Of course, this new feature is enabled in the RPM of PHP 5.4.16 in Fedora 18 and PHP 5.5.0RC3 in Fedora 19, and the ones in my repository (fedora >= 16).

In the service description (unit) file, we now have the line :

Type=notify

This means that systemd will receive start status and health information send by the main php-fpm process. The delay between refresh can be configured using the  systemd_interval option (in /etc/php-fpm.conf), default is 10" (0 to disable).

After the start-up, the "Status" line is:

      Status: "Ready to handle connections"

Then it is refreshed and provides information about the service's health:

      Status: "Processes active: 0, idle: 5, Requests: 101, slow: 2, Traffic: 6.2req/sec"

Which are, the number of active / available processes, the number of requests and slow requests since the service start-up, and the average traffic (requests by sec) since the last update.

I also fix a bug discovered buring my tests:

- FPM:
  . Fixed Bug #64915 (error_log ignored when daemonize=0). (Remi)

Before this fix, whatever the value of error_log option, all the messages from php-fpm were saved in the system log instead of the configured file (/var/log/php-fpm/error.log). Of course, it is possible to use the system log, need to be explicitly set:

error_log = syslog

Another feature available with this systemd integration is the Watchdog mode of systemd.

The service configuration must be changed, by editing a copy of the unit file in /etc:

cp /lib/systemd/system/php-fpm.service /etc/systemd/system/php-fpm.service

Starting with  Fedora 19 it is also possible to simply add a /etc/systemd/system/php-fpm.service.d/watchdog.conf file.

With, at least the option:

[service]
WatchdogSec=30

In the configuration, the value of systemd_interval option is ignored, and half of systemd configured value is used instead.

Here is a good way to monitor a service and have systemd run the necessary action (service restart, for example).

Is there someone who still prefer old SysV scripts ?