Merge pull request #5127 from sgibson91/nasa-cryo/dedicated-nodegroups

nasa-cryo: create hub-specific nodegroups
2i2c-org · Nov 18, 2024 · 63abd57 · 63abd57
2 parents 4c521cd + c33f956
commit 63abd57
Show file tree

Hide file tree

Showing 3 changed files with 62 additions and 5 deletions.
diff --git a/config/clusters/nasa-cryo/prod.values.yaml b/config/clusters/nasa-cryo/prod.values.yaml
@@ -13,6 +13,8 @@ basehub:
         GitHubOAuthenticator:
           oauth_callback_url: https://hub.cryointhecloud.com/hub/oauth_callback
     singleuser:
+      nodeSelector:
+        2i2c/hub-name: prod
       extraEnv:
         SCRATCH_BUCKET: s3://nasa-cryo-scratch/$(JUPYTERHUB_USER)
         PANGEO_SCRATCH: s3://nasa-cryo-scratch/$(JUPYTERHUB_USER)

diff --git a/config/clusters/nasa-cryo/staging.values.yaml b/config/clusters/nasa-cryo/staging.values.yaml
@@ -13,6 +13,8 @@ basehub:
         GitHubOAuthenticator:
           oauth_callback_url: https://staging.hub.cryointhecloud.com/hub/oauth_callback
     singleuser:
+      nodeSelector:
+        2i2c/hub-name: staging
       extraEnv:
         SCRATCH_BUCKET: s3://nasa-cryo-scratch-staging/$(JUPYTERHUB_USER)
         PANGEO_SCRATCH: s3://nasa-cryo-scratch-staging/$(JUPYTERHUB_USER)

diff --git a/eksctl/nasa-cryo.jsonnet b/eksctl/nasa-cryo.jsonnet
@@ -25,13 +25,64 @@ local nodeAz = "us-west-2a";
 // A `node.kubernetes.io/instance-type label is added, so pods
 // can request a particular kind of node with a nodeSelector
 local notebookNodes = [
-    { instanceType: "r5.xlarge", nameSuffix: "b" },
-    { instanceType: "r5.4xlarge" }, // FIXME: tainted, to be deleted when empty, replaced by equivalent during k8s upgrade
-    { instanceType: "r5.4xlarge", nameSuffix: "b" },
-    { instanceType: "r5.16xlarge" },
+    { instanceType: "r5.xlarge", nameSuffix: "b" },  // FIXME: tainted
+    {
+        instanceType: "r5.xlarge",
+        namePrefix: "nb-staging",
+        labels+: { "2i2c/hub-name": "staging" },
+        tags+: { "2i2c:hub-name": "staging" },
+    },
+    {
+        instanceType: "r5.xlarge",
+        namePrefix: "nb-prod",
+        labels+: { "2i2c/hub-name": "prod" },
+        tags+: { "2i2c:hub-name": "prod" },
+    },
+    {
+        instanceType: "r5.4xlarge",
+        namePrefix: "nb-staging",
+        labels+: { "2i2c/hub-name": "staging" },
+        tags+: { "2i2c:hub-name": "staging" },
+    },
+    {
+        instanceType: "r5.4xlarge",
+        namePrefix: "nb-prod",
+        labels+: { "2i2c/hub-name": "prod" },
+        tags+: { "2i2c:hub-name": "prod" },
+    },
+    {
+        instanceType: "r5.16xlarge",
+        namePrefix: "nb-staging",
+        labels+: { "2i2c/hub-name": "staging" },
+        tags+: { "2i2c:hub-name": "staging" },
+    },
+    {
+        instanceType: "r5.16xlarge",
+        namePrefix: "nb-prod",
+        labels+: { "2i2c/hub-name": "prod" },
+        tags+: { "2i2c:hub-name": "prod" },
+    },
+    {
+        instanceType: "g4dn.xlarge",
+        namePrefix: "staging",
+        labels+: { "2i2c/hub-name": "staging" },
+        tags+: {
+            "2i2c:hub-name": "staging",
+            "k8s.io/cluster-autoscaler/node-template/resources/nvidia.com/gpu": "1"
+        },
+        taints+: {
+            "nvidia.com/gpu": "present:NoSchedule"
+        },
+        // Allow provisioning GPUs across all AZs, to prevent situation where all
+        // GPUs in a single AZ are in use and no new nodes can be spawned
+        availabilityZones: masterAzs,
+    },
     {
         instanceType: "g4dn.xlarge",
+        namePrefix: "prod",
+        labels+: { "2i2c/hub-name": "prod" },
         tags+: {
+            "2i2c:hub-name": "prod",
             "k8s.io/cluster-autoscaler/node-template/resources/nvidia.com/gpu": "1"
         },
         taints+: {
@@ -95,7 +146,7 @@ local daskNodes = [
     [
         ng + {
             namePrefix: 'core',
-            nameSuffix: 'b',
+            nameSuffix: 'a',
             nameIncludeInstanceType: false,
             availabilityZones: [nodeAz],
             ssh: {
@@ -108,6 +159,7 @@ local daskNodes = [
                 "hub.jupyter.org/node-purpose": "core",
                 "k8s.dask.org/node-purpose": "core"
             },
+            tags+: { "2i2c:node-purpose": "core" },
         },
     ] + [
         ng + {
@@ -123,6 +175,7 @@ local daskNodes = [
                 "hub.jupyter.org/node-purpose": "user",
                 "k8s.dask.org/node-purpose": "scheduler"
             },
+            tags+: { "2i2c:node-purpose": "user" },
             taints+: {
                 "hub.jupyter.org_dedicated": "user:NoSchedule",
                 "hub.jupyter.org/dedicated": "user:NoSchedule"