[Data] num_rows_per_file doesn't work with small values

[bveeramani]

What happened + What you expected to happen

My dataset contains 28k rows. I tried writing this to Parquet with num_rows_per_file=700. I expected several Parquet files each with 700 rows, but instead got a single file with all the rows.

Versions / Dependencies

4d37e55

Reproduction script

import os

import ray

ray.data.range(100, override_num_blocks=1).write_parquet(
    "/tmp/results", num_rows_per_file=10
)


print(os.listdir("/tmp/results"))

Issue Severity

Medium: It is a significant difficulty but I can work around it.