Can some one please assist, How to choose particul...
# flyte-supporta
acoustic-parrot-22802
08/14/2024, 1:32 PMCan some one please assist, How to choose particular gke node and namespace to run my workflows on flyte
a
average-finland-92144
08/14/2024, 1:41 PMHow to choose particular gke nodeyou could set a node affinity using Pod Templates. What is the reason to choose a particular node? is it for GPUs?
namespace to run my workflows on flytefor every project-domain, Flyte not only creates a namespace but run the execution Pods there. But if you need to indicate a particular namespace, you can set it in the Helm chart using
namespace_mappinga
acoustic-parrot-22802
08/14/2024, 2:04 PMGracias.
For Nodes, we have different nodes gpu, standard compute optimized, etc and I want to run on standard node which is requrest from client.
acoustic-parrot-22802
08/14/2024, 2:06 PMCan't we set through flyte config.yaml (.flyte/config.yaml)?
Also, how to do namespace_mapping any document available
Actually so many teams are working with their different works. I want to run on particular namespace and node without affecting their work
a
average-finland-92144
08/14/2024, 2:31 PMhow to do namespace_mapping any document availablehow did you install Flyte? is it flyte-binary or flyte-core? for flyte-binary for example it is set like this
Copy code
configuration:
inline:
namespace_mapping:
template: "my_namespace"average-finland-92144
08/14/2024, 2:34 PMfor the "imperative" node selection, I'm afraid is not something you can set in the config.yaml.
Unless I miss something, your heterogeneous cluster might be using Kubernetes taints, so from your workflow, you could define a Pod Template that sets matching tolerations for the standard nodes
a
acoustic-parrot-22802
08/14/2024, 5:42 PMI tried installing flyte through install.sh script
and that above configuration need to set in config.yaml or some where else?
For node selection I have to define in workflow.py file ?
To select the particular node, is it correct?
acoustic-parrot-22802
08/14/2024, 5:42 PMalso to create config.yaml file
I use this command
flytectl config init --host localhost:8088
acoustic-parrot-22802
08/16/2024, 10:52 AMHi,
I tried this question in "how to create workflow by mentioning node selector in gke" flyte AskAI
It provided as below:
To create a Flyte workflow with node selectors in GKE, you need to configure the node selectors in the FlytePropeller configuration and then use the @task decorator to specify the node selector for each task. Here’s how you can do it:
Configure FlytePropeller: Update the FlytePropeller configuration to include the node selector labels. This is typically done in the flytepropeller config map.
plugins:
k8s:
gpu-device-node-label: cloud.google.com/gke-accelerator
gpu-partition-size-node-label: cloud.google.com/gke-gpu-partition-size
gpu-unpartitioned-toleration:
effect: NoSchedule
key: cloud.google.com/gke-gpu-partition-size
operator: Equal
value: DoesNotExist
Define Tasks with Node Selectors: Use the @task decorator to specify the node selector for each task.
from flytekit import task, workflow, Resources
@task(requests=Resources(cpu="1", memory="500Mi"), node_selector={"cloud.google.com/gke-accelerator": "nvidia-tesla-t4"})
def task_a() -> str:
return "Output from task A"
@task(requests=Resources(cpu="1", memory="500Mi"), node_selector={"cloud.google.com/gke-accelerator": "nvidia-tesla-v100"})
def task_b(input: str) -> str:
return f"Output from task B with input: {input}"
@workflow
def my_workflow() -> str:
a = task_a()
b = task_b(input=a)
return b
In this example, task_a is scheduled on nodes with the label nvidia-tesla-t4, and task_b is scheduled on nodes with the label nvidia-tesla-v100.
is this correct. If yes, what is the flyte propeller config file and where it will store?
20 Views