The current implementation of the new spliterator was designed with minimal memory consumption in mind, as the ResultSet could be quite large. In YDB version 24.1, settings for response size for readTable have been introduced, so the current spliterator can be improved:

Idea:

• We read the ResultSet from the database and feed it into a Stream. While the Stream processes the ResultSet, we simultaneously read a new ResultSet so that by the time the first one is processed, the second one is already prepared.
• The Stream is almost lock-free. Locks taken only to fetch the ResultSet from the queue.
• The YDB-thread reads the ResultSet and places it in the queue. AFTER appending, if the queue is full, the YDB-thread sleeps until something takes the ResultSet from the queue. This ensures that at any given moment, no more than N + 1 ResultSets are in memory, where N is the size of the queue.
• If the element processing operation from the Stream is heavy, the process can be parallelized. YdbSpliteratorQueue supports fetching the ResultSet from multiple threads.