http://bugzilla.spamassassin.org/show_bug.cgi?id=3983
------- Additional Comments From [EMAIL PROTECTED] 2004-11-23 13:30 ------- Subject: Re: [review] adopt Apache preforking algorithm I asked Dean Gaudet, one of the Apache httpd hackers, about this a ways back but I don't think I ever forwarded this (he gave his okay to forward). ------- start of cut text -------------- From: dean gaudet <[EMAIL PROTECTED]> apache keels over if you don't set MaxClients appropriately for your RAM... it's one of the first few things i mention on the apache tuning page. you really shouldn't attempt to defend against dum setups with some complicated accept() protocol... keep it simple. reacting to load rather than launching a fixed number of children is an improvement though (or more accurately... a bug fix -- since pre-3.0 behaved like this) -- then you can default to launching one child, and that will handle 99% of j. random user's needs. i was honestly happy with post-forking SA... there are already 4 other post-forked daemons in my mail delivery setup (qmail+clamav). it's the most trivial to get post-forking correct in a platform independent manner... and it's trivial to have post-fork react to load changes. but then i'm running SA with one request per child, because i prefer to exchange the perf loss for the extra paranoia of having my children drop root privs entirely early... (and there's that config leak bug at least in the 3.0.1 my install is based on). damn does SA really have 20MB unshared? suck. so yeah sorry, no sage advice here... i'll be disabling most of this myself, security/paranoia is more important. ------- end ---------------------------- ------- You are receiving this mail because: ------- You are the assignee for the bug, or are watching the assignee.
