Improve Horizon Support

[jaydrogers]

Background

I've been working with @nckrtl on optimizing Horizon support

Related Discord Discussion

https://discord.com/channels/910287105714954251/1235899514867683430

### Research Solution
- [x] Ensure Horizon fix is stable
- [x] Test entrypoint script
- [ ] Implement entrypoint script into serversideup/php
- [ ] Change order to be `start-first` for Swarm

### Documentation
- [ ] Show example of Laravel Horizon
- [ ] Show differences between Laravel Horizon and Laravel Queues ([example discussion](https://github.com/serversideup/docker-php/discussions/366#discussioncomment-9475434))
- [ ] Create video how Horizon works
- [ ] Post an update on https://github.com/laravel/horizon/issues/663

[jaydrogers]

@nckrtl: I just added some ToDos for myself above. Any other ideas that need to be added to this list, let me know.

Next week I will be partially out of the office, but I will start diving into this when I return the week after 👍

[nckrtl]

Looks complete to me. Still need to verify assumptions as discussed on Discord. Will give it this weekend a try.

[nckrtl]

Alright, I did a test and can confirm everything is working as expected with setting the horizon docker service order to start-first in docker-compose, but there are a few more requirements which need to be set properly in the Laravel project:

• In the horizon config file horizon.php you must set the timeout key of your queue config to an higher amount than your job may run. So if your job may run 5 minutes (300 seconds) the timeout should be at least set to 301

• In queue.php the retry_after key on your redis connection should be set higher than the timeout amount you set in your horizon config. Else your job will get retried too soon and it will get interrupted before it finished. So setting it to 302 should be sufficient but i'd recommend adding some margin (same goes for the timeout btw.

• In docker-compose the stop_grace_period should be set to an high enough amount, with 5 minute jobs 300 seconds should be sufficient but it wouldn't hurt to set this to 1 hour.

Because of the start-first a new container will start upon a new deployment while the old one will stay active and finish until horizon terminates gracefully. Made a video to demo it all. Might take a bit before the quality is bumped enough by YT for it to be watchable.

https://www.youtube.com/watch?v=NJ0Vj9nwgf4