hot-restarter.py: revert SIGCHLD handler to force kill instead of term

[mpuncel]

PR: #2596 changed the behavior of the SIGTERM and SIGCHLD handlers to
attempt to allow child processes to exit gracefully before force killing
them. This PR reverts the behavior of the SIGCHLD handler back to force
killing children if a child exits uncleanly. This should allow the
supervisor of the python process (e.g. runit) to restart envoy with a
shorter delay (whereas an attempt at graceful TERM might delay up to
TERM_WAIT_SECONDS).

Note: If the child process of hot-restarter.py is a container framework
(e.g. runc), the force kill might result in container state being
leaked. This should hopefully be a rare occurrence.

Signed-off-by: Michael Puncel mpuncel@squareup.com

title: Revert SIGCHLD handler in hot-restarter.py to force killing children instead of attempting graceful shutdown

Description:
Reverts the behavior of the hot-restarter.py
Risk Level: Low | Medium | High
Low

Testing:
I ran hot-restarter.py manually with /usr/bin/false and saw it print that it was sending a TERM
ran it with /usr/bin/true and saw that it exited cleanly without terminating children

Docs Changes:
no docs changes

Release Notes:
no release notes

[htuch]

@mpuncel seems reasonable as per @mattklein123 's ask. Thoughts on adding an integration sh_test with this?

[mpuncel]

sounds good to me, FYI it will probably take me a bit to get the developer environment set up

[mattklein123]

@htuch @danielhochman @mpuncel IMO this is somewhat of a regression. Can we merge this and then file a follow up issue for adding tests?

[mpuncel]

if you'd prefer, I can also revert my previous PR and submit a new one with tests once I get the chance

[mattklein123]

I will defer to @htuch @danielhochman. I don't really think the other PR needs to get reverted but I would rather merge this sooner rather than later.

[htuch]

Fair enough, tests welcome in a followup PR.