PECL/memcached segfaults

I was finally able to reproduce the cause on my dev server. The 
segfaults were caused by the process exceeding its open file limit. This 
explains why only one data center had segfaults (the busier one) and the 
other only saw increasing connections to memcached.

On a dev server, running the worker MPM and mod_php, I configured the 
server with:

         StartServers                2
         MinSpareThreads            20
         MaxSpareThreads            20
         ThreadsPerChild            10
         MaxClients                 20
         MaxRequestsPerChild         0

This creates 2 Apache processes, each with 10 threads for a total of 20 
children.

I am using persistent connections when creating the memcached connections.

The memcached connection count grows and grows and grows until the 
process hits 1024 open files and is killed by the kernel.

I did write the persistent string being sent to the object constructor 
out to the php error log to ensure we were not sending a unique string 
each time. We are not. At most there should be two connections per 
child/process (not sure how the persistence is handled for threads). 
pecl/memcache creates them per child (thread).

I turned off the persistence and things work as they should. The 
connection count does not grow.

So, next I turned off worker thinking this was a threading issue. But, 
the problem persisted.

In the end, I was able to reproduce the issue with a simple script. Here 
is the script and output: http://pastebin.com/40vzVM7D There is a run 
with pecl/memcache there as well to show the difference in behavior.

Tested platforms:

Gentoo Linux
PHP/5.2.13
libmemcached/0.38 thru 0.40
pecl-memcached/1.0.0 thru 1.0.2

Mac - Darwin Kernel Version 10.3.0
PHP 5.3.1
libmemcached/0.40
pecl-memcached/1.0.2

-- 

Brian.
--------
http://brian.moonspot.net/