Nginx – AWS EC2, Ubuntu: upstream timed out (110: Connection timed out) while reading response header from upstream

Use unix sockets, instead of TCP/IP. So, you'd do something like:

listen = /tmp/php.sock

If you don't want to do something like that, check your firewall rules. If you're using any sort of conntrack rules, you could easily be filling up the table for that. See if there's any log messages in dmesg about it. You shouldn't really be using conntrack on a high-load server, so disabling that is probably a good idea.

Also, check your ulimit settings. If the number of open files is set too low, you could run into issues.

Solution for the former issue : increase proxy_read_timeout or check why your backend doens't answer in time.

Your second question : There are two things, the Upgrade header and the Connection header.

Both of these need to be passed to the backend when using websockets and you can use a map to change the value depending on original request headers (i.e. if it's a websocket connection handhaske or standard HTTP traffic). The backend must reply to the Upgrade header with HTTP 101. Then nginx will enter in a specific case where a tunnel is set up between the origin of the request and the backend.

Such a case is explained in the official documentation :

http {

    map $http_upgrade $connection_upgrade {
        default     upgrade;
        ''          close;
    }

}

server {

   ...

   location /foo/ {
        proxy_pass http://backend;
        proxy_http_version 1.1;
        proxy_set_header Upgrade $http_upgrade;
        proxy_set_header Connection $connection_upgrade;
    }

}