I'm getting this error from /var/log/messages on my FreeBSD box. I'm using nginx and spawn-fcgi with memcache and apc modules enabled.
upstream prematurely closed connection while reading response header from upstream,
client HTTP/1.1", upstream: "fastcgi://unix:/tmp/fcgi.sock:", host:
I've had a similar error with unicorn + nginx.
The end result was that the unicorn was timing out due to a firewall misconfiguration, dieing off and leaving NGINX clueless as to what to do (nginx would then throw a 503).
Once the port was open my issue was resolved.
I've also seen this happen with an API call that takes a long time (longer than my 30s unicorn timeout). I ended up shipping it off to a background job so the unicorn didn't timeout.
I had a similar issue with Nginx timing out with a RoR app when using an EC2 + Amazon RDS database instance.
The issue was resolved by editing my security group for the RDS instance to allow the EC2's IP over port 5432. Just edit the security group's rules to add a custom rule for the port you are communicating to the RDS instance over, and whitelist the EC2 server's private IP address. Worked instantly after that!
It was related to the version of PHP. I have used latest version of nginx and slightly old version of PHP. The issue has been fixed by updating PHP to latest version.
