ray/release/serve_tests/compute_tpl_32_cpu_autoscaling.yaml

30 lines
825 B
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"]}}'