Daft icon indicating copy to clipboard operation
Daft copied to clipboard

Account for task crashes during writes on append mode

Open colin-ho opened this issue 11 months ago • 0 comments

Describe the bug If a task crashes during a write on append mode, it will restart and write all the files again, leaving behind dirty files.

To Reproduce Steps to reproduce the behavior: Go to Daft/daft/table/table_io.py and manually insert a probabilistic crash, e.g.

if np.random.random() < failure_probability:
        os._exit(0)

run a write task on ray runner.

To Fix We should implement a mechanism to clear dirty files from possible crashes during writes on append mode.

colin-ho avatar Mar 08 '24 21:03 colin-ho