Daemonising inside an event loop causes hangs

[sj26]

Unfortunately the latest eventmachine release has broken mailcatcher.

With eventmachine 1.0.4, if you install the latest mailcatcher with gem install mailcatcher -v "0.5.9" then start it with mailcatcher _0.5.9_ then nc localhost 1025 you won't get a welcome message, and the process won't respond to SIGTERM. (There's a workaround in 0.6.0.)

Mailcatcher starts an eventmachine loop and binds to a couple of ports before daemonising so we can display messages back to the user if anything fails. Previously this worked fine, and if you run without daemonisation it still works fine, but when daemonising the smtp server stops working. (Thin continues working fine, however).

Some bisecting revealed 10deb49 to be the guilty commit. Not sure exactly how it's interacting to cause SMTP to fail but not Thin.

I assume this is because daemonising will fork, twice, and that is tripping up the pid change handling.

[sodabrew]

Interesting that the fix is simply to restore the original reactor_running? behavior. I will take a look at this.
sj26/mailcatcher@757eafe

[sodabrew]

Oh - SMTP fails but not Thin. Are you using the EventMachine SMTP module?

[sj26]

Yeah, it's using the built in EventMachine::Protocols::SmtpServer in MailCatcher::Smtp. I figured it might be doing something inside using reactor_running? but wasn't sure how far down the rabbit hole it might be.

Thanks for looking into it!

[tmm1]

forking off while the reactor is running is not recommended. But if you must, you should use the EM.fork_reactor method. I would recommend copying the daemonization implementation from Thin.

[sj26]

Daemonization for thin use the daemons gem, which seems at core equivalent to Process.daemon, and doesn't seem to do anything special to the eventmachine reactor? I'm not sure what it would be doing specially. The old processes die immediately after forking so using the same fds should be fine?

[sodabrew]

I think this is pretty unconventional: https://github.com/sj26/mailcatcher/blob/5b3781b44cd3ad82c18fe38f7d15add4a5cf0ef8/lib/mail_catcher.rb#L155-190

Normally, I would do all of my setup (gather resources, open files, open ports) and then daemonize and then begin my run loop.

Forking in the middle of the event loop appears to be something that wasn't supposed to work, but by lucky bug it did allow you to do it. I'm not sure what the consequences of this are, e.g. was something secretly broken but it didn't materialize as a bug for you.

[sodabrew]

Ping - is still a problem for MailCatcher with latest eventmachine (looks like you're on 1.0.9.1 now)?