You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This is to improve Tune usability, namely we should not just let user be stuck in PENDING state. The current PlacementGroup API is very reactive. Tune doesn't know if a certain request is ever gonna be fulfilled or not. Since autoscaler has access to cluster.yaml and has an overview understanding of the resource requests, ideally it should signal if a request cannot be met at all.
Btw there was some discussion around this in #14908 but it's worth revisiting if requirements have changed. I suspect this will be somewhat involved though.
This is to improve Tune usability, namely we should not just let user be stuck in PENDING state. The current PlacementGroup API is very reactive. Tune doesn't know if a certain request is ever gonna be fulfilled or not. Since autoscaler has access to cluster.yaml and has an overview understanding of the resource requests, ideally it should signal if a request cannot be met at all.
For context, see discussions under
https://github.com/ray-project/ray/pull/17533/files/d6150e7f8b3a3b2b64cc056abf45f8e6642e4056#diff-4ed0b87c19cc0a24e8ab5747e611efa147b689312489ad4acfe604d05f17e7ac
which resource is not available, and how much is being requested
what is the total amount of those resource available on the cluster
The text was updated successfully, but these errors were encountered: