ray/benchmarks/benchmark_tests.yaml

- name: single_node
  team: core
  cluster:
    app_config: app_config.yaml
    compute_template: single_node.yaml

  run:
    timeout: 12000
    prepare: sleep 0
    script: python single_node/test_single_node.py

- name: object_store
  team: core
  cluster:
    app_config: app_config.yaml
    compute_template: object_store.yaml

  run:
    timeout: 3600
    prepare: python distributed/wait_cluster.py --num-nodes=50
    script: python object_store/test_object_store.py

- name: many_actors
  team: core
  cluster:
    app_config: app_config.yaml
    compute_template: distributed.yaml

  run:
    timeout: 3600 # 1hr
    prepare: python distributed/wait_cluster.py --num-nodes=65
    script: python distributed/test_many_actors.py

- name: many_actors_smoke_test
  team: core
  cluster:
    app_config: app_config.yaml
    compute_template: distributed_smoke_test.yaml

  run:
    timeout: 3600 # 1hr
    prepare: python distributed/wait_cluster.py --num-nodes=2
    script: SMOKE_TEST=1 python distributed/test_many_actors.py

- name: many_tasks
  team: core
  cluster:
    app_config: app_config.yaml
    compute_template: distributed.yaml

  run:
    timeout: 3600 # 1hr
    prepare: python distributed/wait_cluster.py --num-nodes=65
    script: python distributed/test_many_tasks.py --num-tasks=10000

- name: many_tasks_smoke_test
  team: core
  cluster:
    app_config: app_config.yaml
    compute_template: distributed_smoke_test.yaml

  run:
    timeout: 3600 # 1hr
    prepare: python distributed/wait_cluster.py --num-nodes=2
    script: python distributed/test_many_tasks.py --num-tasks=100

- name: many_pgs
  team: core
  cluster:
    app_config: app_config.yaml
    compute_template: distributed.yaml

  run:
    timeout: 3600 # 1hr
    prepare: python distributed/wait_cluster.py --num-nodes=65
    script: python distributed/test_many_pgs.py

- name: many_pgs_smoke_test
  team: core
  cluster:
    app_config: app_config.yaml
    compute_template: distributed_smoke_test.yaml

  run:
    timeout: 3600 # 1hr
    prepare: python distributed/wait_cluster.py --num-nodes=2
    script: SMOKE_TEST=1 python distributed/test_many_pgs.py

# NOTE: No smoke test since this shares a script with the many_tasks_smoke_test
- name: many_nodes
  team: core
  cluster:
    app_config: app_config.yaml
    compute_template: many_nodes.yaml

  run:
    timeout: 3600 # 1hr
    prepare: python distributed/wait_cluster.py --num-nodes=250
    script: python distributed/test_many_tasks.py --num-tasks=1000

- name: many_tasks_redis_ha
  team: core
  cluster:
    app_config: app_config.yaml
    compute_template: distributed.yaml

  run:
    timeout: 3600 # 1hr
    prepare: python distributed/wait_cluster.py --num-nodes=65
    script: python distributed/test_many_tasks.py --num-tasks=10000

  app_env_vars:
    RAY_gcs_grpc_based_pubsub: "1"
    RAY_gcs_storage: "memory"

- name: many_actors_redis_ha
  team: core
  cluster:
    app_config: app_config.yaml
    compute_template: distributed.yaml

  run:
    timeout: 3600 # 1hr
    prepare: python distributed/wait_cluster.py --num-nodes=65
    script: python distributed/test_many_actors.py

  app_env_vars:
    RAY_gcs_grpc_based_pubsub: "1"
    RAY_gcs_storage: "memory"

- name: many_nodes_redis_ha
  team: core
  cluster:
    app_config: app_config.yaml
    compute_template: many_nodes.yaml

  run:
    timeout: 3600 # 1hr
    prepare: python distributed/wait_cluster.py --num-nodes=250
    script: python distributed/test_many_tasks.py --num-tasks=1000

  app_env_vars:
    RAY_gcs_grpc_based_pubsub: "1"
    RAY_gcs_storage: "memory"

- name: many_pgs_redis_ha
  team: core
  cluster:
    app_config: app_config.yaml
    compute_template: distributed.yaml

  run:
    timeout: 3600 # 1hr
    prepare: python distributed/wait_cluster.py --num-nodes=65
    script: python distributed/test_many_pgs.py

  app_env_vars:
    RAY_gcs_grpc_based_pubsub: "1"
    RAY_gcs_storage: "memory"

- name: scheduling_test_many_0s_tasks_single_node
  team: core
  cluster:
    app_config: app_config.yaml
    compute_template: scheduling.yaml

  run:
    timeout: 3600
    prepare: python distributed/wait_cluster.py --num-nodes=32
    script: python distributed/test_scheduling.py --total-num-task=1984000 --num-cpu-per-task=1 --task-duration-s=0 --total-num-actors=1 --num-actors-per-nodes=1

- name: scheduling_test_many_0s_tasks_many_nodes
  team: core
  cluster:
    app_config: app_config.yaml
    compute_template: scheduling.yaml

  run:
    timeout: 3600
    prepare: python distributed/wait_cluster.py --num-nodes=32
    script: python distributed/test_scheduling.py --total-num-task=1984000 --num-cpu-per-task=1 --task-duration-s=0 --total-num-actors=32 --num-actors-per-nodes=1

- name: scheduling_test_many_5s_tasks_single_node
  team: core
  cluster:
    app_config: app_config.yaml
    compute_template: scheduling.yaml

  run:
    timeout: 3600
    prepare: python distributed/wait_cluster.py --num-nodes=32
    script: python distributed/test_scheduling.py --total-num-task=1984000 --num-cpu-per-task=1 --task-duration-s=5 --total-num-actors=1 --num-actors-per-nodes=1
  stable: false

- name: scheduling_test_many_5s_tasks_many_nodes
  team: core
  cluster:
    app_config: app_config.yaml
    compute_template: scheduling.yaml

  run:
    timeout: 3600
    prepare: python distributed/wait_cluster.py --num-nodes=32
    script: python distributed/test_scheduling.py --total-num-task=1984000 --num-cpu-per-task=1 --task-duration-s=5 --total-num-actors=32 --num-actors-per-nodes=1
  stable: false