Dont start http server in Scheduler.__init__

[fjetter]

This refactors the scheduler init such that the HTTP server is only started once the server is actually up. Otherwise every init call starts an http server and binds sockets, etc.

I tried testing for leaking sockets but that's not easy. Depending on OS this requires root and even with root, I had trouble asserting this properly using psutil.

Closes #4806

[fjetter]

This needs much more work. Ran the wrong tests locally :/

[mrocklin]

I was curious what was going to happen there. I remember having to put the HTTPServer in init for some reason. I was happy to see your initial PR, but not surprised to see it closed ;(

…

On Wed, Jun 9, 2021 at 10:36 AM Florian Jetter ***@***.***> wrote: This needs much more work. Ran the wrong tests locally :/ — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#4896 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AACKZTCKGQGTOKUB53MC7BLTR6DBPANCNFSM46MECLBA> .

[fjetter]

One reason why it is in init is that we can then align security and ensure that if either scheduler comm or http server is secured, the other one must be as well. At least this is one not-trivial thing I found locally. Why the CI deadlocks immediately is beyond my understanding atm