Attempt On demand profile option

dandi · Sep 9, 2024 · 565313c · 565313c
1 parent 45ef110
commit 565313c
Show file tree

Hide file tree

Showing 5 changed files with 67 additions and 2 deletions.
diff --git a/addons.tf b/addons.tf
@@ -168,6 +168,21 @@ module "eks_blueprints_addons" {
         <<-EOT
           clusterName: ${module.eks.cluster_name}
           karpenterRole: ${split("/", module.eks_blueprints_addons.karpenter.node_iam_role_arn)[1]}
+          capacityTypes: ["spot"]
+        EOT
+      ]
+    }
+    karpenter-resources-cpu-on-demand = {
+      name        = "karpenter-resources-cpu-on-demand"
+      description = "A Helm chart for karpenter CPU based resources"
+      chart       = "${path.module}/helm/karpenter-resources"
+      values = [
+        <<-EOT
+          name: cpu-on-demand
+          clusterName: ${module.eks.cluster_name}
+          instanceSizes: ["xlarge", "2xlarge", "4xlarge", "8xlarge", "16xlarge", "24xlarge"]
+          karpenterRole: ${split("/", module.eks_blueprints_addons.karpenter.node_iam_role_arn)[1]}
+          capacityTypes: ["on-demand"]
         EOT
       ]
     }

diff --git a/envs/dandi-staging/managed-jupyterhub.yaml b/envs/dandi-staging/managed-jupyterhub.yaml
@@ -216,6 +216,30 @@ singleuser:
               kubespawner_override:
                 image: ${singleuser_image_repo}:${singleuser_image_tag}
           display_name: Image
+    - default: true
+      description: 6 CPU / 16 GB up to 12C/32G. May take up to 15 mins to start.
+      display_name: Base On Demand
+      kubespawner_override:
+        cpu_guarantee: 6
+        cpu_limit: 12
+        image_pull_policy: Always
+        mem_guarantee: 16G
+        mem_limit: 32G
+        node_selector:
+          NodePool: cpu-on-demand
+      profile_options:
+        image:
+          choices:
+            matlab:
+              display_name: MATLAB (must provide your own license)
+              kubespawner_override:
+                image: ${singleuser_image_repo}:${singleuser_image_tag}-matlab
+            standard:
+              default: true
+              display_name: Standard
+              kubespawner_override:
+                image: ${singleuser_image_repo}:${singleuser_image_tag}
+          display_name: Image
     - default: true
       description: 6 CPU / 16 GB up to 12C/32G. May take up to 15 mins to start.
       display_name: Base

diff --git a/envs/dandi-staging/terraform.tfvars b/envs/dandi-staging/terraform.tfvars
@@ -1,6 +1,6 @@
 name = "eks-hub-staging"
 singleuser_image_repo = "dandiarchive/dandihub"
-region = "us-east-2"
+region = "us-west-2"
 singleuser_image_tag = "latest"
 jupyterhub_domain = "hub-staging.dandiarchive.org"
 dandi_api_domain = "https://api.dandiarchive.org"
diff --git a/envs/shared/jupyterhub.yaml b/envs/shared/jupyterhub.yaml
@@ -59,6 +59,30 @@ singleuser:
         mem_guarantee: 0.5G
         node_selector:
           NodePool: default
+    - display_name: "Base On Demand"
+      description: "6 CPU / 16 GB up to 12C/32G. May take up to 15 mins to start."
+      profile_options:
+        image:
+          display_name: "Image"
+          choices:
+            standard:
+              display_name: "Standard"
+              default: true
+              kubespawner_override:
+                image: "${singleuser_image_repo}:${singleuser_image_tag}"
+            matlab:
+              display_name: "MATLAB (must provide your own license)"
+              kubespawner_override:
+                image: "${singleuser_image_repo}:${singleuser_image_tag}-matlab"
+      default: true
+      kubespawner_override:
+        image_pull_policy: Always
+        cpu_limit: 12
+        cpu_guarantee: 6
+        mem_limit: 32G
+        mem_guarantee: 16G
+        node_selector:
+          NodePool: cpu-on-demand
     - display_name: "Base"
       description: "6 CPU / 16 GB up to 12C/32G. May take up to 15 mins to start."
       profile_options:
@@ -83,6 +107,7 @@ singleuser:
         mem_guarantee: 16G
         node_selector:
           NodePool: default
+
     - display_name: "Medium"
       description: "12C/32G up to 24C/64G. May take up to 15 mins to start."
       profile_options:

diff --git a/helm/karpenter-resources/templates/node-pool.yaml b/helm/karpenter-resources/templates/node-pool.yaml
@@ -38,7 +38,8 @@ spec:
           values: ["amd64"]
         - key: "karpenter.sh/capacity-type"
           operator: In
-          values: ["spot"]
+          values: 
+              - {{ .Values.capacityTypes }}
   disruption:
     consolidationPolicy: WhenEmpty
     consolidateAfter: 30s