Always install default signal handler for SIGINT.

[rayburgemeestre]

Reason for this fix is that in my case I created a systemd unit file for CentOS7.

Issuing systemctl stop digits makes it send SIGTERM to the Python process, but there is no handler for this. In packaging/deb/templates/digits.service I see there is a "hack" for Debian (make systemd use SIGINT instead of SIGTERM), but that doesn't seem to work on CentOS7. So what I did is create a bash script that traps SIGTERM from systemd and translates it into a SIGINT for Python.

Anyway in this case, the way Python is launched (as a subshell) it won't register the default handler for SIGINT. This is a small fix that simply always registers the SIGINT handler.. More info: http://stackoverflow.com/questions/40775054/capturing-sigint-using-keyboardinterrupt-exception-works-in-terminal-not-in-scr/40785230#40785230

I have another suggestion, maybe it's a better idea to additionally listen for SIGTERM?
If that is preferred I can update this PR with that solution?

[lukeyeager]

Interesting link. I'll have to think through the implications.

If the scheduler is running, SIGTERM will shut it down. Have you seen this line?
https://github.com/NVIDIA/DIGITS/blob/v5.0.0/digits/scheduler.py#L358

[rayburgemeestre]

Actually I didn't see that! So scratch my suggestion to additionally listen to SIGTERM :)

It's nice that SIGINT can be used for the server and SIGTERM for the jobs.

Edit: was a bit too hasty with writing a reply