Improved php performance

Personally I am not a big fan of PHP not that it is a bad programming language, however that simplicity of development has brought are lot of page badly programmed. Result I have a negative bias of this language, I am honest in my remarks, however it is a reality that pages php are very present in our daily lives.

So we’ll see how we can improve the performance of our apache system when we need to run php. We have seen how handled the CPU problem using a cache system, we saw how to deal with the problem of IO when writing file using a file system in memory. On the other hand the ideal solution for the problem of memory remains the use of the module of process management worker unfortunately mod_php can not work with this last one. Let’s see a solution to this problem, we will use php5-fpm.

Use in cgi mode (php5-fpm)

PHP-FPM (FastCGI Process Manager) is an SAPI interface for communication between a Web server and PHP, based on the FastCGI protocol and written by Andrei Nigmatulin.

Installation & configuration of php5-fpm

Thank you Debian we have a pasting for that: P.

$ sudo apt-get install php5-fpm libapache2-mod-fastcgi

This will allow us to have a Fastcgi service available for apache, but that’s what it Fastcgi in other words instead that apache is able to interpret itself php files run the / usr / bin / php command to build the page. The downside if you use this method is that I’ve already seen in the pass a small change of behavior in php pages when the latter expects to “live” in apache. We had encountered a problem mainly in environment variables that are not present because not running under apache.

Following the operation you can see that the FastCGI module is already active:

$ ls -l /etc/apache2/mods-enabled/*fast* lrwxrwxrwx 1 root root 30 Jul 8 17:05 /etc/apache2/mods-enabled/fastcgi.conf -> ../mods-available/fastcgi.conf lrwxrwxrwx 1 root root 30 Jul 8 17:05 /etc/apache2/mods-enabled/fastcgi.load -> ../mods-available/fastcgi.load

We will disable the mod_php and put the system in worker mode:

$ sudo a2dismod php5 $ sudo a2dismod mpm_prefork $ sudo a2enmod mpm_worker

We could restart apache at this point but we will set up the configuration for php-fpm. We will create the configuration file /etc/apache2/conf-available/php5-fpm.conf

AddHandler php5-fcgi .php Action php5-fcgi /php5-fcgi Alias /php5-fcgi /usr/lib/cgi-bin/php5-fcgi FastCgiExternalServer /usr/lib/cgi-bin/php5-fcgi -socket /var/run/php5-fpm.sock -pass-header Authorization

Let’s analyze the configuration file:

• AddHandler: we add an interpreter for files whose extension ends with .php when this type of file will be processed we will call the handler (the interpreter) php-fcgi

• Action: Following the identification of the file extension we define the action to take we tell him to communicate with / php5-fcgi

• Alias: We have already seen Alias mode this allows us to redirect a destination in the URL to another PATH on the server. In our case it will be / usr / lib / cgi-bin / php5-fcgi

• FastCgiExternalServer: To finish we specify that when communicating to the / usr / lib / cgi-bin / php5-fcgi file it must pass it to the /var/run/php5-fpm.sock socket via the FastCgi system.

WOOT, I almost understood: P, let’s see the files now:

$ ls -l /usr/lib/cgi-bin/php5-fcgi ls: cannot access /usr/lib/cgi-bin/php5-fcgi: No such file or directory $ ls -l /var/run/php5-fpm.sock ls: cannot access /var/run/php5-fpm.sock: No such file or directory

Well there if you just copy pasted configuration files you say to yourself, here is Thomas wrote the documentation after the victory of France against Germany at EURO 2016. He was in the potatoes … No panic j have cuvée since: P

The / usr / lib / cgi-bin / php5-fcgi file is not true but only to force the FastCgiExternalServer call, we will create an empty file.

$ sudo touch /usr/lib/cgi-bin/php5-fcgi $ ls -ld /usr/lib/cgi-bin/php5-fcgi -rw-r--r-- 1 root root 0 Jul 8 17:24 /usr/lib/cgi-bin/php5-fcgi

The file /var/run/php5-fpm.sock is a socket defined by the php5-fpm service so we will start it, of course you also have to make sure that this service restarts during a restart of the server all as apache.

$ sudo /etc/init.d/php5-fpm start $ sudo /etc/init.d/php5-fpm status * php5-fpm is running $ ls -l /var/run/php5-fpm.sock srw-rw---- 1 www-data www-data 0 Jul 8 17:26 /var/run/php5-fpm.sock

Let’s activate the configuration:

$ cd /etc/apache2/conf-enabled/ $ sudo ln -s ../conf-available/php5-fpm.conf

Good it looks good GO?

$ sudo apachectl configtest && sudo /etc/init.d/apache2 restart AH00526: Syntax error on line 3 of /etc/apache2/conf-enabled/php5-fpm.conf: Invalid command 'Action', perhaps misspelled or defined by a module not included in the server configuration Action 'configtest' failed. The Apache error log may have more information.

Oupss it was missing a module

$ sudo a2enmod actions

RE GO ?

$ sudo apachectl configtest && sudo /etc/init.d/apache2 restart Syntax OK * Restarting web server apache2 ...done.

Validate the URL: https://www.linux202-sitea.com/info.php Issh: You do not have permission to access /php5-fcgi/info.php on this server.

Validate the logs, I like to put some errors in my training this allows to see a little practice analysis because unfortunately it rarely works the first time: P, it will probably not the same mistakes but it offers a demonstration of analysis.

$ tail /data/vhosts/siteA/logs/error-ssl.log [... OUPUT COUPÉ ...] [Fri Jul 08 17:29:39.669126 2016] [authz_core:error] [pid 473:tid 3051350848] [client 172.17.42.1:43232] AH01630: client denied by server configuration: /usr/lib/cgi-bin/php5-fcgi

Why ?? Remember at the level of security I like blocked the whole system with since the root result all new configuration obliges me to open the access … We will therefore modify the configuration to allow access to the directory / usr / lib / cgi-bin /. Open the file

AddHandler php5-fcgi .php Action php5-fcgi /php5-fcgi Alias /php5-fcgi /usr/lib/cgi-bin/php5-fcgi FastCgiExternalServer /usr/lib/cgi-bin/php5-fcgi -socket /var/run/php5-fpm.sock -pass-header Authorization # Authorization de l'access au repertoire cgi-bin Require all granted

It’s gone: D:

$ sudo apachectl configtest && sudo /etc/init.d/apache2 restart

performance test with php5-fpm

Visualization of mpm_worker configurations:

$ cat /etc/apache2/mods-enabled/mpm_worker.conf StartServers 2 MinSpareThreads 25 MaxSpareThreads 75 ThreadLimit 64 ThreadsPerChild 25 MaxRequestWorkers 25 MaxConnectionsPerChild 0

We go on the URL: https://www.linux202-sitea.com/info.php: WOOT it works … Go is the time of stress tests. For the calculation of the use of the RAM, I changed a little the command because in addition to apache we must be able to calculate the service php:

$ ps aux | tr -s " " | grep ^www-data www-data 422 0.0 0.2 86620 4700 ? S 17:26 0:00 php-fpm: pool www www-data 423 0.0 0.4 86620 9912 ? S 17:26 0:00 php-fpm: pool www www-data 551 0.0 0.1 8380 3200 ? S 17:36 0:00 /usr/sbin/fcgi-pm -k start www-data 552 0.0 0.3 232920 7016 ? Sl 17:36 0:00 /usr/sbin/apache2 -k start $ ps aux | tr -s " " | grep ^www-data | cut -d " " -f 6,11- | egrep -v "grep|htcache" | awk '{ sum+=$1} END {print sum/1024 " Megs"}' 24.2461 Megs

So we have the apache2 process but also all processes related to php management

I can not wait to do the load test so GO :).

[server]$ uptime 17:41:51 up 47 min, 1 user, load average: 0.53, 0.47, 0.41 [server]$ ps aux | tr -s " " | grep ^www-data | cut -d " " -f 6,11- | egrep -v "grep|htcache" | awk '{ sum+=$1} END {print sum/1024 " Megs"}' 24.2461 Megs [client]$ ab -n 100 -c 10 -l https://www.linux202-siteA.com/info.php Percentage of the requests served within a certain time (ms) 50% 241 66% 263 75% 287 80% 294 90% 313 95% 339 98% 353 99% 361 100% 361 (longest request) [server]$ ps aux | tr -s " " | grep ^www-data | cut -d " " -f 6,11- | egrep -v "grep|htcache" | awk '{ sum+=$1} END {print sum/1024 " Megs"}' 29.543 Megs [server]$ uptime 17:43:26 up 49 min, 1 user, load average: 0.95, 0.59, 0.46

Second test:

[server]$ $ uptime 17:46:41 up 52 min, 1 user, load average: 0.45, 0.57, 0.48 [server]$ ps aux | tr -s " " | grep ^www-data | cut -d " " -f 6,11- | egrep -v "grep|htcache" | awk '{ sum+=$1} END {print sum/1024 " Megs"}' 29.543 Megs [client]$ ab -n 300 -c 30 -l https://www.linux202-siteA.com/info.php Percentage of the requests served within a certain time (ms) 50% 781 66% 842 75% 886 80% 933 90% 1106 95% 1229 98% 1348 99% 1404 100% 1417 (longest request) [server]$ uptime 17:46:55 up 52 min, 1 user, load average: 2.42, 1.00, 0.62 [server]$ ps aux | tr -s " " | grep ^www-data | cut -d " " -f 6,11- | egrep -v "grep|htcache" | awk '{ sum+=$1} END {print sum/1024 " Megs"}' 39.4375 Megs

REF : https://www.digitalocean.com/community/questions/apache-2-4-with-php5-fpm

Comparative with prefork and mod_php5

Let’s resume the test with the mod_prefork with 100 requests including 10 competitors, here is the configuration of the prefork:

$ cat /etc/apache2/mods-enabled/mpm_prefork.conf StartServers 5 MinSpareServers 5 MaxSpareServers 10 # MaxRequestWorkers 150 MaxRequestWorkers 25 MaxConnectionsPerChild 0

It’s gone for the charge test

[server]$ uptime 08:40:58 up 22 min, 1 user, load average: 0.17, 0.44, 0.41 [server]$ ps aux | tr -s " " | cut -d " " -f 6,11- | grep apache | egrep -v "grep|htcache" | awk '{ sum+=$1} END {print sum/1024 " Megs"}' 59.332 Megs [client] $ ab -n 100 -c 10 -l https://www.linux202-siteA.com/info.php Connection Times (ms) min mean[+/-sd] median max Connect: 51 191 65.0 197 334 Processing: 9 41 23.0 34 133 Waiting: 1 21 16.0 18 95 Total: 65 232 73.9 230 407 Percentage of the requests served within a certain time (ms) 50% 230 66% 269 75% 289 80% 296 90% 327 95% 340 98% 383 99% 407 100% 407 (longest request) [server]$ ps aux | tr -s " " | cut -d " " -f 6,11- | grep apache | egrep -v "grep|htcache" | awk '{ sum+=$1} END {print sum/1024 " Megs"}' 130.27 Megs [server]$ uptime 08:41:58 up 23 min, 1 user, load average: 0.14, 0.38, 0.39

Second test with 300 queries including 30 competitors:

[server]$ uptime 08:30:49 up 13 min, 1 user, load average: 0.11, 0.18, 0.21 [server]$ ps aux | tr -s " " | grep ^www-data | cut -d " " -f 6,11- | egrep -v "grep|htcache" | awk '{ sum+=$1} END {print sum/1024 " Megs"}' 43.6719 Megs [client]$ ab -n 300 -c 30 -l https://www.linux202-siteA.com/info.php Percentage of the requests served within a certain time (ms) 50% 736 66% 810 75% 845 80% 867 90% 940 95% 986 98% 1135 99% 1233 100% 1284 (longest request) [server]$ $ ps aux | tr -s " " | grep ^www-data | cut -d " " -f 6,11- | egrep -v "grep|htcache" | awk '{ sum+=$1} END {print sum/1024 " Megs"}' 324.855 Megs [server]$ uptime 08:31:56 up 14 min, 1 user, load average: 2.36, 0.67, 0.36

conclusion

In the current situation with a server that has very little CPU we have very similar results in both modes, in terms of the response time of web pages. We can also see that the load on the CPU is very similar if we refer to the load average of the system. We can see a big improvement in the use of the system at the memory level we have a huge economic using the mpm_worker mode.