Frequent lockups in Nginx/php-fpm after move from Apache/mod_php

[Arthur Pemberton]

I moved hosting of roughly 150 Wordpress sites from Apache/mod_php to
Nginx/php-fpm after having tested and found measurably better performance
with Nginx. (hosted on Centos 6/7)

The first problem I ran into was that the server ran completely out of
memory (30GB of RAM) and came to a complete crash one night, I presume when
all the bots came by -- never hard this issue before. I changed from static
pm to ondemand, with 10 and 25 children per pool max, each website has it's
own pool. The new problem is that I start getting bad gateway errors just
as regularly, as if there's no queuing of new requests. I get frequent
lockups where it would seem that somehow every child in every pool is in
use, and there is no queuing of requests from Nginx to php-fpm, and this
state is maintained until I restart php-fpm

There are no errors in besides infrequent "unable to read what child say:
Bad file descriptor" which is pretty vague.

This is currently on Centos 7 64bit with Nginx 1.6.0 (from nginx's repos)
and php-fpm 5.5.15 (from the Remi repos)