CODEWITHSUNDEEP
pythonray
Disenchanted
Disenchanted

Reputation: 613

How to stop ray from running multiple tasks on the same cluster node

I have a ray cluster that is started manually on several nodes using ray start. How can I schedule tasks to run on the cluster, such that they are exclusive, i.e., no tasks are ran in parallel on one node?

One option would be to specify each node as having only 1 CPU. Another would be to introduce a custom resource 'node', with 1 instance per node.

But this seems like a common scenario, is their a cleaner way to handle this?

Upvotes: 0

Views: 996

Answers (2)

Sang
Sang

Reputation: 925

As ^ said, you can use the custom resources. For example,

In the terminal,

# Head
ray start --head --resources="{<name_of_resources>: 1.0}"
# Worker
ray start --resources="{<name_of_resources>: 1.0}"

In the ray driver (main Python entrypoint that calls ray.init),

ray.init("auto")
@ray.remote(resources={<name_of_resources>: 1})
def ...

Upvotes: 1

Yard1
Yard1

Reputation: 53

Using custom resources is the way to go right now.

Upvotes: 1

Related Questions