Another question on Daft... When running large jo...
# general
y
Another question on Daft... When running large jobs on ray, I am seeing that the denominator keeps moving:
Copy code
ScanWithTask-Aggregate [Stage:3]:  94%|█████████▍| 3246/3437 [30:25<01:20,  2.38it/s]
ScanWithTask-Aggregate [Stage:3]:  94%|█████████▍| 3250/3441 [30:26<01:11,  2.68it/s]
ScanWithTask-Aggregate [Stage:3]:  94%|█████████▍| 3255/3446 [30:27<01:03,  3.03it/s]
ScanWithTask-Aggregate [Stage:3]:  94%|█████████▍| 3259/3450 [30:30<01:23,  2.28it/s]
ScanWithTask-Aggregate [Stage:3]:  94%|█████████▍| 3265/3456 [30:31<01:07,  2.83it/s]
ScanWithTask-Aggregate [Stage:3]:  94%|█████████▍| 3269/3460 [30:33<01:17,  2.46it/s]
ScanWithTask-Aggregate [Stage:3]:  94%|█████████▍| 3272/3463 [30:35<01:16,  2.50it/s]
ScanWithTask-Aggregate [Stage:3]:  94%|█████████▍| 3277/3468 [30:36<01:05,  2.93it/s]
Is this expected? Any insight as to what is happening under the hood? Is Daft only scheduling a few partitions at a time?
j
Correct! Daft is scheduling your partitions in waves, and sometimes will be able to short-circuit certain executions (e.g. during a limit)
Another quick point of consideration here is that if we fire off all our tasks at once, Ray can get pretty overwhelmed and so we’ve had to build our own scheduling mechanisms here