Buddy Lee Haystack <[EMAIL PROTECTED]> wrote:
> The memory consumption increases by about 1 megabyte for each child
> process every time I issued the USR1 signal. Which is kind of what I
> expected given Jens-Uwe Mager's explanation of DSO's failings with
> mod_perl.
FWIW, I get a 240k increase in httpd size every time I send a USR1
signal to apache, and that is with a 100% static Apache/mod_perl,
without even mod_so compiled in.
Doing a few "ps"s during the graceful restart process, it looks like all
child processes die, then the parent grows and starts respawning new
children.
It smells to me like the leak is in mod_perl reloading the perl modules
themselves. If I trim httpd.conf down so it only loads a very minimal
"hello world" handler, then the leak goes down to 64k per restart.
I've gathered from this list that the 'canonical' answer to these
problems is Apache::Reload, but I've been reluctant to use it because it
would presumably reload the modules separately in each child, and
therefore leaving them unshared.
So for the moment what I do is restart apache for real every couple
weeks. It's only a couple seconds of downtime...
--
Roger Espel Llima, [EMAIL PROTECTED]
http://www.iagora.com/~espel/index.html