try vllm eks test with new test path

Author: jinyan-li1

test/platforms/entrypoint.py

@@ -3,6 +3,7 @@
 from src.config import is_new_test_structure_enabled
 from src.buildspec import Buildspec
 from test.platforms.infra.ec2.setup import EC2Platform
+from test.platforms.infra.eks.setup import EKSPlatform
 from test.test_utils import get_framework_from_image_uri
 
 
@@ -116,10 +117,7 @@ def main():
         raise
 
     # Filter for applicable tests
-    if test_type == "ec2":
-        applicable_tests = [test for test in buildspec_data["tests"] if test["platform"].startswith("ec2")]
-    else:
-        applicable_tests = [test for test in buildspec_data["tests"] if test["platform"] == test_type]
+    applicable_tests = [test for test in buildspec_data["tests"] if test["platform"].startswith(test_type)]
 
     print(f"Found {len(buildspec_data['tests'])} test configurations")
     print(f"Found {len(applicable_tests)} applicable test configurations for {test_type}")
@@ -144,6 +142,22 @@ def main():
             except Exception as e:
                 print(f"Test failed: {e}")
                 raise
+        elif test_type == "eks" and platform_name.startswith("eks"):
+            print(f"Executing EKS test for platform: {platform_name}")
+            platform = EKSPlatform()
+            try:
+                setup_params = {**test_config["params"], **buildspec_data["globals"]}
+                print(f"Setup parameters: {setup_params}")
+                platform.setup(setup_params)
+                print(f"Platform setup completed")
+
+                print(f"Executing {len(test_config['run'])} commands:")
+                for cmd in test_config["run"]:
+                    print(f"  - {cmd}")
+                    platform.execute_command(cmd)
+            except Exception as e:
+                print(f"Test failed: {e}")
+                raise
         else:
             print(f"Skipping test config {i+1}: test_type={test_type}, platform={platform_name}")
 

test/platforms/infra/eks/__init__.py

@@ -0,0 +1,44 @@
+import os
+from invoke import run
+from test.test_utils import LOGGER
+
+class EKSPlatform:
+    def __init__(self):
+        self.resources = None
+        self.region = os.getenv("REGION", "us-west-2")
+        self.build_context = os.getenv("BUILD_CONTEXT")
+        self.cluster_name = None
+        self.namespace = None
+
+    def setup(self, params):
+        """
+        Setup EKS infrastructure and return any resources needed for tests
+        """
+        print(f"Setting up EKS platform with params: {params}")
+
+        framework = params.get("framework")
+        cluster_prefix = params.get("cluster") 
+        self.cluster_name = f"{cluster_prefix}-{self.build_context}"
+        self.namespace = params.get("namespace")
+        
+        print(f"EKS Platform - Framework: {framework}")
+        print(f"EKS Platform - Cluster: {self.cluster_name}")
+        print(f"EKS Platform - Namespace: {self.namespace}")
+
+        if not os.getenv("DLC_IMAGE"):
+            raise ValueError("DLC_IMAGE environment variable not set")
+
+    def execute_command(self, cmd):
+        """
+        Execute a test command with proper environment setup
+        """
+        env = {
+            "AWS_REGION": self.region,
+            "CLUSTER_NAME": self.cluster_name,
+            "NAMESPACE": self.namespace,
+            "BUILD_CONTEXT": self.build_context,
+            "DLC_IMAGE": os.getenv("DLC_IMAGE"),
+        }
+        
+        LOGGER.info(f"Executing command with EKS environment: {cmd}")
+        run(cmd, env=env)

test/platforms/infra/eks/setup.py

@@ -0,0 +1,26 @@
+apiVersion: eksctl.io/v1alpha5
+kind: ClusterConfig
+
+metadata:
+  name: ${CLUSTER_NAME}
+  region: ${AWS_REGION} 
+  version: "${EKS_VERSION}" # Latest stable EKS version
+
+# Specify availability zones
+availabilityZones: ["${AWS_REGION}a", "${AWS_REGION}b"]
+
+# Enable CloudWatch logging
+cloudWatch:
+  clusterLogging:
+    enableTypes: ["api", "audit", "authenticator", "controllerManager", "scheduler"]
+
+# Add-ons for the cluster
+addons:
+  - name: vpc-cni
+    version: latest
+  - name: coredns
+    version: latest
+  - name: kube-proxy
+    version: latest
+  - name: aws-ebs-csi-driver
+    version: latest

test/platforms/infra/eks/vllm/__init__.py

@@ -0,0 +1,18 @@
+apiVersion: v1
+kind: PersistentVolume
+metadata:
+  name: fsx-lustre-pv
+spec:
+  capacity:
+    storage: 1200Gi  # Adjust based on your FSx Lustre filesystem size
+  volumeMode: Filesystem
+  accessModes:
+    - ReadWriteMany
+  persistentVolumeReclaimPolicy: Retain
+  storageClassName: fsx-sc
+  csi:
+    driver: fsx.csi.aws.com
+    volumeHandle: <fs-id>  # FSx Lustre filesystem ID
+    volumeAttributes:
+      dnsname: <dns-name>  # FSx Lustre DNS name
+      mountname: <mount-name>  # The mount name of your FSx Lustre filesyst

test/platforms/infra/eks/vllm/eks-cluster.yaml

@@ -0,0 +1,12 @@
+apiVersion: v1
+kind: PersistentVolumeClaim
+metadata:
+  name: fsx-lustre-pvc
+  namespace: vllm
+spec:
+  accessModes:
+    - ReadWriteMany
+  storageClassName: fsx-sc
+  resources:
+    requests:
+      storage: 1200Gi  # Should match the PV capacity

test/platforms/infra/eks/vllm/fsx-lustre-pv.yaml

@@ -0,0 +1,18 @@
+apiVersion: storage.k8s.io/v1
+kind: StorageClass
+metadata:
+  name: fsx-sc
+provisioner: fsx.csi.aws.com
+parameters:
+  subnetId: <subnet-id>
+  securityGroupIds: <sg-id>
+  deploymentType: SCRATCH_2
+  automaticBackupRetentionDays: "0"
+  dailyAutomaticBackupStartTime: "00:00"
+  copyTagsToBackups: "false"
+  perUnitStorageThroughput: "50"
+  dataCompressionType: "NONE"
+reclaimPolicy: Retain
+volumeBindingMode: Immediate
+mountOptions:
+  - flock

test/platforms/infra/eks/vllm/fsx-lustre-pvc.yaml

@@ -0,0 +1,45 @@
+apiVersion: eksctl.io/v1alpha5
+kind: ClusterConfig
+
+metadata:
+  name: ${CLUSTER_NAME}
+  region: ${AWS_REGION}
+
+managedNodeGroups:
+  - name: vllm-p4d-nodes-efa
+    instanceType: p4d.24xlarge
+    minSize: 0
+    maxSize: 2
+    desiredCapacity: 0
+    availabilityZones: ["${AWS_REGION}a"] # EFA-enabled nodegroups must have only one availability zone
+    volumeSize: 100
+    privateNetworking: true
+    amiFamily: AmazonLinux2023
+    ami: ami-009600863d7ec0b6a
+    preBootstrapCommands:
+      - dnf install -y lustre-client
+      - lfs --version
+    labels:
+      role: large-model-worker
+      nvidia.com/gpu: "true"
+      k8s.amazonaws.com/accelerator: nvidia-gpu
+      aws.amazon.com/efa: "true"  # Add EFA label
+    tags:
+      nodegroup-role: large-model-worker
+      k8s.io/cluster-autoscaler/node-template/label/role: large-model-worker
+      k8s.io/cluster-autoscaler/enabled: "true"
+      k8s.io/cluster-autoscaler/${CLUSTER_NAME}: "owned"
+      k8s.io/cluster-autoscaler/node-template/resources/vpc.amazonaws.com/efa: "4"
+      k8s.io/cluster-autoscaler/node-template/label/aws.amazon.com/efa: "true"
+    iam:
+      withAddonPolicies:
+        autoScaler: true
+        albIngress: true
+        cloudWatch: true
+        ebs: true
+        imageBuilder: true
+    # Enable EFA interfaces
+    efaEnabled: true
+    capacityReservation:
+      capacityReservationTarget:
+        capacityReservationID: "cr-08e4079a2d40aee96"

test/platforms/infra/eks/vllm/fsx-storage-class.yaml

@@ -0,0 +1,29 @@
+apiVersion: networking.k8s.io/v1
+kind: Ingress
+metadata:
+  name: vllm-deepseek-32b-lws-ingress
+  namespace: vllm
+  annotations:
+    # Use AWS Load Balancer Controller with ALB
+    alb.ingress.kubernetes.io/scheme: internet-facing
+    alb.ingress.kubernetes.io/target-type: ip
+    alb.ingress.kubernetes.io/security-groups: <sg-id>
+    alb.ingress.kubernetes.io/healthcheck-path: /health
+    alb.ingress.kubernetes.io/healthcheck-port: '8000'
+    alb.ingress.kubernetes.io/healthcheck-protocol: HTTP
+    alb.ingress.kubernetes.io/listen-ports: '[{"HTTP": 80}]'
+    alb.ingress.kubernetes.io/load-balancer-attributes: load_balancing.cross_zone.enabled=true
+    # Specify ALB class
+    kubernetes.io/ingress.class: alb
+spec:
+  ingressClassName: alb
+  rules:
+  - http:
+      paths:
+      - path: /
+        pathType: Prefix
+        backend:
+          service:
+            name: vllm-deepseek-32b-lws-leader
+            port:
+              number: 8000