RFR: 8294916: Cancelling a request must eventually cause its response body subscriber to be unregistered

[Daniel Fuchs]

On Tue, 11 Oct 2022 15:49:14 GMT, Daniel Fuchs <dfuchs at openjdk.org> wrote:

> When [JDK-8277969](https://bugs.openjdk.org/browse/JDK-8277969) was implemented, a list of outstanding response subscribers was added to `HttpClientImpl`. A body subscriber is added to the list after being created and is removed from the list when it is completed, either successfully or exceptionally.
> 
> It appears that in the case where the subscription is cancelled before the subscriber is completed, the subscriber might remain registered in the list forever, or at least until the HttpClient gets garbage collected. This can be easily reproduced using streaming subscribers, such as BodySubscriber::ofInputStream. In the case where the input stream is closed without having read all the bytes, Subscription::cancel will be called. Whether the subscriber gets unregistered or not at that point becomes racy.
> 
> Indeed, the reactive stream specification doesn't guarantee whether onComplete or onError will be called or not after a subscriber cancels its subscription. Any cleanup that would have been performed by onComplete/onError might therefore need to be performed when the subscription is cancelled too.

That's exactly what this change is going to fix. What happens in the code you're showing here is that if the status code is not 200, then the input stream is closed without having read any bites. This will cause the underlying subscriber to cancel its subscription. And that is where the leak comes from, as there's no guarantee that the subscriber's onComplete/onError will ever be called in that case. With this fix, the subscriber will be taken off the list when the subscription is cancelled, and that should fix the leak.

-------------

PR: https://git.openjdk.org/jdk/pull/10659