Re: Processes dying prematurely

Ben,

I think it would be worth finding out who on the PHP team wrote this code, and 
ask them why:

-  it's exiting so much, and;
-  why your PHP and my PHP seem to function differently

In my case, I have the fcgi-pm forking the php-pm, which forks the php-workers. 
In your case, the fcgi-pm forks the php-workers directly.  This is BAD, because 
under high load the workers are killing themselves too quickly.  You could try 
tuning the dynamic proc manager to be more tolerant of things dying quickly (ie, 
move it down from 30 seconds to 5 seconds) which will at least stop the symptom.

I have a php-pm under 4.3.X, 5.0.X, and 5.1.X - it works the same across the 
board.  We seem to have them built the same way, yet the end result is 
manifestly different.

David.

-----

On Thu, 24 Aug 2006, Ben Chabot wrote:

> Well, after more investigation, I saw that on the good vhost, the daemon
> was resetting every 500 requests and restarting.  On the other one, it
> was resetting after 500, and not restarting.  I wonder if it's because
> on the one that was not working, it seems to require multiple php
> scripts to run for each page load.  Like it was dying in the middle of a
> page load?
>
> At any rate, I looked at the php cgi source :
>
> [root@berkana cgi]# pwd
> /usr/amnet/php-4.4.2/sapi/cgi
>
> And changed what was 500 to 2000 :
>
> [root@berkana cgi]# grep 2000 cgi_main.c
>        int max_requests = 2000;
>
> I know you can do this in the environment too, but this seems to have
> fixed it on the vhost that was not working.  It's as if, fcgi-pm
> restarts php before it hits the 2000 limit.  Is fcgi-pm's restart based
> on time? In the log file it's the normal sequence now... (1) termination
> signaled (2) terminated (3) restarted.
>
> For all the vhosts, none are dying spontaneously... yet anyway.  Do you
> have any idea why this seems to have improved the situation?  It seems
> to me that if fastcgi is going to be restarting the php-fcgi daemon,
> then why does it need to kill itself after X many requests?
>
> (And David Birnbaum, you were right, even though I hadn't set the
> environment variable, it was still dying after 500 requests, very odd.)
>
> Thanks,
> Ben
>
> Chris Lightfoot wrote:
>> build a copy of PHP with debugging symbols and attach gdb
>> to the instance which dies frequently; put a break on the
>> exit function and _exit syscall, and wait for the process
>> to exit. When it does, examine the state of the process to
>> identify what's causing it to do so.
>>
>>
> ___________________________________
> fastcgi-developers mailing list
> http://fastcgi.com/fastcgi-developers/
>
___________________________________
fastcgi-developers mailing list
http://fastcgi.com/fastcgi-developers/