wassname/ray
1
0
Fork 0
mirror of https://github.com/wassname/ray.git synced 2026-08-06 13:31:10 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Autoscaler] Multi node commands (#10236)

Browse Source
This commit is contained in:
Alex Wu
2020-08-25 23:35:38 -07:00
committed by GitHub
parent 0dae50b5eb
commit 9ca159aa0b
4 changed files with 93 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/ray/tests/test_resource_demand_scheduler.py
+54 -1
View File
@@ -200,6 +200,10 @@ class AutoscalingTest(unittest.TestCase):
return path
def testGetOrCreateMultiNodeType(self):
config = MULTI_WORKER_CLUSTER.copy()
# Commenting out this line causes the test case to fail?!?!
config["min_workers"] = 0
config_path = self.write_config(config)
config_path = self.write_config(MULTI_WORKER_CLUSTER)
self.provider = MockProvider()
runner = MockProcessRunner()
@@ -214,7 +218,7 @@ class AutoscalingTest(unittest.TestCase):
_runner=runner)
self.waitForNodes(1)
runner.assert_has_call("1.2.3.4", "init_cmd")
runner.assert_has_call("1.2.3.4", "head_setup_cmd")
runner.assert_has_call("1.2.3.4", "setup_cmd")
runner.assert_has_call("1.2.3.4", "start_ray_head")
self.assertEqual(self.provider.mock_nodes[0].node_type, "empty_node")
self.assertEqual(
@@ -348,6 +352,55 @@ class AutoscalingTest(unittest.TestCase):
runner.assert_has_call("172.0.0.1", "CPU: 32")
runner.assert_has_call("172.0.0.1", "GPU: 8")
def testCommandPassing(self):
t = "custom"
config = MULTI_WORKER_CLUSTER.copy()
config["available_node_types"]["p2.8xlarge"][
"worker_setup_commands"] = ["new_worker_setup_command"]
config["available_node_types"]["p2.xlarge"][
"initialization_commands"] = ["new_worker_initialization_cmd"]
config["available_node_types"]["p2.xlarge"]["resources"][t] = 1
# Commenting out this line causes the test case to fail?!?!
config["min_workers"] = 0
config["max_workers"] = 10
config_path = self.write_config(config)
self.provider = MockProvider()
runner = MockProcessRunner()
autoscaler = StandardAutoscaler(
config_path,
LoadMetrics(),
max_failures=0,
process_runner=runner,
update_interval_s=0)
assert len(self.provider.non_terminated_nodes({})) == 0
autoscaler.update()
self.waitForNodes(0)
autoscaler.request_resources([{"CPU": 1}])
autoscaler.update()
self.waitForNodes(1)
assert self.provider.mock_nodes[0].node_type == "m4.large"
autoscaler.request_resources([{"GPU": 8}])
autoscaler.update()
self.waitForNodes(2)
assert self.provider.mock_nodes[1].node_type == "p2.8xlarge"
autoscaler.request_resources([{"GPU": 1}] * 9)
# autoscaler.request_resources([{t: 1}])
autoscaler.update()
self.waitForNodes(3)
assert self.provider.mock_nodes[2].node_type == "p2.xlarge"
autoscaler.update()
sleep(0.1)
runner.assert_has_call(self.provider.mock_nodes[1].internal_ip,
"new_worker_setup_command")
runner.assert_not_has_call(self.provider.mock_nodes[1].internal_ip,
"setup_cmd")
runner.assert_not_has_call(self.provider.mock_nodes[1].internal_ip,
"worker_setup_cmd")
runner.assert_has_call(self.provider.mock_nodes[2].internal_ip,
"new_worker_initialization_cmd")
runner.assert_not_has_call(self.provider.mock_nodes[2].internal_ip,
"init_cmd")
if __name__ == "__main__":
import sys