nginx + pagespeed + php-fpm doesn't work after second request

[frutik]

Hi guys.

I have strange issue when trying to setup pagespeed

stock nginx + stock php-fpm works as expected nginx built from source (without pagespeed) + stock php-fpm works as expected nginx built from source with pagespeed + stock php-fpm - first 2 requests works as expected, after that all next responses contains full php source (instead of html) until restart of nginx

root@dev:/usr/local/nginx/conf# lsb_release -a No LSB modules are available. Distributor ID: Ubuntu Description: Ubuntu 12.10 Release: 12.10 Codename: quantal

nginx-1.4.0.tar.gz release-1.5.27.1-beta.zip stock php-fpm

http { include /usr/local/nginx/conf/mime.types; default_type application/octet-stream;

root           /usr/local/nginx/html;

sendfile        on;
keepalive_timeout  65;

server {
    listen       80;
pagespeed off;

location / {
    try_files $uri $uri/ /index.php$is_args$args;
}

    location ~ \.php$ {
        fastcgi_pass   unix:/var/run/php5-fpm.sock;
        fastcgi_index  index.php;
    fastcgi_param  SCRIPT_FILENAME  $document_root$fastcgi_script_name;
        include        /usr/local/nginx/conf/fastcgi_params;
    }
}

[jeffkaufman]

Yikes; that shouldn't happen. Quickly testing now with php-cgi I'm not seeing this:

php-cgi -b 127.0.0.1:9000

location ~ \.php$ {
  fastcgi_param SCRIPT_FILENAME $request_filename;
  fastcgi_param QUERY_STRING $query_string;
  fastcgi_param REQUEST_METHOD $request_method;
  fastcgi_param CONTENT_TYPE $content_type;
  fastcgi_param CONTENT_LENGTH $content_length;
  fastcgi_pass 127.0.0.1:9000;
}

curl localhost:8050/mod_pagespeed_test/php_withoutflush.php
curl localhost:8050/mod_pagespeed_test/php_withoutflush.php
curl localhost:8050/mod_pagespeed_test/php_withoutflush.php
curl localhost:8050/mod_pagespeed_test/php_withoutflush.php
curl localhost:8050/mod_pagespeed_test/php_withoutflush.php

I'm going to try to reproduce your setup fully to figure out what's going on.

[jeffkaufman]

I set up a test [1] that's much closer to your setup: I copied your config, used php5-fpm on a unix socket, but I'm still not seeing this.

When you restarted nginx, this happened again?

[1] https://github.com/pagespeed/ngx_pagespeed/tree/jefftk-test-php-more

[frutik]

Hi, Jeff.

Do you need any additional configs etc? Perhaps I can try to debug something on my own to help you?

[frutik]

yes. every time after restart of nginx it serves successfully only two first requests after that sends source of php file

[frutik]

I'll try to make screencast

[jeffkaufman]

Is there anything in the error log? Is there anything unusual about the php url in question?

[jeffkaufman]

(You might need to set your error log to "debug" in order to see relevant errors.)

[jeffkaufman]

If I'm reading it right it's printing:

<?php
    phpinfo();
>

This is the content of index.php on your system?

[frutik]

yep. nothing interesting in logs, btw

[jeffkaufman]

Trying to reproduce this more, still failing.

Here's the nginx.conf I'm using:

worker_processes  1;

events {
  worker_connections  1024;
}

error_log /home/jefftk/php-error.log;

http {
  access_log /home/jefftk/php-access.log;

  include /home/jefftk/mime.types;
  default_type application/octet-stream;

  root /home/jefftk/nginx/html;

  sendfile        on;
  keepalive_timeout  65;

  server {
    listen       8090;
    pagespeed off;

    location / {
      try_files $uri $uri/ /index.php$is_args$args;
    }

    location ~ \.php$ {
      fastcgi_pass   unix:/var/run/php5-fpm.sock;
      fastcgi_index  index.php;
      fastcgi_param  SCRIPT_FILENAME  $document_root$fastcgi_script_name;
      include /home/jefftk/fastcgi_params;
    }
  }
}

My fastcgi params are the 1.4.0 default:

fastcgi_param  QUERY_STRING       $query_string;
fastcgi_param  REQUEST_METHOD     $request_method;
fastcgi_param  CONTENT_TYPE       $content_type;
fastcgi_param  CONTENT_LENGTH     $content_length;

fastcgi_param  SCRIPT_NAME        $fastcgi_script_name;
fastcgi_param  REQUEST_URI        $request_uri;
fastcgi_param  DOCUMENT_URI       $document_uri;
fastcgi_param  DOCUMENT_ROOT      $document_root;
fastcgi_param  SERVER_PROTOCOL    $server_protocol;
fastcgi_param  HTTPS              $https if_not_empty;

fastcgi_param  GATEWAY_INTERFACE  CGI/1.1;
fastcgi_param  SERVER_SOFTWARE    nginx/$nginx_version;

fastcgi_param  REMOTE_ADDR        $remote_addr;
fastcgi_param  REMOTE_PORT        $remote_port;
fastcgi_param  SERVER_ADDR        $server_addr;
fastcgi_param  SERVER_PORT        $server_port;
fastcgi_param  SERVER_NAME        $server_name;

# PHP only, required if PHP was built with --enable-force-cgi-redirect
fastcgi_param  REDIRECT_STATUS    200;

My index.php is:

<?php
    phpinfo();
?>

Running:

curl localhost:8090/index.php

repeatedly I just get the output of phpinfo().

[jeffkaufman]

Remaining differences between our setups:

• you're running as root on 80, I'm as me on 8090
• you're fetching from a different computer
• you're on 12.10, I'm on 12.04

Is there anything you're including in your ./configure that I wouldn't be?

[frutik]

Sorry, I'll be back in a few hours

[eleshar]

I'm experiencing the same issue on the 1.5.27.1-beta release, if I switch to the alpha-r2338 branch it works.

[oschaaf]

I tried to reproduce, but failed using nginx1.4.0 on ubuntu12.04.2 (x86_64). Both debug and release configured with either pagespeed on; and pagespeed off; don't reproduce this for me. The try_files statement doesn't work for me, but if I remove $uri/ it does. The try_files path also does not reproduce the problem for me either.

[frutik]

I am trying to create vagrant/puppet stuff to have reproducible environment which could help reproduce issue

[jeffkaufman]

if I switch to the alpha-r2338 branch it works

Did you try any of the others? In particular, alpha-2618 or alpha-2748? That would be helpful for trying to narrow down when this started.

[eleshar]

I only checked that one since i recall it was working.

I've tested with alpha-2618 and alpha-2748 both working for me.

[jeffkaufman]

@eleshar thanks for checking them!

alpha-2748 wasn't very long ago. The only two significant changes I see since then are:

• we're built off of mod_pagespeed 1.5.27.1 instead of mod_pagespeed trunk@2748
• @chaizhenhua's #241, which switched us to using a single pipe to communicate with nginx

[jeffkaufman]

@frutik

vagrant/puppet stuff to have reproducible environment

That would be great.

[jeffkaufman]

For now people should use alpha-2748 (and the install instructions say to). Master is updated to fix this, and 1.5.27.2 which I'm trying to get out soon will have this fixed.