Ray is a unified framework for scaling AI and Python applications. Ray consists of a core distributed runtime and a set of AI Libraries for accelerating ML workloads.
[Data] Pending object memory store usage in the progress bar should incorporate estimated size of output blocks (obj_store_mem_pending_task_outputs) #47125
What happened + What you expected to happen
Pending resource usage is introduced to the progress bar in #46729 -- number of pending actors scheduled on CPU and GPU are now displayed in progress bar. We should also include
obj_store_mem_pending_task_outputs
as the pending object store usage. https://github.com/ray-project/ray/blob/82af058f3456dd7088b94a8d29a95450f73d215b/python/ray/data/_internal/execution/interfaces/op_runtime_metrics.py#L338Versions / Dependencies
ray master
Reproduction script
Use the following script to spin up actors to be executed on GPUs.
Issue Severity
Low: It annoys or frustrates me.