How to safely tell Rx IScheduler that it is time to exit

[mwpowellhtx]

In a WPF application, or really any application, when it is time to exit, how do we tell IScheduler to exit? Is that by disposing of any observers currently in flight?

[idg10]

In general, schedulers do not have any notion of shutting down.

There is one exception: virtual schedulers, or more specifically anything that derives from VirtualTimeSchedulerBase<TAbsolute, TRelative>, have a Stop method, although even that isn't telling the scheduler that it is time to exit. It is really asking the scheduler to let us know when it is ready to stop.

The main reason we have this on virtual schedulers is that those are designed for use in test scenarios where we need to run to the end of all activity. (Basically, we to know that everything has finished before checking the results) We definitely need that in tests, because the intent of the tests is to verify that the correct behaviour occurs eventually, however long that may take, but we also want the tests to complete as soon as they have reached that 'eventually'. Draining the scheduler work queue is an efficient and reliable way to achieve this.

Note that in this case, shutdown is not about enabling the scheduler to release any resources it has required. (So this is not like Dispose.) This is about ensuring that all of the work initiated by the test has run to completion. The scheduler will continue to run for as long as there is work to do, and Stop will return only once its work queues are empty. So although the method is called Stop, it might be more accurate to think of it as WaitForOutstandingWorkToComplete. The scheduler will stop when it decides it's time to stop, which might be never.)

I don't think we've previously had any request for a similar 'drain' all outstanding work outside of testing scenarios. I can see one obvious problem with such a thing: if you use anything that schedules recurring work, then unless you've shut down all active subscriptions (Disposed every subscription that was still running) then a Stop like we have on the virtual schedulers will never return. For tests that's OK because we just need to make sure we write the tests in such a way that things do come to a stop. But Rx applications don't necessarily do that. It's common to set up a timer and leave it running forever, with the expectation that the process will eventually exit without ever needing to shut that timer down.

Is there an application problem you're looking to solve with this? Or are you just looking to understand how to shut down Rx 'properly' on exit? If it's the latter, then you don't need to do anything.