[BNM] Apache / Apache 2 / FastCGI / Lighttpd advice

[Dominic Mitchell]

On Mon, Dec 04, 2006 at 02:39:23PM +0000, Tom Hume wrote:
> On 4 Dec 2006, at 14:21, Jay Caines-Gooby wrote:
> 
> >> What does "crashing" actually mean?  Apache dieing? some bits not
> >> responding?
> > Kernel death and/or whole networking stack :-/
> > Sometimes the box reboots othertimes I have to use my remote re-boot
> > capability as I can't ssh in. :P
> 
> Wow. That sounds seriously screwy to me; I've seen Apache/FCGI  
> processes fall over and dump core thanks to library mismatches etc,  
> but bringing down the whole box? I'd look at hardware and check the  
> memory in the machine as a first step. There's no way user-level  
> processes should be doing that sort of thing.

Yup.  The OS shouldn't crash, no matter what.  It might be worth asking
your hosting provider about it.  Maybe they have access to the console
and can capture a kernel stack trace or something which might help to
diagnose what's going on...

>  From my (limited) experience lighttpd is ace: we benchmarked it  
> comfortably handling ludicrous loads (2000 req/second) on a small LVS  
> cluster of 1 load balancer and 2 years-old compaq boxes with about  
> 128mb ram.

Yeah, lighttpd sounds great.  It's in the same class of servers as
thttpd -- a single process doing async i/o, instead of the convulted
multi-thread or multi-process Apache setup.  It can be really fast.  In
Jay's situation, my only query is whether or not it works well with
mongrel...   Hmmm, the mongrel author disagrees:

  http://mongrel.rubyforge.org/docs/lighttpd.html

Shame.  But if you are interested in lighttpd, there's a really good
interview with the author on bsdtalk:

  http://bsdtalk.blogspot.com/2006/08/bsdtalk061-interview-with-lighttpd.html

Ditto for the cool varnish cache the Sevan mentioned.  :-)

  http://bsdtalk.blogspot.com/2006/10/bsdtalk072-interview-with-poul-henning.html

-Dom