When using dask, is "loadbalance spillback" still best practice?

as in:

RAY_SCHEDULER_LOADBALANCE_SPILLBACK=1 ray start --head --num-cpus=0

The instructions specifically mention ray 1.3 - is this env setting also still recommended for ray 2.0?

From 2.0, we made a better scheduling policy, and it is likely that you don’t need this option.

good to know, thanks! :+1:

Please let us know if you see any issue (like object store memory usage is not well load balanced).