Laravel Forge php-fpm Optimization

Hi everyone,

I have recently redesigned one of my legacy CodeIgniter systems running on PHP 5 to run on Laravel 5.3 and PHP 7.1. At any given time, about 60-110 users are online at a single time. This is not a realtime app; it’s just a normal load-a-page-and-serve app with some AJAX requests here and there.

Here was my CodeIgniter + PHP 5 DigitalOcean droplet specs:

Web: 2 GB RAM + 2 CPU

Dedicated MySQL: 2 GB RAM + 2 CPU

My CodeIgniter + PHP 5 DigitalOcean Apache 2 configuration was:

<IfModule mpm_prefork_module>
        StartServers 5
        MinSpareServers 5
        MaxSpareServers 50
</IfModule>

Rarely do my load averages ever spike above 1 on either my web or dedicated DB server.

When I switched DNS to begin serving my new Laravel app with Laravel Forge, here were my server specs:

Web: 8 GB RAM + 4 CPU

Dedicated MySQL: 4 GB RAM + 2 CPU

php-fpm.conf for web was:

listen = /run/php/php7.1-fpm.sock
pm = dynamic
pm.max_children = 1024
pm.start_servers = 300
pm.min_spare_servers = 100
pm.max_spare_servers = 300
pm.max_requests = 1024

And php-fpm spectacularly failed and generated tons of "www pool seems busy (consider increasing pm.start_servers)" errors, alongside of a 100-400 load average (all three numbers) and near 100% CPU utilization constantly. It used only about only 50-70% RAM usage (about 5-6GB of 8GB on the new web machine) and no matter how much I increased these numbers, php-fpm still couldn’t handle it. The site was virtually not loading with our average user load of 60-110 and I had to redirect the traffic back to our old servers and app with CodeIgniter and abort the transition.

MySQL load stayed under 1 the entire time. It was the web server that was overwhelmed.

Can anybody explain to me why, with twice the amount of hardware as my current setup, nginx and php-fpm failed so badly? What am I missing here?