I can see dask task stream when I use dask.compute in dask.distributed.Client
so how can I see task stream when I use ray_dask_get(dsk_map, task_list)
here is some examples:
from dask.distributed import Client
from ray.util.dask import ray_dask_get
- dask.compute(delayed_dict) I can see task streams and progress.
- client.get(custom_graph_dict, list) I can see task streams too, buy it does not use ray.
3 .ray_dask_get(custom_graph_dict, list) I can see nothing from dask status web
so what can I do for example 3 to see task streams?
@Asuka Hi and welcome to Discourse!
Dask Distributed is a Dask Scheduler, which has the diagostics dashboard (where you’re seeing the task stream and progress bar) built-in.
Dask-on-Ray is a different, alternative “scheduler” you can use, which doesn’t provide a dashboard.
In your question, examples 1 and 2 are using the Dask Distributed Scheduler. Whereas example 3 is using the Dask-on-Ray scheduler, that’s why you’re not seeing activity in the Dask Dashboard. Does this help answer your question?
Side note: I also noticed the dask-on-ray docs have a note that recommends you to not use the dask.distributed client.
Thank you for your answer. It helps a lot.
I want to know is there any daskboard or something like tools for dask on ray which I can see task streams, progress,left tasks in it(like what dask distributed dashboard shows) .
@Asuka I’m not an expert, but the dask-on-ray docs mention custom callbacks for diagnostics, maybe you can check that out? I also think the Ray community, who developed this scheduler, might be able help you better with this question.