def test_update_with_resources(self, client, created_endpoints,
experiment_run, model_for_deployment):
endpoint_name = _utils.generate_default_name()
endpoint = client.set_endpoint(endpoint_name)
created_endpoints.append(endpoint)
original_status = endpoint.get_status()
original_build_ids = get_build_ids(original_status)
experiment_run.log_model(model_for_deployment['model'],
custom_modules=[])
experiment_run.log_requirements(['scikit-learn'])
resources = '{"cpu": 0.25, "memory": "100M"}'
runner = CliRunner()
result = runner.invoke(
cli,
[
'deployment', 'update', 'endpoint', endpoint_name, '--run-id',
experiment_run.id, "-s", "direct", '--resources', resources
],
)
assert not result.exception
resources_dict = Resources._from_dict(json.loads(
resources))._as_dict() # config is `cpu`, wire is `cpu_millis`
assert endpoint.get_update_status(
)['update_request']['resources'] == resources_dict
def test_create_update_body(self):
endpoint = Endpoint(None, None, None, None)
resources = Resources(cpu=.25, memory="512Mi")
env_vars = {
'CUDA_VISIBLE_DEVICES': "1,2",
"VERTA_HOST": "app.verta.ai",
"GIT_TERMINAL_PROMPT": "1"
}
parameter_json = endpoint._create_update_body(DirectUpdateStrategy(),
resources, None,
env_vars)
assert parameter_json == {
'env': [{
"name": 'CUDA_VISIBLE_DEVICES',
'value': '1,2'
}, {
'name': 'GIT_TERMINAL_PROMPT',
'value': '1'
}, {
"name": 'VERTA_HOST',
'value': 'app.verta.ai'
}],
'resources': {
'cpu_millis': 250,
'memory': '512Mi'
},
'strategy':
'rollout',
}
def test_update_with_parameters(self, client, created_endpoints,
experiment_run, model_for_deployment):
experiment_run.log_model(model_for_deployment['model'],
custom_modules=[])
experiment_run.log_requirements(['scikit-learn'])
path = verta._internal_utils._utils.generate_default_name()
endpoint = client.set_endpoint(path)
created_endpoints.append(endpoint)
original_status = endpoint.get_status()
original_build_ids = get_build_ids(original_status)
strategy = CanaryUpdateStrategy(interval=1, step=0.5)
strategy.add_rule(MaximumAverageLatencyThresholdRule(0.8))
updated_status = endpoint.update(experiment_run,
strategy,
resources=Resources(cpu=.25,
memory="512Mi"),
env_vars={
'CUDA_VISIBLE_DEVICES': "1,2",
"VERTA_HOST": "app.verta.ai"
})
# Check that a new build is added:
new_build_ids = get_build_ids(updated_status)
assert len(new_build_ids) - len(
new_build_ids.intersection(original_build_ids)) > 0
def test_update_from_json_config_with_params(self, client, in_tempdir, created_entities, experiment_run, model_for_deployment):
yaml = pytest.importorskip("yaml")
experiment_run.log_model(model_for_deployment['model'], custom_modules=[])
experiment_run.log_environment(Python(['scikit-learn']))
path = verta._internal_utils._utils.generate_default_name()
endpoint = client.set_endpoint(path)
created_entities.append(endpoint)
original_status = endpoint.get_status()
original_build_ids = get_build_ids(original_status)
# Creating config dict:
config_dict = {
"run_id": experiment_run.id,
"strategy": "direct",
"autoscaling": {
"quantities": {"min_replicas": 1, "max_replicas": 4, "min_scale": 0.5, "max_scale": 2.0},
"metrics": [
{"metric": "cpu_utilization", "parameters": [{"name": "target", "value": "0.5"}]},
{"metric": "memory_utilization", "parameters": [{"name": "target", "value": "0.7"}]}
]
},
"env_vars": {"VERTA_HOST": "app.verta.ai"},
"resources": {"cpu": 0.25, "memory": "100M"}
}
filepath = "config.json"
with open(filepath, 'w') as f:
json.dump(config_dict, f)
endpoint.update_from_config(filepath)
update_status = endpoint.get_update_status()
# Check autoscaling:
autoscaling_parameters = update_status["update_request"]["autoscaling"]
autoscaling_quantities = autoscaling_parameters["quantities"]
assert autoscaling_quantities == config_dict["autoscaling"]["quantities"]
autoscaling_metrics = autoscaling_parameters["metrics"]
assert len(autoscaling_metrics) == 2
for metric in autoscaling_metrics:
assert metric["metric_id"] in [1001, 1002, 1003]
if metric["metric_id"] == 1001:
assert metric["parameters"][0]["value"] == "0.5"
else:
assert metric["parameters"][0]["value"] == "0.7"
# Check env_vars:
assert update_status["update_request"]["env"][0]["name"] == "VERTA_HOST"
assert update_status["update_request"]["env"][0]["value"] == "app.verta.ai"
# Check resources:
resources_dict = Resources._from_dict(config_dict["resources"])._as_dict() # config is `cpu`, wire is `cpu_millis`
assert endpoint.get_update_status()['update_request']['resources'] == resources_dict
def test_download_endpoint_manifest(client, data, strategy, in_tempdir):
resources = Resources(cpu=data)
# test that `resources` can actually be used
client.download_endpoint_manifest(
"deployment.yaml",
"/production-prediction",
"production-prediction",
strategy=strategy,
resources=resources,
)
def test_memory(client, data, in_tempdir):
resources = Resources(memory=data)
# test that `resources` can acutally be used
client.download_endpoint_manifest(
"deployment.yaml",
"/production-prediction",
"production-prediction",
strategy=DirectUpdateStrategy(),
resources=resources,
)
def test_download_manifest(self, client, in_tempdir):
download_to_path = "manifest.yaml"
path = verta._internal_utils._utils.generate_default_name()
name = verta._internal_utils._utils.generate_default_name()
strategy = CanaryUpdateStrategy(interval=10, step=0.1)
strategy.add_rule(MaximumAverageLatencyThresholdRule(0.1))
resources = Resources(cpu=.1, memory="128Mi")
autoscaling = Autoscaling(min_replicas=1,
max_replicas=10,
min_scale=0.1,
max_scale=2)
autoscaling.add_metric(CpuUtilizationTarget(0.75))
env_vars = {'env1': "var1", 'env2': "var2"}
filepath = client.download_endpoint_manifest(
download_to_path=download_to_path,
path=path,
name=name,
strategy=strategy,
autoscaling=autoscaling,
resources=resources,
env_vars=env_vars,
)
# can be loaded as YAML
with open(filepath, 'rb') as f:
manifest = yaml.safe_load(f)
assert manifest['kind'] == "Endpoint"
# check environment variables
containers = manifest['spec']['function']['spec']['templates'][
'deployment']['spec']['template']['spec']['containers']
retrieved_env_vars = {
env_var['name']: env_var['value']
for env_var in containers[0]['env']
}
assert retrieved_env_vars == env_vars
def test_memory_negative_type(data):
with pytest.raises(TypeError):
Resources(memory=data)
def test_memory_negative(data):
with pytest.raises(ValueError):
Resources(memory=data)
def test_cpu_milli_negative_type(data):
with pytest.raises(TypeError):
Resources(cpu=data)
def test_cpu_milli_negative(data):
with pytest.raises(ValueError):
Resources(cpu=data)
def test_cpu_milli(client, data, in_tempdir):
Resources(cpu=data)
def test_cpu_milli(data):
Resources(cpu=data)
def test_memory(data):
Resources(memory=data)