What is a subsized stream?

A stream is sized when its size is known upfront without consuming it — for example, a stream opened on any collection is sized, since you can just call size() on the collection.

A stream is subsized when splitting it produces substreams that are themselves sized. Not every sized stream is subsized: it depends entirely on how the underlying data structure splits.

List gives you subsized streams: if you split a list into two sublists, you know exactly how many elements landed in each one.

Set does not. Splitting a set means splitting its internal array, and there's no guarantee of how the elements land across the two halves. In the worst case, one half could end up with zero elements and the other with all of them — a completely unbalanced split.

Splitting a data source is exactly what parallel streams do internally: they divide the source into chunks and process each chunk on a different thread. If a source can't be split evenly — like a Set — a parallel stream over it can end up with badly unbalanced work, where one thread does almost everything while others sit idle.

java
List<Integer> list = List.of(1, 2, 3, 4, 5, 6);
Set<Integer> set = Set.of(1, 2, 3, 4, 5, 6);

list.stream().spliterator().hasCharacteristics(Spliterator.SUBSIZED); // true
set.stream().spliterator().hasCharacteristics(Spliterator.SUBSIZED);  // false

A List's spliterator reports SUBSIZED; a Set's does not, since its internal array can't be divided predictably.

Don't reach for a parallel stream on a source that isn't easily splittable — a HashSet, for instance, is a poor candidate. And even when a source is subsized, like a List, think twice before parallelizing: the overhead of splitting and coordinating threads can easily outweigh the benefit, and a parallel stream can end up hurting performance rather than improving it.