helm: logging improvements

[derekperkins]

This PR adds log rotation via logrotate to the helm chart. By default, it rotates logs at 100M, running on an hourly cron.

This also uses the same debian:stretch-slim to tail the error and slow logs, and adds a vitess user/group.

@sougou - Can you create a new Docker repo for logrotate and logtail and have them auto-build like the others?

[dkhenry]

LGTM

[derekperkins]

There are two things still bothering me about logging.

1. I don't like that the logrotate container is running as root
2. I don't think that the tailing containers (and maybe the logrotate container) are listening for SIGTERM, meaning that upgrades/restarts wait the entire termination deadline before executing.

[derekperkins]

Ok, I solved those two problems.

1. I changed the logrotate container to use a simple sleep loop instead of a cron and now it can run as the vitess user. I also trap for signals so logrotate doesn't block the pod from restarting.
2. I created another logtail container that does the tailing, but inside of a bash script that traps signals, and I can confirm that that solves the issue where the pod takes the full deadline before restarting. I have the logging containers start running mysqladmin ping in a loop after receiving a signal but before exiting. That makes sure that we don't miss any logs - especially error logs in the event that something goes wrong during shutdown.

This has been bugging me for a year and I feel pretty good about these changes. @dkhenry - if you could take another look, I made a lot of changes after your last LGTM.

cc @hmcgonig @leoxlin @acharis @msolters @trevex

[trevex]

I created another logtail container that does the tailing, but inside of a bash script that traps signals, and I can confirm that that solves the issue where the pod takes the full deadline before restarting. I have the logging containers start running mysqladmin ping in a loop after receiving a signal but before exiting. That makes sure that we don't miss any logs - especially error logs in the event that something goes wrong during shutdown.

This is a great quality of life improvement 👍