How to wait for outstanding queries to finish?

[Mr-Pepe]

First, a big thank you for this amazing library!

I am writing integration tests to test my app's interaction with the database.

On app startup, a bunch of listeners to the database get initialized asynchronously. The problem is that I have to close the database connection by calling db.close() after each test, or otherwise the tests try to establish multiple connections. Sometimes there seems to be a race condition between establishing the listeners and closing the database, leading to these error messages:

The following StateError was thrown running a test (but after the test had completed):
Bad state: Tried to send Request (id = 98): StatementMethod.deleteOrUpdate: UPDATE "categories" SET
"user_id" = ? WHERE "user_id" IS NULL; with [eb9f893d-b557-4bb2-9b07-817a2f487798] over isolate
channel, but the connection was closed!

Is there a way to somehow wait for queued queries to finish before closing the database?

I saw mentions of similar issues here but that comment did not link to a solution.

[simolus3]

Waiting for in-flight request before closing the database sounds reasonable, but that's not technically what's happening here. The error you've posted is thrown here, before submitting requests

drift/drift/lib/src/remote/communication.dart

Lines 127 to 130 in 4bd9fef

	if (isClosed) {
	throw StateError('Tried to send $msg over isolate channel, but the '
	'connection was closed!');
	}

For in-flight requests, you'd get a Channel was closed before receiving a response error.

This actually makes it kind of tricky to fix this issue in drift. You have one asynchronous task closing the database and other independent tasks not realizing that and continuing to emit statements. Drift already closes streams when the database closes, but if you emit multiple statements in response to query streams updating, it's possible that the database closes in-between.
Usually the way to fix that would be to have some two-way communication channel where

1. At the end of your test (but before closing the database), you request to the listeners that they stop.
2. Once the listeners are at a state where they're done with one iteration of whatever they're doing, they check the cancellation signal and complete a Completer.
3. That completer's future is awaited in tests before closing the database.

That's kind of tricky to set up, but I don't really see a solution that could live in drift. Judging from the error this isn't a weird intermediate half-closed state of the database internal to drift, you are closing the database and then have other tasks still using it.
But if I'm misunderstanding something or you have ideas on how to make this easier with some drift additions, I'd love to hear them.

[Mr-Pepe]

Thank you very much for the detailed response. It turned out that I had a listener listening to something other than the database and then trying to write those changes to the database. And I didn't cancel that listener correctly :\