Nginx and PHP-FPM Together: Sockets, Pools, fastcgi_params

How Nginx passes requests to PHP-FPM

Nginx cannot execute PHP itself — it serves static files and proxies dynamic requests to a separate PHP-FPM (FastCGI Process Manager) process over the FastCGI protocol. This is handled by the fastcgi_pass directive inside a location block that matches files with a .php extension. PHP-FPM runs the script and sends the result back through the same channel.

The link between Nginx and PHP-FPM can be built two ways: through a Unix socket or through a TCP address such as 127.0.0.1:9000. On a single server a socket is usually faster and does not occupy a port, while TCP is convenient when PHP-FPM runs in a separate container or on another machine. On ZevsHost.net shared hosting and VDS with cPanel and ISPmanager 6, a socket is used more often — the panel creates one automatically for every user.

Configuring the location for .php files

The basic block for a PHP site looks like this — a regex match, passing the socket, and a mandatory SCRIPT_FILENAME.

location ~ \.php$ {
    include fastcgi_params;
    fastcgi_pass unix:/run/php/php8.3-fpm.sock;
    fastcgi_index index.php;
    fastcgi_param SCRIPT_FILENAME $document_root$fastcgi_script_name;
}

Without SCRIPT_FILENAME, PHP-FPM has no idea which file to run and returns a blank page or a "No input file specified" error. The fastcgi_params file usually lives in /etc/nginx/ and is loaded with the include directive — there is no need to rewrite it entirely, just add the missing parameters on top.

Socket or TCP: how to choose without dropping connections

The difference between the two connection methods lies in speed and in limits on the number of simultaneous connections.

If Nginx logs show the error connect() to unix:/run/php/php8.3-fpm.sock failed (11: Resource temporarily unavailable), the socket queue is full: the PHP-FPM pool cannot process requests fast enough. The fix is to raise listen.backlog in the pool or increase the number of worker processes.

Tuning a PHP-FPM pool for load

A pool is described in a file such as /etc/php/8.3/fpm/pool.d/www.conf. Three parameters control how many requests PHP-FPM handles at once.

pm = dynamic
pm.max_children = 20
pm.start_servers = 4
pm.min_spare_servers = 2
pm.max_spare_servers = 8

pm.max_children is a hard cap on worker processes. It is calculated as: available memory minus memory for the system and Nginx, divided by the average size of one PHP process (usually 40-80 MB for a CMS like WordPress). Set it too high and the server swaps under peak load; set it too low and Nginx starts returning 502 on a traffic spike.

Diagnostics: where to look when the site will not open

The check order for a 502 or 504 error on a PHP-FPM site:

• Service status: systemctl status php8.3-fpm.
• The socket path in the pool config matches the path in fastcgi_pass.
• The PHP-FPM log in /var/log/php8.3-fpm.log for "max_children reached".
• The Nginx log in /var/log/nginx/error.log for "upstream timed out".

A detailed breakdown of specific 502 and 504 error codes with commands for each case is covered in a separate article on 502 and 504 diagnostics. General rules on which location wins for .php files are covered in the article on location priority.

Summary: checklist for the Nginx and PHP-FPM link

• The socket path or TCP address in fastcgi_pass matches listen in the PHP-FPM pool.
• The SCRIPT_FILENAME parameter is set explicitly, not inherited from an old config template.
• pm.max_children is calculated from the server's RAM, not left at the default.
• After editing configs, nginx -t and systemctl reload php8.3-fpm were both run.
• The link matches the PHP version actually installed on the server — check with the article on installing Nginx and the output of php -v.