awsome-distributed-training/1.architectures/4.amazon-eks/eks-g5-node-autorepair.yaml at main · awslabs/awsome-distributed-training · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
apiVersion: eksctl.io/v1alpha5
kind: ClusterConfig

metadata:
  name: eks-g5-autorepair
  region: PLACEHOLDER_AWS_REGION
  version: "1.32"

# List availability zones where cluster subnets will be created
availabilityZones:
  - PLACEHOLDER_AZ_1
  - PLACEHOLDER_AZ_2

iam:
  withOIDC: true

# EKS-managed node group(s)
managedNodeGroups:
  # Nodegroup for system pods
  - name: sys
    instanceType: c5.2xlarge
    desiredCapacity: 1
    iam:
      withAddonPolicies:
        autoScaler: true
        cloudWatch: true
    nodeRepairConfig:
      enabled: true

  # GPU nodegroup
  # List availability zones where instances in from this nodegroup will be launched
  - name: g5
    instanceType: g5.8xlarge
    instancePrefix: g5
    privateNetworking: true
    efaEnabled: true
    minSize: 0
    desiredCapacity: 2
    maxSize: 10
    availabilityZones: ["PLACEHOLDER_AZ_2"]
    # Utilize the local instance store volume(s)
    overrideBootstrapCommand: |
      apiVersion: node.eks.aws/v1alpha1
      kind: NodeConfig
      spec:
        instance:
          localStorage:
            strategy: RAID0
    iam:
      withAddonPolicies:
        autoScaler: true
        cloudWatch: true
        ebs: true
        fsx: true
    nodeRepairConfig:
      enabled: true

addons:
  - name: eks-node-monitoring-agent
    resolveConflicts: overwrite
  - name: amazon-cloudwatch-observability
    resolveConflicts: overwrite
    attachPolicyARNs:
      - arn:aws:iam::aws:policy/CloudWatchFullAccess