php-fpm falls, reveal the cause

Hello guys (and girls)!

We have: 3 VPS with the same (synchronized) content and configured MariaDB Galera Cluster. Servers are located far from each other (Japan, Netherlands, USA).

Goal: uninterrupted operation and geo-distributed return of content.

Problem: At a random point in time, nginx freezes, and then gives out a 504 bad gateway, its logs show that the php-fpm socket is no longer available. Personally, I could not configure the output of php-fpm logs to understand the cause of the fall, so I ask you for help for payment.

Perhaps this is due to the long mysql slick due to update or insert, but the logs will show it better.
You need to find the cause of the crash and configure all 3 VPS (mariadb, php-fpm, nginx).