Red Hat Bugzilla – Bug 1623
kill -HUP `cat /var/run/httpsd.pid` aggrevates memory leak
Last modified: 2008-05-01 11:37:49 EDT
It appears that the mod_perl shipped with secureweb
has a memory leak when configured as a DSO that can
be demostrated on a server that has the mod_perl
module included but doesn't actually call any perl
by hupping the server while watching the process
size. The mod_perl version grows considerably faster
than the non-mod_perl version.
I've tried a few different combinations:
perl 5004_01, perl 5004_04, mod_perl 1.16, 1.16_02,
This is all with redhat 4.2
I've read that it may be better to make mod_perl not
be a DSO but have it in the binary instead. Some of
the folks on the mod_perl list claim that helps.
This is inconvenient for you folks but perhaps you
would consider shipping the rpm either with two
binaries or in two flavors?
mod_perl 1.19 still has this memory leak. I am discussing things with
I'm seeing major problems with busy servers without using mod_perl,
too. My httpsd processes are growing to over 50 megs total size
pretty quickly if they see significant load.
The one test server I had with mod_perl included grew to over 150 megs
without any load at all. Any progress on this?
the concensus among mod_perl people seems to be to restart the server
periodically in a cron job if you are experiencing memory leaks. While the
problem may be fixed in a better way down the road, this is the current
solution, and while sub-optimal, it works.