Possible race condition with shutdown() not actually shutting down

[kiplingw]

Unless I am mis-reading the code (entirely possible), it appears as though Pistache::Http::Endpoint's shutdown() method can return before all service threads have completed. This method in turn calls SyncImpl::shutdown(). The latter is defined like so:

        void shutdown() override
        {
            shutdown_.store(true);
            shutdownFd.notify();
        }

If this happens, that may create a race condition with some endpoint handlers still executing after the user thinks the endpoint has "shutdown".

I don't know enough about the reactor pattern @oktal employed, but my guess is the behaviour of shutdown() should probably block until it actually has shut down.

[Fabio3rs]

I think it may happen, the problem I described in this comment seens to be related:
#59 (comment)

[kiplingw]

@oktal, you're likely the only one who can advise us on whether this behaviour was intentional or not.

[dennisjenkins75]

Can someone create a PR that adds a unit test which reliably reproduces the problem? The PR might not get merged (certainly won't until the bug is fixed), but a reliable repro would be super handy. Its ok it the PR needs to expose/hack/modify the core of pistache in order to rig the failure, since the existing class hierarchy does not lend it self to C++ mocks.

[kiplingw]

Feedback welcome @dennisjenkins75.