It is ok for me to use transfer manager in s3storage. But I don't understand why you need haproxy and round-robin properties. Actually, you could define multiple main stages, each has its own auth/storage section, where you could use different host:port.
We have a N node cluster and to balance load across all nodes, haproxy is used. We are trying to do some benchmarking with large objects, 16M, 32M, 64M, and I am trying to eliminate all the intermediate hops between cosbench and my cluster. The change I put in, the Host header is set to whatever was configured in the xml file, but when I make the actual connection, I use the ip addr. Does that make sense? Pls let me know tf there is any other way to bypass haproxy/LB and send requests directly to nodes.