compute_tpl_32_cpu_autoscaling.yaml

cloud_id: {{env["ANYSCALE_CLOUD_ID"]}}
region: us-west-2

# 1k max replicas (1000 / 32 = 31.25 containers needed)
max_workers: 35

head_node_type:
    name: head_node
    # 32 cpus, x86, 128G mem, 10Gb NIC
    instance_type: m5.8xlarge

worker_node_types:
    - name: worker_node
      # 32 cpus, x86, 128G mem, 10Gb NIC
      instance_type: m5.8xlarge
      # smaller min workers will make the head node cpu usage very high, and crash the head node.
      # issue: https://github.com/ray-project/ray/issues/18908
      min_workers: 5
      # 1k max replicas
      max_workers: 35
      use_spot: false

aws:
  TagSpecifications:
    - ResourceType: "instance"
      Tags:
        - Key: anyscale-user
          Value: '{{env["ANYSCALE_USER"]}}'
        - Key: anyscale-expiration
          Value: '{{env["EXPIRATION_1D"]}}'