ray/release/tune_tests/scalability_tests/tune_tests.yaml
2021-06-01 11:19:15 -07:00

97 lines
2 KiB
YAML

- name: bookkeeping_overhead
owner:
mail: "kai@anyscale.com"
slack: "@tune-team"
cluster:
app_config: app_config.yaml
compute_template: tpl_1x16.yaml
run:
timeout: 1200
script: python workloads/test_bookkeeping_overhead.py
- name: durable_trainable
owner:
mail: "kai@anyscale.com"
slack: "@tune-team"
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 ray-release-test-results
- name: long_running_large_checkpoints
owner:
mail: "kai@anyscale.com"
slack: "@tune-team"
cluster:
app_config: app_config.yaml
compute_template: tpl_1x32_hd.yaml
run:
timeout: 86400
script: python workloads/test_long_running_large_checkpoints.py
time_key: last_update
max_update_delay: 120
- name: network_overhead
owner:
mail: "kai@anyscale.com"
slack: "@tune-team"
cluster:
app_config: app_config.yaml
compute_template: tpl_200x2.yaml
run:
timeout: 1200
prepare: python wait_cluster.py 200 600
script: python workloads/test_network_overhead.py
- name: result_throughput_cluster
owner:
mail: "kai@anyscale.com"
slack: "@tune-team"
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
owner:
mail: "kai@anyscale.com"
slack: "@tune-team"
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
owner:
mail: "kai@anyscale.com"
slack: "@tune-team"
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