| Author |
|
glsmith
Moderator
Joined: 16 Oct 2007
Posts: 2268
Location: Sun Diego, USA
|
 Posted: Sat 08 Feb '14 19:53 Post subject: |
 |
|
That binary you have is SNAP-0910052141 sort of. It's weird in that the actual snap's changes are just 1 change shy of that module. The one extra change in that binary you have is;
*) Update to compile against httpd-2.4.3. [Peter Danenberg]
yet at the top of the changes file that module was built with it says mod_fastcgi-SNAP-0910052141.
Anyhow, that tells me the SNAP-0910052141 and earlier probably will not compile against Apache 2.4 so I doubt anyone is going to compile them for you. |
|
| Back to top |
|
jimski
Joined: 18 Jan 2014
Posts: 196
Location: USSA
|
| Posted: Sun 09 Feb '14 2:12 Post subject: |
|
|
| glsmith wrote: | | That binary you have is SNAP-0910052141 sort of. |
I'm looking for precompiled binaries for Apache 2.2
(Win32) mod_fastcgi-SNAP-0811090952 binary
(Win32) mod_fastcgi-SNAP-0910052141 binary
If you have them please post the a link. Thank you. |
|
| Back to top |
|
jimski
Joined: 18 Jan 2014
Posts: 196
Location: USSA
|
| Posted: Sun 09 Feb '14 2:33 Post subject: |
|
|
This will surprise you
After downgrading from php 5.4.24 to php 5.3.8
All the failed requests under mod_FCGID went away but the overall performance degraded. I used the same php.ini settings
Here are the latest results.
Test file phpinfo.php
Test using mod_FCGID Apache 2.4 php 5.3.8 ts
10 processes
Time taken for tests: 13.1 sec.
Completed requests: 10000
Failed requests: 0
Requests per second: 768
Test using mod_FCGID Apache 2.4 php 5.3.8 nts
10 processes
Time taken for tests: 9.1 sec.
Completed requests: 10000
Failed requests: 0
Requests per second: 1109
Test using mod_FastCGI Apache 2.4 php 5.3.8 nts
10 processes
Time taken for tests: 3.35 sec.
Completed requests: 10000
Failed requests: 1099
Requests per second: 2976
Although mod_FastCGI stalls and generates failed requests it is clearly superior when it comes to speed of processing concurrent requests. Unfortunately is not usable for production.
The premature exiting of php-cgi.exe processes seem to be related to a limit set at compilation stage and so far I couldn't find directive to override it. Looks like each process last about 500 requests and then exits. If this issue is fixed this will be one hell of solution for php on windows which will beat the pants off of FCGID.
Last edited by jimski on Sun 09 Feb '14 21:26; edited 2 times in total |
|
| Back to top |
|
Steffen
Moderator
Joined: 15 Oct 2005
Posts: 3057
Location: Hilversum, NL, EU
|
|
| Back to top |
|
glsmith
Moderator
Joined: 16 Oct 2007
Posts: 2268
Location: Sun Diego, USA
|
| Posted: Sun 09 Feb '14 12:51 Post subject: |
|
|
| jimski wrote: |
I'm looking for precompiled binaries for Apache 2.2
|
You did not specify 2.2 earlier, anyhow, VC9 compiled against ApacheHaus' 2.2.26. Modules are in the Release folders.
Moderator Edit: Remove stale link
Last edited by glsmith on Fri 02 Jan '15 3:49; edited 2 times in total |
|
| Back to top |
|
jimski
Joined: 18 Jan 2014
Posts: 196
Location: USSA
|
| Posted: Sun 09 Feb '14 20:01 Post subject: |
|
|
'
Steffen, if this is the case then how do you override 500 requests limit with mod_fastcgi on windows.
There is no FCGI_MAX_REQUESTS directive, or anything similar, in the specifications for mod_FastCGI http://www.fastcgi.com/mod_fastcgi/docs/mod_fastcgi.html.
The mod_FastCGI supposed to keep the process alive by default and only destroy it as needed using -killInterval or other process control directive.
Here is what Lighty writes about it:
FastCGI removes this limitation by keeping the process running and handling the requests by this always running process. This removes the time used for the fork() and the overall startup and cleanup time which is necessary to create and destroy a process. http://redmine.lighttpd.net/projects/1/wiki/Docs_ModFastCGI
| glsmith wrote: |
You did not specify 2.2 earlier, anyhow, VC9 compiled against ApacheHaus' 2.2.26. Modules are in the Release folders.
|
Thank you glsmith, I will give it a spin and if it works I will post results.
Moderator Edit: Remove stale link |
|
| Back to top |
|
CamaroSS
Joined: 24 Jan 2013
Posts: 78
Location: RF, Tver
|
| Posted: Mon 10 Feb '14 12:55 Post subject: |
|
|
I think
| Code: | | FastCgiServer -initial-env PHP_FCGI_MAX_REQUESTS=0 |
should fix it. It's just another environment variable. |
|
| Back to top |
|
Steffen
Moderator
Joined: 15 Oct 2005
Posts: 3057
Location: Hilversum, NL, EU
|
| Posted: Mon 10 Feb '14 13:45 Post subject: |
|
|
Also advised to set env: FcgidInitialEnv PHP_FCGI_CHILDREN 0
For mod_fcgid from the docu:
PHP child process management (PHP_FCGI_CHILDREN) should always be disabled with mod_fcgid, which will only route one request at a time to application processes it has spawned; thus, any child processes created by PHP will not be used effectively. (Additionally, the PHP child processes may not be terminated properly.) By default, and with the environment variable setting PHP_FCGI_CHILDREN=0, PHP child process management is disabled. |
|
| Back to top |
|
CamaroSS
Joined: 24 Jan 2013
Posts: 78
Location: RF, Tver
|
| Posted: Mon 10 Feb '14 14:55 Post subject: |
|
|
| I don't get it. The last time I've checked, PHP for Windows couldn't fork and therefore couldn't have child processes, so this variable has no effect. Additionally, a single php-fcgi process couldn't handle more than one request at a time. Did anything change? |
|
| Back to top |
|
jimski
Joined: 18 Jan 2014
Posts: 196
Location: USSA
|
| Posted: Mon 10 Feb '14 17:03 Post subject: |
|
|
Thank you CamaroSS and Steffen I will try it.
| CamaroSS wrote: | | I don't get it. The last time I've checked, PHP for Windows couldn't fork and therefore couldn't have child processes, so this variable has no effect. Additionally, a single php-fcgi process couldn't handle more than one request at a time. Did anything change? |
Windows server versions can fork with Interix subsystem which is build in by default in windows server 2003R2 and 2008 http://en.wikipedia.org/wiki/Interix . But CamarroSS is right that forking is not used by php for windows. There is however a fastcgi.dll that should spawn a new thread for each requests (at least that's what I've been told)
Here is cryptic and convoluted tutorial about how to use this dll http://forum.lazarus.freepascal.org/index.php?topic=13312.0
Using mod_FastCGI on windows is definitely poorly documented and not easy to figure out. |
|
| Back to top |
|
C0nw0nk
Joined: 07 Oct 2013
Posts: 241
Location: United Kingdom, London
|
| Posted: Mon 10 Feb '14 19:53 Post subject: |
|
|
| CamaroSS wrote: |
| Code: | | FastCgiServer -initial-env PHP_FCGI_MAX_REQUESTS=0 |
should fix it. It's just another environment variable. |
Does that actualy work if you already have this set ?
| Code: | FcgidInitialEnv PHP_FCGI_MAX_REQUESTS 0
FcgidInitialEnv PHP_FCGI_CHILDREN 0 |
Or do they both do seperate things ? |
|
| Back to top |
|
jimski
Joined: 18 Jan 2014
Posts: 196
Location: USSA
|
| Posted: Mon 10 Feb '14 20:13 Post subject: |
|
|
| C0nw0nk wrote: |
Does that actualy work if you already have this set ?
Or do they both do seperate things ? |
CamaroSS was referring to mod_fastcgi and you are using mod_fcgid, they have different directives and CamaroSS suggestion will not work with mod_fcgid.
Meanwhile Steffen was referring to mod_fcgid while I was asking about mod_fastcgi.
We have some miscommunication going on here.  |
|
| Back to top |
|
C0nw0nk
Joined: 07 Oct 2013
Posts: 241
Location: United Kingdom, London
|
|
| Back to top |
|
jimski
Joined: 18 Jan 2014
Posts: 196
Location: USSA
|
| Posted: Tue 11 Feb '14 8:59 Post subject: |
|
|
New status on mod_FastCGI
The suggestion of CamaroSS worked and the processes are not dying anymore. However the errors still persist with phpinfo.php although at much lower rate.
Here are the latest tests performed on Windows Server 2003R2 x64
Test file phpinfo.php
Test using mod_FastCGI Apache 2.4.4 x64 php 5.3.22 x64 nts
10 processes
Time taken for tests: 3.61 sec.
Completed requests: 10000
Failed requests: 996
Requests per second: 2759
Randomly this error appears when pointing firefox to localhost/phpinfo.php The entire phpinfo page is being displayed but the browser says "waiting for localhost" and after 30 seconds this appears in the error log.
FastCGI: comm with server "C:/Server/Apache24/www/php-cgi.exe" aborted: idle timeout (30 sec), referer: http://localhost/phpinfo.php
FastCGI: incomplete headers (0 bytes) received from server "E:/Server/Apache24/www/php-cgi.exe", referer: http://localhost/phpinfo.php
Last edited by jimski on Tue 11 Feb '14 11:58; edited 1 time in total |
|
| Back to top |
|
James Blond
Moderator
Joined: 19 Jan 2006
Posts: 7294
Location: Germany, Next to Hamburg
|
| Posted: Tue 11 Feb '14 11:51 Post subject: |
|
|
can you please try to adjust the values for mod_fcgid
| Code: |
<IfModule fcgid_module>
FcgidConnectTimeout 10
FcgidMaxProcesses 300
FcgidMaxProcessesPerClass 300
FcgidOutputBufferSize 64
ProcessLifeTime 0
FcgidMaxRequestsPerProcess 0
FcgidMinProcessesPerClass 0
FcgidFixPathinfo 0
FcgidProcessLifeTime 0
FcgidZombieScanInterval 20
FcgidMaxRequestLen 536870912
FcgidIOTimeout 120
FcgidTimeScore 3
</IfModule>
|
Give it a shot of love <3
-- edit 1 --
for mod_fcgid
Last edited by James Blond on Tue 11 Feb '14 12:30; edited 1 time in total |
|
| Back to top |
|
jimski
Joined: 18 Jan 2014
Posts: 196
Location: USSA
|
| Posted: Tue 11 Feb '14 11:55 Post subject: |
|
|
James, My apologies for the typo. I meant mod_FastCGI and not mod_FCGID.
If you could kindly adjust your post to comment on FastCGI it would be great.
I thought that errors may be related to a buffer and I added -flush directive to send unbuffered data but this didn't fix the errors.
I also increased memory in php but also with no result.
I don't see a directive to adjust the buffer sidze in FastCGI docs.
Last edited by jimski on Tue 11 Feb '14 12:38; edited 1 time in total |
|
| Back to top |
|
James Blond
Moderator
Joined: 19 Jan 2006
Posts: 7294
Location: Germany, Next to Hamburg
|
|
| Back to top |
|
jimski
Joined: 18 Jan 2014
Posts: 196
Location: USSA
|
| Posted: Tue 11 Feb '14 12:42 Post subject: |
|
|
I have -appConnTimeout set to 30 seconds that's why the error is reported after 30 seconds. This is plenty of time to execute almost any script.
Here are my latest settings:
| Code: | FastCgiConfig -initial-env PATH -idle-timeout 400 -killInterval 3600 -maxClassProcesses 10 -maxProcesses 10 -startDelay 1 -minProcesses 10
FastCgiServer "C:/Server/Apache24/www/php-cgi.exe" -initial-env PATH -initial-env PHP_FCGI_MAX_REQUESTS=0 -initial-env PHP_FCGI_CHILDREN=0 -appConnTimeout 30 -processes 10 -init-start-delay 1 -restart-delay 1 -min-server-life 6000 -listen-queue-depth 10000 -flush |
Last edited by jimski on Tue 11 Feb '14 15:45; edited 3 times in total |
|
| Back to top |
|
jimski
Joined: 18 Jan 2014
Posts: 196
Location: USSA
|
| Posted: Tue 11 Feb '14 13:09 Post subject: |
|
|
I think we are getting closer to finding the problem.
1. If I specify -processes 1 , then there is 0 Failed Requests
2. If I specify -processes 3, then there is 33% failed requests
3. If I specify -processes 10, then there is 10% failed requests
4. If I specify -processes 100 , then there is 1% failed requests
5. After the server is shut down there is always one orphan zombie php-cgi process lingering around that I have to shut down manually.
So the problem is that if more than one process is running then out of all the running processes there is always one which is dead and it is causing failed requests at a rate described by this equation:
failed_requests = total_requests * (1 / number_of_processes )
The more processes are running, the lower the error rate.
Last edited by jimski on Tue 11 Feb '14 19:51; edited 2 times in total |
|
| Back to top |
|
James Blond
Moderator
Joined: 19 Jan 2006
Posts: 7294
Location: Germany, Next to Hamburg
|
| Posted: Tue 11 Feb '14 15:40 Post subject: |
|
|
Do I get it right that if you define -processes 1 you have one single instance of php-cgi.exe running in your task manager?
What does happen if that instances crashes? Does fastcgi start a new one? Do requests fail it that case? |
|
| Back to top |
|
