Home > Storage > ObjectScale and ECS > Product Documentation > ECS with HAProxy Load Balancer > NFS Configuration Definitions
Note: The definitions in this section to handle NFS requests will only work with HAProxy version 1.7.10 or later.
The HAProxy definitions for NFS discussed in HAProxy configuration for single setup has been further simplified by combining all frontend and backend definitions for each port into a single handler section as shown in Figure 54. In this example, it is not required for the port to be specified in the server line. If there is no port defined on the server line, HAProxy will automatically forward the traffic to the same port that the request came in. Also, stick tables are implemented in this example. By using stick tables, a client is “sticked” to a server such that requests are redirected to the same backend server, unless the server is down or the entry expires in the stick table. For NFS requests, this stickiness is important since ECS utilizes server-side metadata caching and the cache exists locally on each node. Stick tables stores the source IP address in memory. In this example, the size of this table is 10k.
For redundancy and reload safe purposes, a “peers” section is defined to synchronize the content of the stick table between multiple HAProxy instances. In this example only one HAProxy instance is being used where the IP address is the IP of the HAProxy host and the peer name is actual hostname of server running HAProxy instance which is “glb”. The port of the peer can be any number as long as firewalls are open for this port.
For more information on stick tables, refer to this blog by Baptiste, https://www.haproxy.com/blog/client-ip- persistence-or-source-ip-hash-load-balancing/.