Hi all, Is there a way to configure resource requests and limits for

flyte-binary

pods? We are seeing the flyte-binary pods failing healthchecks sometimes causing the pod to be terminated and a new one brought up which causes some downtime. One recommendation from our platform team is to set requests and limits for memory and CPU. But I am not sure how to do that for flyte-binary. We are setting these requests and limits for tasks via

task_resources

in the values.yaml though but not sure it works for the main

flyte-binary

pods.

You can this is in the helm

Can you please point where to put the resource requests and limits in the values.yaml? I tried putting it under deployment, but it didn't show up in the terraform plan output:

deployment:
  annotations:
    downscaler/exclude: "true"
  readinessProbe:
    tcpSocket:
      port: 8088
    initialDelaySeconds: 300
    periodSeconds: 30
  image:
    repository: "${flyte_repo}"
    tag: "${flyte_tag}"
  resources:
    requests:
      memory: 256Mi
      cpu: 100m
    limits:
      memory: 8Gi
      cpu: 500m

here's an example: https://github.com/davidmirror-ops/flyte-the-hard-way/blob/483c0cd7635106588e3a19ea69971eecf9360f53/docs/05-deploy-with-helm.md?plain=1#L45C1-L51

Hi @Samhita Alla we have already set the requests and limits for task pods via

task_resources

I want to know if this applies to the

flyte-binary

pods as well. The current problem we are facing is the

flyte-binary

pods (that host the UI) being evicted / reloaded over and over, due to which the UI is inaccessible, which our platform team feels might be due to requests / limits not being set. Hope that clarifies.

sorry, my bad. @David Espejo (he/him) any idea why adding resources config to deployment isn't working?

@Nandakumar Raghu according to the template definition, the rendered deployment should include the resources you define under

deployment.resources

Are you using a Helm resource with Terraform?

Thanks @David Espejo (he/him), found it and were able to define resource requests and limits for flyte-binary.

Ok, but just adding the resources to that section worked? Or did you have to do something else?

Yes, we added resource limits and requests. Karpenter was evicting the flyte-binary pods as the resource requests and limits were not defined and we were using spot instances.