Description:
------------
I'm running nginx 1.2.4 with php5-fpm (5.4.8) issuing

 ab -n 20000 -c 1000 http://localhost/index.php


When 

pm = dynamic
pm.max_children = 3000
pm.start_servers = 100  <- this is important
pm.max_spare_servers = 1000

I'm having a lot of errors in nginx talking to php


[error] 26125#0: *49177 upstream prematurely closed connection while reading 
response header from upstream, ... "fastcgi://127.0.0.1:9000"

[error] 26125#0: *56030 recv() failed (104: Connection reset by peer) while 
reading response header from upstream, ... upstream: "fastcgi://127.0.0.1:9000"


The fpm log shows a lot of 

WARNING: [pool www] seems busy (you may need to increase pm.start_servers, or 
pm.min/max_spare_servers), spawning 32 children, there are 0 idle, and X total 
children


When spawning new children reaches a concrete amount nginx stops issuing such 
errors. This leads me to believe freshly spawned processes do not execute and 
respond correctly.


Now, when I start with 3000 children straight ahead, all connections work 
properly. No errors in nginx or fpm.

pm = dynamic
pm.max_children = 3000
pm.start_servers = 3000  <- Straight ahead
pm.max_spare_servers = 3000

Test script:
---------------
The php code on the server is relatively simple.

apc_inc('count111'); sleep(1); echo 'x';