Fix hanging streams

[rkistner]

A race conditions with websocket handling could cause the route handler to "hang" indefinitely, after the underlying connection is closed. This surfaces in two ways:

1. The "concurrent connections" metric reports these as connections, giving a slow increase in connection count over time.
2. The syncSemaphore could permanently hold on to a lock for these, eventually leading to no locks being available, and no users able to sync.

While the race condition is rare, it does build up over time.

In tests I could trigger the race condition in two places:

1. Closing a connection before observer.registerListener() is called, by setting lifetime: 0 on the client. I'm not sure whether this ever happened in practice.
2. Close the connection after next batch of data is retrieved, but before the if (requestedN <= 0) check. This caused the handler to indefinitely wait for the next request() or cancel() to be triggered.

The fix here is to:

1. Use an AbortController instead of the observer to handle stream cancellation.
2. Check if (signal.aborted) to see whether the stream is already cancelled, wherever we listen for cancellation.
3. Throw an error when unable to acquire the syncSemaphore within 30 seconds, instead of just waiting silently.

[changeset-bot]

🦋 Changeset detected

Latest commit: 306b6d8

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 4 packages

Name	Type
@powersync/service-rsocket-router	Patch
@powersync/service-core	Patch
@powersync/service-image	Patch
test-client	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[stevensJourney]

Looks good to me 🙏 Thanks for the fix.