From f5ae88b8d08da8d321cea511b5c962fdc38db720 Mon Sep 17 00:00:00 2001 From: Vara Bonthu Date: Fri, 5 Apr 2024 00:10:12 -0700 Subject: [PATCH] updated Karpenter resources --- ai-ml/jark-stack/terraform/addons.tf | 8 ++++---- .../ray-serve-stablediffusion.yaml | 10 +--------- 2 files changed, 5 insertions(+), 13 deletions(-) diff --git a/ai-ml/jark-stack/terraform/addons.tf b/ai-ml/jark-stack/terraform/addons.tf index 47cb96725..ea16ea717 100644 --- a/ai-ml/jark-stack/terraform/addons.tf +++ b/ai-ml/jark-stack/terraform/addons.tf @@ -231,7 +231,7 @@ module "data_addons" { nodePool: labels: - type: karpenter - - NodeGroupType: spark-executor-gpu-karpenter + - NodeGroupType: g5-gpu-karpenter taints: - key: nvidia.com/gpu value: "Exists" @@ -253,7 +253,7 @@ module "data_addons" { cpu: 1000 disruption: consolidationPolicy: WhenEmpty - consolidateAfter: 30s + consolidateAfter: 180s expireAfter: 720h weight: 100 EOT @@ -276,7 +276,7 @@ module "data_addons" { nodePool: labels: - type: karpenter - - NodeGroupType: spark-driver-cpu-karpenter + - NodeGroupType: x86-cpu-karpenter requirements: - key: "karpenter.k8s.aws/instance-family" operator: In @@ -294,7 +294,7 @@ module "data_addons" { cpu: 1000 disruption: consolidationPolicy: WhenEmpty - consolidateAfter: 30s + consolidateAfter: 180s expireAfter: 720h weight: 100 EOT diff --git a/ai-ml/jark-stack/terraform/examples/inference/stablediffusion-rayserve/ray-serve-stablediffusion.yaml b/ai-ml/jark-stack/terraform/examples/inference/stablediffusion-rayserve/ray-serve-stablediffusion.yaml index c554b956c..a75f20fa0 100644 --- a/ai-ml/jark-stack/terraform/examples/inference/stablediffusion-rayserve/ray-serve-stablediffusion.yaml +++ b/ai-ml/jark-stack/terraform/examples/inference/stablediffusion-rayserve/ray-serve-stablediffusion.yaml @@ -150,18 +150,10 @@ spec: # To access the dashboard: # Option 1: Use the `kubectl port-forward` command. # Option 2: Deploy a public-facing NLB by modifying the configuration file "ai-ml/jark-stack/terraform/helm-values/ingress-nginx-values.yaml". - - path: /dashboard/(.*) + - path: /dashboard/(.*) pathType: ImplementationSpecific backend: service: name: stablediffusion-service port: number: 8265 - # This service endpoint used by the Gradio app to make requests to the Ray Serve deployment. - - path: /serve/(.*) - pathType: ImplementationSpecific - backend: - service: - name: stablediffusion-service - port: - number: 8000