 3.0, however, this results in an error. It fails in 3.0 because the code to hit the API wraps the get_connection call with async_to_sync, which is forbidden in the asyncio event loop.

What we do here is try to detect when this happens and when it does, we run it through greenback portal which makes it not fail.

Related: #55568

[ashb]

The reason is, the code to hit the API wraps the get_connection call with async_to_sync, which is forbidden in the asyncio event loop.

Yes but not quite.

The issue is itit wasn't first wrapped in sync_to_async (which is what the async_to_sync is meant to "bubble out" of

Most triggers are fine and so correctly call it in the right way. The ones they don't were always at risk of blocking the event loop (but in practice works have not blocked it does more than a few ms if they did. Most of the time)

[dstandish]

@ashb re

The reason is, the code to hit the API wraps the get_connection call with async_to_sync, which is forbidden in the asyncio event loop.

Yes but not quite.

The issue is itit wasn't first wrapped in sync_to_async (which is what the async_to_sync is meant to "bubble out" of

Most triggers are fine and so correctly call it in the right way. The ones they don't were always at risk of blocking the event loop (but in practice works have not blocked it does more than a few ms if they did. Most of the time)

isn't that what i wrote just above?

sometimes get_connection (which goes to the database) might be called without wrapping in sync_to_async.

[ashb]

Yes, I guess it is. Forgive me it's late 😴