Dynamic server spawning broken in 1.3?

[ra...@bellglobal.com]

I am stress-testing the current 1.3b1 code.  It doesn't seem to spawn new
servers when needed.

Despite getting repeated messages like the following:
[Wed Sep 17 10:32:04 1997] [error] server seems busy, spawning 8 children (you
may need to increase StartServers, or Min/MaxSpareServers)

and with the following settings:

   MinSpareServers 25
   MaxSpareServers 50
   StartServers 25
   MaxClients 250

A 'ps -ef' throughout the testing shows that the number of httpd's running
at any one time is never more than 25.  So I guess it is lying when it is
telling me it is spawning 8 new children?

This is on an Irix 6.2 box using httpd compiled with the SGI optimizing
C compiler.

My stress tester launches 10 concurrent recursive wget's and it does this
10 times for a total of 100 gets of the entire document tree.  This
should mean that there are actually never more than 10 active http sessions
at any one time.  Anybody have some insight?  Am I misinterpreting
something here?

-Rasmus

Re: Dynamic server spawning broken in 1.3?

[Dean Gaudet <dg...@arctic.org>]

I know it definately spawns more under linux ... you'd really need to hit
it with a lot of concurrency to prove that it's not just satisfying the
load with a smaller number of children.  My test case for testing the
spawning is to open two hundred connections and not ask a thing on any of
them, just hold 'em open.

It's pretty safe to run a par (irix for strace/ktrace/truss) against the
parent, you'll see it spawning and you'll see it killing children off.
It's not very active so you don't have to worry about it spamming you.

Dean

On Wed, 17 Sep 1997 rasmus@bellglobal.com wrote:

> I am stress-testing the current 1.3b1 code.  It doesn't seem to spawn new
> servers when needed.
> 
> Despite getting repeated messages like the following:
> [Wed Sep 17 10:32:04 1997] [error] server seems busy, spawning 8 children (you
> may need to increase StartServers, or Min/MaxSpareServers)
> 
> and with the following settings:
> 
>    MinSpareServers 25
>    MaxSpareServers 50
>    StartServers 25
>    MaxClients 250
> 
> A 'ps -ef' throughout the testing shows that the number of httpd's running
> at any one time is never more than 25.  So I guess it is lying when it is
> telling me it is spawning 8 new children?
> 
> This is on an Irix 6.2 box using httpd compiled with the SGI optimizing
> C compiler.
> 
> My stress tester launches 10 concurrent recursive wget's and it does this
> 10 times for a total of 100 gets of the entire document tree.  This
> should mean that there are actually never more than 10 active http sessions
> at any one time.  Anybody have some insight?  Am I misinterpreting
> something here?
> 
> -Rasmus
>