Re: Minimize startup timeout on dynamic fastcgi

Hi Stephan!

I tried to install latest snapshot php5.1-200604051630 to my debug server:
i386 + Linux Fedocra Core 4 + Apache 2.0.55 + mod_fastcgi-SNAP-0404142202
PHP was compiled with minimal options:
'./configure' '--enable-fastcgi' '--enable-force-cgi-redirect' 
'--with-mysql=shared' '--prefix=/usr/local/php5'
It works and identifies itself as PHP Version 5.1.3RC3-dev.

But there's some additional strange moments:

Here's access_log:
212.176.x.x - - [06/Apr/2006:13:31:53 +0400] "GET /1.php HTTP/1.1" 1 33417
212.176.x.x - - [06/Apr/2006:13:31:54 +0400] "GET 
/1.php?=PHPE9568F35-D428-11d2-A769-00AA001ACF42 HTTP/1.1" 200 2146
212.176.x.x - - [06/Apr/2006:13:33:10 +0400] "GET /1.php HTTP/1.1" 1 33417
212.176.x.x - - [06/Apr/2006:13:33:11 +0400] "GET 
/1.php?=PHPE9568F35-D428-11d2-A769-00AA001ACF42 HTTP/1.1" 200 2146
24.87.52.26 - - [06/Apr/2006:13:37:01 +0400] "GET 
http://proxyking.servehttp.com:8080/pk/service?service=Echo HTTP/1.0" 
404 373
212.176.x.x - - [06/Apr/2006:13:40:09 +0400] "GET /1.php HTTP/1.1" 1 33417
212.176.x.x - - [06/Apr/2006:13:40:10 +0400] "GET 
/1.php?=PHPE9568F35-D428-11d2-A769-00AA001ACF42 HTTP/1.1" 200 2146
212.176.x.x - - [06/Apr/2006:13:40:38 +0400] "GET /1.php HTTP/1.1" 1 33417
212.176.x.x - - [06/Apr/2006:13:40:43 +0400] "GET 
/1.php?=PHPE9568F35-D428-11d2-A769-00AA001ACF42 HTTP/1.1" 200 2146

BTW: whats' the strange HTTP code 1 ?

Here's error_log:

[Thu Apr 06 13:31:50 2006] [notice] SIGHUP received.  Attempting to restart
[Thu Apr 06 13:31:51 2006] [notice] FastCGI: wrapper mechanism enabled 
(wrapper: /usr/local/apache2/bin/suexec)
[Thu Apr 06 13:31:51 2006] [notice] FastCGI: process manager initialized 
(pid 5730)
[Thu Apr 06 13:31:51 2006] [notice] Apache/2.0.55 (Unix) mod_ssl/2.0.55 
OpenSSL/0.9.7f DAV/2 mod_fastcgi/mod_fastcgi-SNAP-0404142202 configured 
-- resuming normal operations
[Thu Apr 06 13:31:53 2006] [warn] FastCGI: (dynamic) server 
"/usr/local/apache2/htdocs/fcgi-bin/php5.fcgi" (uid 504, gid 504) 
started (pid 5736)
[Thu Apr 06 13:31:54 2006] [error] [client 212.176.x.x] (104)Connection 
reset by peer: FastCGI: comm with server 
"/usr/local/apache2/htdocs/fcgi-bin/php5.fcgi" aborted: read failed
[Thu Apr 06 13:33:11 2006] [error] [client 212.176.x.x] (104)Connection 
reset by peer: FastCGI: comm with server 
"/usr/local/apache2/htdocs/fcgi-bin/php5.fcgi" aborted: read failed
[Thu Apr 06 13:37:01 2006] [error] [client 24.87.52.26] File does not 
exist: /usr/local/apache2/htdocs/pk
[Thu Apr 06 13:40:09 2006] [error] [client 212.176.x.x] (104)Connection 
reset by peer: FastCGI: comm with server 
"/usr/local/apache2/htdocs/fcgi-bin/php5.fcgi" aborted: read failed
[Thu Apr 06 13:40:10 2006] [warn] FastCGI: (dynamic) server 
"/usr/local/apache2/htdocs/fcgi-bin/php5.fcgi" (pid 5736) termination 
signaled
[Thu Apr 06 13:40:10 2006] [warn] FastCGI: (dynamic) server 
"/usr/local/apache2/htdocs/fcgi-bin/php5.fcgi" (pid 5736) terminated by 
calling exit with status '0'
[Thu Apr 06 13:40:40 2006] [warn] FastCGI: (dynamic) server 
"/usr/local/apache2/htdocs/fcgi-bin/php5.fcgi" (uid 504, gid 504) 
restarted (pid 5791)
[Thu Apr 06 13:40:42 2006] [error] [client 212.176.x.x] (104)Connection 
reset by peer: FastCGI: comm with server 
"/usr/local/apache2/htdocs/fcgi-bin/php5.fcgi" aborted: read failed
[Thu Apr 06 13:46:52 2006] [warn] FastCGI: (dynamic) server 
"/usr/local/apache2/htdocs/fcgi-bin/php5.fcgi" (pid 5791) termination 
signaled
[Thu Apr 06 13:46:52 2006] [warn] FastCGI: (dynamic) server 
"/usr/local/apache2/htdocs/fcgi-bin/php5.fcgi" (pid 5791) terminated by 
calling exit with status '0'


First run is fine and quick. But after I see error 104 - connection 
reset by peer.
After some time PHP was said to stop and it stopped. But socket resided 
(and have lastmod time 13:31 - i.e. Apache startup).
On next page load @13:40:09 apache uses it. Hmm... just after that php 
was killed - because timout is over. But it lookes like process cound be 
killed during the request! Is it possible to use idle to timeouts rather 
that work timout to secide what must be killed?
Later at 13:40:38 apache tried to use the socket but insuccessful - so 
after timeout it had to restart php.
Very strange moment - why socket resides while peer is killed?

May be we need more fresh version of mod_fastcgi which is aware of peer 
process status or can drop the socket at least?

Currently I'm not sure bugs are because of  the PHP only. It dies as it 
receives signal, but the problem resides.

Thank you.

Yours,
Eugene


>
>
> The patch does not directly apply to php 5.1.2. I am using this patch 
> ontop of 4.4.2.
> If you compare the difference between both version:
> (diff php-5.1.2/sapi/cgi/cgi_main.c php-4.4.2/sapi/cgi/cgi_main.c)
> Then you see that they share the essential code, most of the changes
> seam to be of cosmetic nature.
> If you understand the basics, then you should be able to patch that 
> file manually (untested).
>
> Alternatively, you could also go with the latest snapshot of the 
> 5.1.x.(>5.1.2) The fast_cgi
> code was written from scratch.
>
> Stefan Wendt
>
>
>
>
>

___________________________________
fastcgi-developers mailing list
http://fastcgi.com/fastcgi-developers/