ray/doc/source/data/dataset-spill.svg
Eric Liang 460cf86858
Split blocks automatically into 500MB chunks on file read and transformation (#20235)
This PR adds support for automatic block splitting on read and map transforms, to keep block size bounded to ~500MiB. This avoids potential OOM situations where a map task may consume too much intermediate Python heap memory, or too much object store shared memory for one block.
2021-11-15 22:25:11 -08:00

1 line
No EOL
45 KiB
XML