keep-alive message flood.

[Ian]

Hi all,

I am running an application that uses the nginx_http_push_module-0.712 
push module.

The relevant set up is
server {
     listen   443  default ssl;
     ## SSL Certs
     ssl on;
     ssl_certificate /ssl_keys/coachmaster.co.uk.crt;
     ssl_certificate_key /ssl_keys/coachmaster.co.uk.key;
     ssl_ciphers HIGH:!ADH:!MD5;
     ssl_prefer_server_ciphers on;
     ssl_protocols TLSv1;
     ssl_session_cache shared:SSL:1m;
     ssl_session_timeout 5m;
     #
     server_name example.co.uk www.example.co.uk;
     root /var/www/example.co.uk/htsecure;
     access_log /var/www/example.co.uk/access.log;
     index index.php;
     #
     # serve php via fastcgi if it exists
     location ~ \.php$ {
         # try_files $uri =404;
         include /etc/nginx/fastcgi_params;
         fastcgi_pass 127.0.0.1:9000;
         fastcgi_param  SCRIPT_FILENAME $document_root$fastcgi_script_name;
         fastcgi_param  CENTRAL_ROOT       $document_root;
         fastcgi_param  RESELLER_ROOT      $document_root;
         fastcgi_param  ENVIRONMENT        production;
         fastcgi_param  HTTPS ON;
     }
     # serve static files
     try_files $uri $uri/ /index.php ;
     expires 30m;
     # set up publish/subscribe
     push_store_messages on;
     location /publish {
         push_publisher;
         set $push_channel_id $arg_id;
         push_message_timeout 10s;
         push_max_message_buffer_length 30;
     }
     location /activity {
         push_subscriber;
         push_subscriber_concurrency broadcast;
         set $push_channel_id $arg_id;
         default_type text/plain;
     }
}

The trouble is that this has been working fine for about a year, but a 
new user is reporting problems. Everyone else continues to have no 
problems.

He is reporting long delays in both IE11 and Chrome. He is in Germany
behind a corporate proxy/firewall. When he tries it from home, all works 
OK, so the proxy/firewall is prime suspect.

For political reasons (quoting security - ha!) the proxy cannot be
altered.

Looking at the server logs from when he was having problems, he is 
sending a lot of AJAX requests to the /activity URL, and receiving what 
appear to be empty replies (with reply code 200 OK). These rattle 
through at top speed.

After anything from 10 to 200 such exchanges, he gets a larger message
(also with 200 OK) that is so delayed it crashes the application with
a "Missing messages" report.

The jquery call his code it making, already has nocache, so the URL 
contains a "&_=<uniqueNo> parameter to break any caching.

Oddly when I added a similar tweak to break the cache, I managed to 
trigger the same behaviour. There was no  proxy in my setup, and I have 
disabled browser caching. My round trip was 6ms, so I did not see long 
enough delays to crash things.  However I could not find out what was 
causing the problem so I took out my tweak. It only went in, because
I suspected the nocache was being ignored.

Ideas anyone? I'm stumped.

Anyone understand how a proxy could mess things up for Nginx? How
can I prove its the proxy? Crucially how can I compensate?

Thanks
Ian

-- 
Ian Hobson
Mid Auchentiber, Auchentiber, Kilwinning, North Ayrshire KA13 7RR
Tel: 0203 287 1392
Preparing eBooks for Kindle and ePub formats to give the best reader 
experience.