ray/release/tune_tests/scalability_tests/tune_tests.yaml

- name: bookkeeping_overhead
  cluster:
    app_config: app_config.yaml
    compute_template: tpl_1x16.yaml

  run:
    timeout: 1200
    script: python workloads/test_bookkeeping_overhead.py


- name: durable_trainable
  cluster:
    app_config: app_config.yaml
    compute_template: tpl_16x2.yaml

  run:
    timeout: 900
    prepare: python wait_cluster.py 16 600
    script: python workloads/test_durable_trainable.py --bucket data-test-ilr

- name: long_running_large_checkpoints
  cluster:
    app_config: app_config.yaml
    compute_template: tpl_1x32_hd.yaml

  run:
    timeout: 86400
    script: python workloads/test_long_running_large_checkpoints.py
    long_running: True

  smoke_test:
    run:
      timeout: 3600


- name: network_overhead
  cluster:
    app_config: app_config.yaml
    compute_template: tpl_100x2.yaml

  run:
    timeout: 900
    prepare: python wait_cluster.py 100 1200
    script: python workloads/test_network_overhead.py

  smoke_test:
    cluster:
      compute_template: tpl_20x2.yaml

    run:
      timeout: 400
      prepare: python wait_cluster.py 20 1200

- name: result_throughput_cluster
  cluster:
    app_config: app_config.yaml
    compute_template: tpl_16x64.yaml

  run:
    timeout: 600
    prepare: python wait_cluster.py 16 600
    script: python workloads/test_result_throughput_cluster.py

- name: result_throughput_single_node
  cluster:
    app_config: app_config.yaml
    compute_template: tpl_1x96.yaml

  run:
    timeout: 600
    script: python workloads/test_result_throughput_single_node.py

- name: xgboost_sweep
  cluster:
    app_config: app_config_data.yaml
    compute_template: tpl_16x64.yaml

  run:
    timeout: 3600
    prepare: python wait_cluster.py 16 600
    script: python workloads/test_xgboost_sweep.py