Optimize PHP-FPM on VestaCP
/usr/local/vesta/data/templates/web/php-fpm/default.tpl /usr/local/vesta/data/templates/web/php-fpm/socket.tpl
Now the template files are default configurations. You can make the following changes to the template files to reduce the memory usage
In both default.tpl and socket.tpl change the line pm = dymanic to pm = ondemand and add the following lines below pm.max_spare_servers
pm.process_idle_timeout = 10s pm.max_requests = 500
Essentially, what this does is make PHP-FPM spawn PHP processes when needed, and let others die. By changing pm = dynamic to pm = ondemand we can use PHP-FPM’s feature to reduce its memory usage.
We have also added in pm.process_idle_timeout = 10s which will kill the PHP process after 10 seconds of inactivity, in an attempt to reduce PHP-FPM memory usage, and set pm.max_requests = 500. The pm.max_requests setting is advised as covered here (basically helping to stop memory leaks).
Rebuild All The PHP-FPM Configuration Files
Now this is where things get a bit tricky. Earlier it was not possible to rebuild all your web domain configurations at once. But you can now do the same with the v-rebuild-web-domains.
Run the following to command to rebuild all the config
Make sure that you backup the nginx.conf and snginx.conf as these will also be rebuilt. In case you have custom rules you can just restore the configs from your backup.
$ /usr/local/vesta/bin $ ./v-rebuild-web-domains [USER]
The argument [USER] for v-rebuild-web-domains will be the username for which the web domains will be regenerated.
Once all the domains are regenerated restart PHP-FPM with
$ service php-fpm restart
After this anytime you create a new web domain it will be created with the optimized configuration and rebuilding again is not required.
That’s it. You have now successfully optimized PHP-FPM.
Conclusion
After we made these changes, the datacenter resource utilization was approximately 40% lower of what it used to be.
We use NodeQuery to monitor our servers and below is the hourly RAM & SWAP utilization graph
