Make PHP-FPM Listen at 'IPAddress:Port' Instead of '/var/run/php5-fpm.sock;'

Question

As I was load testing my site today (using blitz.io); despite lots of RAM (more than 50%) and CPU power (over 70%) available, results showed that my site started timing out at a certain number of concurrent users per second.

Nginx error log for my site (/var/log/nginx/example.com.error.log) showed something like this:

2013/02/12 19:03:57 [error] 13749#0: *3175 connect() to unix:/var/run/php5-fpm.sock failed (11: Resource temporarily unavailable) while connecting to upstream, client: 54.123.456.46, server: example.com, request: "GET / HTTP/1.1", upstream: "fastcgi://unix:/var/run/php5-fpm.sock:", host: "example.com"

Googling the error led me to this answer which states using TCP\IP connection instead of unix socket as the solution to the problem; as unix socket's "problems on high-load cases is well-known".

So, as suggested by the answer:

• I replaced listen = /var/run/php5-fpm.sock with listen 127.0.0.1:9000 in /etc/php5/fpm/pool.d/www.conf

• As there's no /etc/nginx/php_location on my distrio (Debian Wheezy), I did nothing about it.

• Since I use fastcgi_pass unix:/var/run/php5-fpm.sock; in the Nginx configuration file for my site, i.e., /etc/nginx/sites-available/example.com, I replaced it with fastcgi_pass 127.0.0.1:9000;

Now the problem is, I get a 502 Bad Gateway error when I visit my website. Yes, I did reload Nginx and PHP-FPM. What am I doing wrong? (A total newbie here, doing my best to learn by doing.)

In case this is relevant, when I do sudo service php5-fpm restart, I get this error:

[FAIL] Restarting PHP5 FastCGI Process Manager: php5-fpm failed!

And this is happening only since I made the aforementioned changes. How can I fix this?

Please let me know if I should get more information.

---

UPDATE

The file /etc/nginx/sites-available/default says this:

#   # With php5-cgi alone:
#   fastcgi_pass 127.0.0.1:9000;

#   # With php5-fpm:
#   fastcgi_pass unix:/var/run/php5-fpm.sock;

So, does that mean, if my server is running PHP-FPM, it SHOULD, without a choice, use /var/run/php5-fpm.sock?

Solution 1

I used $ sudo php5-fpm -t command to test if PHP-FPM's settings are fine (if not, it'll show me some error/info).

So, here's what the output looked like:

[13-Feb-2013 18:35:00] ERROR: [/etc/php5/fpm/pool.d/www.conf:33] value is NULL for a ZEND_INI_PARSER_ENTRY
[13-Feb-2013 18:35:00] ERROR: Unable to include /etc/php5/fpm/pool.d/www.conf from /etc/php5/fpm/php-fpm.conf at line 33
[13-Feb-2013 18:35:00] ERROR: failed to load configuration file '/etc/php5/fpm/php-fpm.conf'
[13-Feb-2013 18:35:00] ERROR: FPM initialization failed

The error says, something's wrong in Line 33 of of /etc/php5/fpm/pool.d/www.conf, which happens to be this: listen 127.0.0.1:9000 (not so much of a coincidence, is it?).

After seeing it, I immediately compared it with other lines, and then it struck me, an = (equal-to sign) is missing!

So, this is what it's supposed to be: listen = 127.0.0.1:9000 and that fixed everything!

Solution 2

Sounds like (based on the 502 and the error message) that php-fpm service is failing to start.

Could something else be using port 9000? To check run:

sudo lsof -P | grep TCP | grep LISTEN

and look for something like TCP localhost:9000 (LISTEN). If there is, you could just use a different port, say 9001.

Good place to check would be the php-fpm error log. It may not be enabled by default. If you look at your php-fpm conf file (/etc/php5/fpm/php-fpm.conf on my system) you'll find the 'error_log' setting. This is the path to your error log. If this is commented out, un-comment it, restart the php-fpm service again and check what the contents of the log file say.

Solution 3

You can use IP/port or Unix socket on your choices.

This depends on your php-fpm config, the file is www.conf, in my system (Ubuntu 16.04), this file located at /etc/php/7.0/fpm/pool.d.

Within this file, you can find a directive named listen, if you set this directive to /run/php/php7.0-fpm.sock, for example, then in your nginx site config fastcgi_pass should be unix:/run/php/php7.0-fpm.sock;

But if you set listen to 127.0.0.1:9000, then fastcfg_pass should be 127.0.0.1:9000.

Comments

[mgorven]

Please paste your entire www.conf.

[its_me]

(1) No, port 9000 doesn't seem to be used by any other service. Here's the output of $ sudo lsof -P | grep TCP | grep LISTEN (2) here are the contents of PHP-FPM log file at /var/log/php5-fpm.log -- paste.kde.org/670802, I still dont understand what to do...?!

[chrskly]

Looks like it's still trying to use the file socket NOTICE: using inherited socket fd=6, "/var/run/php5-fpm.sock". Is there a listen setting in /etc/php5/fpm/php-fpm.conf?

[its_me]

No. But it has this: include=/etc/php5/fpm/pool.d/*.conf which probably means it includes /etc/php5/fpm/pool.d/www.conf, in which I did make changes as mentioned in my question.

[its_me]

I got more information from sudo php5-fpm -t: paste.kde.org/670826 -- as shown in the error, line 33 of /etc/php5/fpm/pool.d/www.conf is this: listen 127.0.0.1:9000

[its_me]

The line 33 of www.conf is listen 127.0.0.1:9000 which I am using instead of listen = /var/run/php5-fpm.sock as the latter is causing time outs under load (as mentioned in my question).