"Hadoop" is never going to die. Even if it fades into the background, some vestigial remnant within Spark will use a HDFS wrapper to access S3 or something like that.

But MapReduce will eventually, finally, kick the can.