Facebook
Twitter
On Plesk 12.5, we're starting to experiment with using PHP-FPM instead of FastCGI for our customers. However, we are frequently running into an issue where php is broken across every site on the server due to the PHP-FPM daemon not starting. This has been tracked back to Plesk allowing subdomains to be deleted without cleaning up the config file in /etc/php-fpm.d/. That causes the FPM daemon to fail with:
[24-May-2016 12:32:22] ERROR: [pool test.domain.com] the prefix '/var/www/vhosts/system/test.domain.com' does not exist or is not a directory
[24-May-2016 12:32:22] ERROR: failed to post process the configuration
[24-May-2016 12:32:22] ERROR: FPM initialization failed
At that point, php is broken on every site on the server.
This issue is documented in KB https://kb.plesk.com/en/127183 where Plesk's solution is just oh, your php is broken, well go remove the file that's breaking it.
The Self Repair tool does NOT fix the issue; i.e., if there is a now-bogus file in /etc/php-fpm.d/ referencing a site that does not exist, and you tell Plesk to do a self repair of all web server config, it will not go remove the file, so php-fpm remains broken.
Having watchdog monitor it doesn't matter because it can't be restarted at this point anyway.
The value that seems to cause the problem is this:
php_value[open_basedir] = "/var/www/vhosts/test.domain.com/:/tmp/"
since the failure is based on a directory, and that appears to be the only site-specific directory Plesk puts in the config file for each php-fpm enabled domain.
Other than setting every site to have an open_basedir of none, has anyone found a workaround for this? Any chance the php-fpm config files support conditionals so you could test for the directory's existence before the config directive is parsed?
[24-May-2016 12:32:22] ERROR: [pool test.domain.com] the prefix '/var/www/vhosts/system/test.domain.com' does not exist or is not a directory
[24-May-2016 12:32:22] ERROR: failed to post process the configuration
[24-May-2016 12:32:22] ERROR: FPM initialization failed
At that point, php is broken on every site on the server.
This issue is documented in KB https://kb.plesk.com/en/127183 where Plesk's solution is just oh, your php is broken, well go remove the file that's breaking it.
The Self Repair tool does NOT fix the issue; i.e., if there is a now-bogus file in /etc/php-fpm.d/ referencing a site that does not exist, and you tell Plesk to do a self repair of all web server config, it will not go remove the file, so php-fpm remains broken.
Having watchdog monitor it doesn't matter because it can't be restarted at this point anyway.
The value that seems to cause the problem is this:
php_value[open_basedir] = "/var/www/vhosts/test.domain.com/:/tmp/"
since the failure is based on a directory, and that appears to be the only site-specific directory Plesk puts in the config file for each php-fpm enabled domain.
Other than setting every site to have an open_basedir of none, has anyone found a workaround for this? Any chance the php-fpm config files support conditionals so you could test for the directory's existence before the config directive is parsed?
