Closed amitschang closed 9 months ago
All modified and coverable lines are covered by tests :white_check_mark:
Comparison is base (
63df416
) 100.00% compared to head (e41c5be
) 100.00%.
:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.
@xiangchenjhu , I will update the comment, but these are just what I deemed reasonable values to balance catching real issues and giving enough chances to flush the data. Hasn't seemed to be a real issue with linux, for whatever reason mac may be faster/slower such that it pops up as an issue. If this causes issues later, we can attempt to be more deterministic by shutting ray down and passing files, but that seems overly complex for the present case.
@amitschang Got it, thanks for the insight. It's intriguing to observe the differing performance between Linux and Mac in this scenario.
Meant to fix https://github.com/ssec-jhu/dplutils/issues/27. Ray doesn't seem to provide an API to wait for all pending tasks on an actor, so we give some time and triggers to flush out the system.