redis sentinel is aimed at providing failover: elect new master node and provide it to a client. Why slave nodes are unable to elect by themselves ?(through consensus protocol)
Clustered redis nodes, in contrast, can elect new master. Why don't sentinel nodes do the election in clustered redis?
It's just because our manual failover API call eventually is directed to sentinels by internal algorithms. It's quite natural for sentinel to do so as sentinel has responsibilities to monitor nodes.
However, as you wonder, some people try to make changes so that we can deal with it like cluster (ask replicas instead of sentinels). For example, this is an issue in valkey project. https://github.com/redis/redis/issues/13118