Search code examples
kubeflowkubeflow-pipelineskalekubeflow-kale

How to specify task pod resources in Kubeflow Kale

Is it possible to specify the pod resources (memory and CPU requests and limits), for a task in a kubeflow pipeline created with Kale?

To provide more details, when writing the kubflow pipeline in python using the DSL, I can specify the task pod resources as follows:

    task1 = (
        train_op(...)
        .set_memory_request('600M')
        .set_memory_limit('1200M')
        .set_cpu_request('700m')
        .set_cpu_limit('1400m')
    )

Is it possible to do the same with Kale?

Solution

  • From what I can see, setting pod resource requests and limits for each task in the kubeflow pipeline is not possible with Kale. The best alternative is to set default resources for the namespace, which will set the pod requests and limits for all tasks in the kubeflow pipeline.