httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Dean Gaudet <>
Subject Re: fcntl problem in apache-apr.
Date Tue, 13 Apr 1999 19:39:57 GMT
Uh, you should need at most one worker to accept at most one more
connection from the acceptor thread... no? 

Well, then I also think the whole idea of an acceptor thread is a waste ;)


On Tue, 13 Apr 1999, Ryan Bloom wrote:

> The problem is seen most notably when you have many ports, and some of
> them aren't used as often as others.  We use one file per socket to do our
> locking, instead of one file for the whole server.
> Because of this, if one of the accept threads is in fcntl for a rarely
> used socket, that could keep the whole process from exiting.  This becomes
> a really big problem on OS's like Linux 2.0.x, where there is a limit on
> the number of threads per user.  Because we have to kill our workers
> after we kill our acceptor threads (otherwise, we will lose requests), we
> have to our acceptors as quickly as possible.  If we wait for the process
> to exit, we can hit our thread limit very easily.
> Ryan
> On Tue, 13 Apr 1999, Dean Gaudet wrote:
> > Uh, isn't it a lot easier to redefine MaxRequestsPerChild as "a child will
> > start shutting down after this many requests have been served, but will
> > remain until all of its threads have exited".  Then for people who are
> > totally anal then can set MaxRPC to N - K where K is the number of threads
> > in the child.  Or some other crud like that. 
> > 
> > i.e. I don't see why it has to be totally accurate, and the problem would
> > seem to be solved trivially by just letting the process serve another
> > request.
> > 
> > But then I haven't looked at apache-apr. 
> > 
> > Dean
> > 
> > 
> _______________________________________________________________________
> Ryan Bloom
> 4205 S Miami Blvd	
> RTP, NC 27709		It's a beautiful sight to see good dancers 
> 			doing simple steps.  It's a painful sight to
> 			see beginners doing complicated patterns.	

View raw message