As far as I remember, mostly Flume (open sourced as Apache Beam and also known as Cloud Dataflow on GCP).

Flume/Beam still provides map/reduce as operations, mind you, but with a much richer processing model.