Currently, with multi raft feature ON, a data node can participate in multiple pipelines. But, same set of dns can be part of multiple pipelines . This leads to 2 problems:
1) In case of let's say 5 datanodes, as and when 3 datanodes register initially with multi-raft feature ON, those 3 datanodes will pair among themselves to create two pipelines thereby satisfying the default pipeline limit of 2 per dn. Next datanodes even after registering cannot take part in writes bcoz it needs a 6th datanode to come up to form the nest tri-set , thereby remain unutilized.
2) With same set of Datanodes being part of multiple plpelines, if one datanode fails, the source for container re-replication will be only the other set of 2 nodes and can beocme a bottleneck.
3) This uneven distribution of pipelines among dns will lead to load distribution getting uneven as well.