LEMP stack, weekly needs reset

I'm currently running a linode 2048 with ubuntu / nginx / php5-fpm / mariadb with varnish.

I keep running into the same issues over and over again with this setup.. Runs flawlessly for a couple weeks then all the sudden dies. I then have to restart nginx / php5-fpm / and varnish about 10 times to get it back into a stable state.

Looking at CPU I'm never above 40%, Memory never above 30%, Swap 0%, Load 4% and below, and network 1.3-2 mbs. (From Longview)

Here's my nginx error log (below lines repeated numerous times… And the below is sporadic):

open socket #11 left in connection 3

*20 recv() failed (104: Connection reset by peer) while reading response header from upstream, client: 127.0.0.1

I've tried using fastcgipass 127.0.0.1:9000; instead of fastcgipass unix:/var/run/php5-fpm.sock; but that doesn't seem to solve the problem either.

This is driving me nuts, thanks for any insights!