Memory leak in combination with Starman

There seems to be a serious memory leak when running Foswiki on starman. After some time (~1day) of operation this message appears to the user

Foswiki detected an internal error - please check your Foswiki logs and webserver logs for more information.

Can't fork: Cannot allocate memory

and the process table looks like

app      21927  0.0  0.9  15648 10092 ?        Ss   Mar27   0:03 starman master /srv/web/propaganda.pm/srv/www/foswiki.psgi --workers 5 --pid /var/run/fos
app      21928  0.6 16.1 172400 166932 ?       S    Mar27   6:00 starman worker /srv/web/propaganda.pm/srv/www/foswiki.psgi --workers 5 --pid /var/run/fos
app      21929  0.5 12.9 138356 133772 ?       S    Mar27   4:51 starman worker /srv/web/propaganda.pm/srv/www/foswiki.psgi --workers 5 --pid /var/run/fos
app      21930  0.4 11.2 121232 116628 ?       S    Mar27   4:04 starman worker /srv/web/propaganda.pm/srv/www/foswiki.psgi --workers 5 --pid /var/run/fos
app      21931  0.5 13.3 142700 138012 ?       S    Mar27   4:48 starman worker /srv/web/propaganda.pm/srv/www/foswiki.psgi --workers 5 --pid /var/run/fos
app      21932  0.6 14.8 159824 153376 ?       S    Mar27   5:31 starman worker /srv/web/propaganda.pm/srv/www/foswiki.psgi --workers 5 --pid /var/run/fos

A simple restart helps, but a more sustainable solution would be welcome.

-- RichardJelinek - 28 Mar 2013

Using FastCGIEngineContrib, the fcgi backends restart themselves every 1000 calls (configurable).

-- MichaelDaum - 28 Mar 2013

You could use the https://metacpan.org/pod/Plack::Middleware::SizeLimit for the automatic worker restarts, until the leak will be closed. Works nicely, just determine the right memory size. (Use the log_when_limits_exceeded => 1 parameter to see, when the harakiri is executed).

-- JozefMojzis - 08 Dec 2015