Slow reads on MongoDB from Spark - weird task allocation

Question

I have a MongoDB 4.2 cluster with 15 shards; the database stores a sharded collection of 6GB (i.e., about 400MB per machine).

I\'m trying to read the whole collection fro