On 5/20/2015 9:59 AM, Alex wrote: > Do I need rpcbind runing on haproxy , ? we are trying to use the haproxy to > load balance across GlusterFS for NFS using mode tcp > but i get connection refused port 111 .
NFS as a protocol is fairly complicated, like passive FTP in some ways. An initial connection goes to RPC, the client is informed about what port the requested service is using, and then a new connection is made. Making a protocol like that work with haproxy is challenging. It's possible, but not completely straightforward. I have a gluster install. We are phasing it out, but we had a similar need -- high availability. What I did to achieve this was install ucarp on a pair of the gluster peers. If the active machine fails, the shared IP address migrates to the other one, where NFS is still available. I have separate monitoring to make sure that the NFS process is active on each of those servers. Here's something else advising a similar approach: http://unix.stackexchange.com/a/77102 Thanks, Shawn
