haproxy issue with request headers

When doing several attempts to login to erp5 at the same time (20 users or more), haproxy corrupts around 20% of the requests bad-inserting the http-send-name-header key-value. Changing the way this information is passed in the request avoids the issue.

stack/erp5/haproxy.cfg.in

{%   for address, connection_count, webdav in backend_list -%}
{%     if webdav %}{% do has_webdav.append(None) %}{% endif -%}
{%     set server_name = name ~ '-' ~ loop.index0 -%}
+ http-request set-header X-Balancer-Current-Server {{ server_name }}

[Sebastien Robin]

This is going to generate such configuration :

listen user
  bind 10.0.15.62:2152
  http-request set-header X-Balancer-Current-Cookie SERVERID
  http-request set-header X-Balancer-Current-Server user-0
  server user-0 10.0.5.183:2200 cookie user-0 check inter 3s rise 1 fall 2 maxqueue 5 maxconn 1
  http-request set-header X-Balancer-Current-Server user-1
  server user-1 10.0.5.183:2201 cookie user-1 check inter 3s rise 1 fall 2 maxqueue 5 maxconn 1

haproxy.cfg is a configuration file, it's not a script that would be executed line by line. Thus there is good chance that your change works only in the case we have a single server in the backend...

[Cédric Le Ninivin]

I agree, I had wrongly read.

[Roque]

I see. So, this doesn't have the same behavior of "http-send-name-header X-Balancer-Current-Server". I'll investigate if there is a another way to pass the server name using a different keyword in the default part.

[Sebastien Robin]

Roque, only good way is to go to discuss with haproxy guys