Lighttpd

I think this is an issue with php-cgi spawn from lighttpd. When lighttpd restarts, the old lighttpd shuts down the port it listens to so the new process can bind to it. However the old php-cgi parent still binds to the same port (1026 for example), which the new php-cgi cannot bind, that leads to failure to start.

My solution is to use spawn-fcgi even in localhost so you can manage your PHP FastCGI and lighttpd processes independently.

See http://hostingfu.com/article/use-spawn-fcgi-for-lighttpd-php-fastcgi

-- Scott

If you killed all PHP procecsses, the PHP port should be free.

Same problem here for 1.4.16 (Debian backport). From time to time, php-cgi processes are not linked to lighttpd, so when a restart is initiated, php doesn't suppress its children, and no php process is created again.
I must point out the fact that when running php from within a php script (using system('php-cgi ...') or system('/a/php/file/with/good/headers.php')) the process goes directly to a zombie one and stop working at all. I couldn't manage to make it work correctly so far...

-- Yvan

Same problem here for 1.4.16 (Debian backport).

Can you try 1.4.18?

I must point out the fact that when running php from within a php script (using system('php-cgi ...') or system('/a/php/file/with/good/headers.php'))

Shouldn't you use php-cli? Running it that way is not CGI.

Same here
lighttpd-1.4.18
CentOS 5.1
PHP 5.2.5
even killall php-cgi does nothing and it keeps holding memory, only restarting the server would solve the problem.

-- masryalex

Replying to masryalex@gmail.com:

Same here
lighttpd-1.4.18
CentOS 5.1
PHP 5.2.5
even killall php-cgi does nothing and it keeps holding memory, only restarting the server would solve the problem.

try killall -HUP php-cgi