def test_latest_version(self, download_two_model_versions, get_test_dir,
start_server_update_flow_latest,
create_channel_for_update_flow_latest):
resnet_v1, resnet_v2 = download_two_model_versions
dir = get_test_dir + '/saved_models/' + 'update/'
resnet_v1_copy_dir = copy_model(resnet_v1, 1, dir)
time.sleep(8)
stub = create_channel_for_update_flow_latest
print("Getting info about resnet model")
model_name = 'resnet'
out_name = 'resnet_v1_50/predictions/Reshape_1'
expected_input_metadata = {
'input': {
'dtype': 1,
'shape': [1, 3, 224, 224]
}
}
expected_output_metadata = {out_name: {'dtype': 1, 'shape': [1, 1000]}}
request = get_model_metadata(model_name=model_name)
response = stub.GetModelMetadata(request, 10)
input_metadata, output_metadata = model_metadata_response(
response=response)
print(output_metadata)
assert model_name == response.model_spec.name
assert expected_input_metadata == input_metadata
assert expected_output_metadata == output_metadata
shutil.rmtree(resnet_v1_copy_dir)
resnet_v2_copy_dir = copy_model(resnet_v2, 2, dir)
time.sleep(3)
out_name = 'resnet_v2_50/predictions/Reshape_1'
expected_input_metadata = {
'input': {
'dtype': 1,
'shape': [1, 3, 224, 224]
}
}
expected_output_metadata = {out_name: {'dtype': 1, 'shape': [1, 1001]}}
request = get_model_metadata(model_name=model_name)
response = stub.GetModelMetadata(request, 10)
input_metadata, output_metadata = model_metadata_response(
response=response)
print(output_metadata)
assert model_name == response.model_spec.name
assert expected_input_metadata == input_metadata
assert expected_output_metadata == output_metadata
shutil.rmtree(resnet_v2_copy_dir)
def test_latest_version_rest(self, download_two_model_versions,
get_test_dir,
start_server_update_flow_latest):
resnet_v1, resnet_v2 = download_two_model_versions
dir = get_test_dir + '/saved_models/' + 'update/'
resnet_v1_copy_dir = copy_model(resnet_v1, 1, dir)
time.sleep(8)
print("Getting info about resnet model")
model_name = 'resnet'
out_name = 'resnet_v1_50/predictions/Reshape_1'
expected_input_metadata = {
'input': {
'dtype': 1,
'shape': [1, 3, 224, 224]
}
}
expected_output_metadata = {out_name: {'dtype': 1, 'shape': [1, 1000]}}
rest_url = 'http://localhost:5562/v1/models/resnet/metadata'
response = get_model_metadata_response_rest(rest_url)
input_metadata, output_metadata = model_metadata_response(
response=response)
print(output_metadata)
assert model_name == response.model_spec.name
assert expected_input_metadata == input_metadata
assert expected_output_metadata == output_metadata
shutil.rmtree(resnet_v1_copy_dir)
resnet_v2_copy_dir = copy_model(resnet_v2, 2, dir)
time.sleep(3)
out_name = 'resnet_v2_50/predictions/Reshape_1'
expected_input_metadata = {
'input': {
'dtype': 1,
'shape': [1, 3, 224, 224]
}
}
expected_output_metadata = {out_name: {'dtype': 1, 'shape': [1, 1001]}}
response = get_model_metadata_response_rest(rest_url)
input_metadata, output_metadata = model_metadata_response(
response=response)
print(output_metadata)
assert model_name == response.model_spec.name
assert expected_input_metadata == input_metadata
assert expected_output_metadata == output_metadata
shutil.rmtree(resnet_v2_copy_dir)
def test_get_model_metadata(self, resnet_2_out_model_downloader,
create_channel_for_port_mapping_server,
start_server_with_mapping):
print("Downloaded model files:", resnet_2_out_model_downloader)
stub = create_channel_for_port_mapping_server
model_name = 'resnet_2_out'
expected_input_metadata = {
'new_key': {
'dtype': 1,
'shape': [1, 3, 224, 224]
}
}
expected_output_metadata = {
'mask': {
'dtype': 1,
'shape': [1, 2048, 7, 7]
},
'output': {
'dtype': 1,
'shape': [1, 2048, 7, 7]
}
}
request = get_model_metadata(model_name=model_name)
response = stub.GetModelMetadata(request, 10)
print("response", response)
input_metadata, output_metadata = model_metadata_response(
response=response)
assert model_name == response.model_spec.name
assert expected_input_metadata == input_metadata
assert expected_output_metadata == output_metadata
def test_get_model_metadata_rest(self, download_two_model_versions,
start_server_multi_model):
print("Downloaded model files:", download_two_model_versions)
urls = ['http://localhost:5561/v1/models/resnet/metadata',
'http://localhost:5561/v1/models/resnet/versions/1/metadata']
expected_outputs_metadata = \
[{'resnet_v2_50/predictions/Reshape_1':
{'dtype': 1, 'shape': [1, 1001]}},
{'resnet_v1_50/predictions/Reshape_1':
{'dtype': 1, 'shape': [1, 1000]}}
]
for x in range(len(urls)):
print("Getting info about resnet model version:".format(
urls[x]))
model_name = 'resnet'
expected_input_metadata = {'input': {'dtype': 1,
'shape': [1, 3, 224, 224]}}
expected_output_metadata = expected_outputs_metadata[x]
response = get_model_metadata_response_rest(urls[x])
input_metadata, output_metadata = model_metadata_response(
response=response)
print(output_metadata)
assert model_name == response.model_spec.name
assert expected_input_metadata == input_metadata
assert expected_output_metadata == output_metadata
def test_get_model_metadata(self, resnet_8_batch_model_downloader,
start_server_batch_model,
create_channel_for_batching_server):
print("Downloaded model files:", resnet_8_batch_model_downloader)
stub = create_channel_for_batching_server
model_name = 'resnet'
out_name = 'resnet_v1_50/predictions/Reshape_1'
expected_input_metadata = {
'input': {
'dtype': 1,
'shape': [8, 3, 224, 224]
}
}
expected_output_metadata = {out_name: {'dtype': 1, 'shape': [8, 1000]}}
request = get_model_metadata(model_name='resnet')
response = stub.GetModelMetadata(request, 10)
input_metadata, output_metadata = model_metadata_response(
response=response)
print(output_metadata)
assert model_name == response.model_spec.name
assert expected_input_metadata == input_metadata
assert expected_output_metadata == output_metadata
def test_get_model_metadata(self, start_server_single_model_from_gc,
create_channel_for_port_single_server):
result = start_server_single_model_from_gc
print("docker starting status:", result)
time.sleep(30) # Waiting for inference service to load models
assert result == 0, "docker container was not started successfully"
stub = create_channel_for_port_single_server
model_name = 'resnet'
out_name = 'resnet_v1_50/predictions/Reshape_1'
expected_input_metadata = {
'input': {
'dtype': 1,
'shape': [1, 3, 224, 224]
}
}
expected_output_metadata = {out_name: {'dtype': 1, 'shape': [1, 1, 1]}}
request = get_model_metadata(model_name='resnet')
response = stub.GetModelMetadata(request, 10)
input_metadata, output_metadata = model_metadata_response(
response=response)
print(output_metadata)
assert model_name == response.model_spec.name
assert expected_input_metadata == input_metadata
assert expected_output_metadata == output_metadata
def test_get_model_metadata_rest(self, resnet_2_out_model_downloader,
start_server_with_mapping):
print("Downloaded model files:", resnet_2_out_model_downloader)
model_name = 'resnet_2_out'
expected_input_metadata = {
'new_key': {
'dtype': 1,
'shape': [1, 3, 224, 224]
}
}
expected_output_metadata = {
'mask': {
'dtype': 1,
'shape': [1, 2048, 7, 7]
},
'output': {
'dtype': 1,
'shape': [1, 2048, 7, 7]
}
}
rest_url = 'http://localhost:5556/v1/models/resnet_2_out/metadata'
response = get_model_metadata_response_rest(rest_url)
print("response", response)
input_metadata, output_metadata = model_metadata_response(
response=response)
assert model_name == response.model_spec.name
assert expected_input_metadata == input_metadata
assert expected_output_metadata == output_metadata
def test_get_model_metadata(self, download_two_model_versions,
start_server_multi_model,
create_channel_for_port_multi_server):
print("Downloaded model files:", download_two_model_versions)
# Connect to grpc service
stub = create_channel_for_port_multi_server
versions = [None, 1]
expected_outputs_metadata = \
[{'resnet_v2_50/predictions/Reshape_1':
{'dtype': 1, 'shape': [1, 1001]}},
{'resnet_v1_50/predictions/Reshape_1':
{'dtype': 1, 'shape': [1, 1000]}}
]
for x in range(len(versions)):
print("Getting info about resnet model version:".format(
versions[x]))
model_name = 'resnet'
expected_input_metadata = {'input': {'dtype': 1,
'shape': [1, 3, 224, 224]}}
expected_output_metadata = expected_outputs_metadata[x]
request = get_model_metadata(model_name='resnet',
version=versions[x])
response = stub.GetModelMetadata(request, 10)
input_metadata, output_metadata = model_metadata_response(
response=response)
print(output_metadata)
assert model_name == response.model_spec.name
assert expected_input_metadata == input_metadata
assert expected_output_metadata == output_metadata
def test_get_model_metadata(self, download_two_models,
start_server_multi_model,
create_channel_for_port_multi_server):
"""
<b>Description</b>
Execute inference request using gRPC interface hosting multiple models
<b>input data</b>
- directory with 2 models in IR format
- docker image
<b>fixtures used</b>
- model downloader
- input data downloader
- service launching
<b>Expected results</b>
- response contains proper response about model metadata for both
models set in config file:
model resnet_v1_50, pnasnet_large
- both served models handles appropriate input formats
"""
print("Downloaded model files:", download_two_models)
# Connect to grpc service
stub = create_channel_for_port_multi_server
versions = [None, 1]
expected_outputs_metadata = \
[{'resnet_v2_50/predictions/Reshape_1':
{'dtype': 1, 'shape': [1, 1001]}},
{'resnet_v1_50/predictions/Reshape_1':
{'dtype': 1, 'shape': [1, 1000]}}
]
for x in range(len(versions)):
print("Getting info about resnet model version:".format(
versions[x]))
model_name = 'resnet'
expected_input_metadata = {
'input': {
'dtype': 1,
'shape': [1, 3, 224, 224]
}
}
expected_output_metadata = expected_outputs_metadata[x]
request = get_model_metadata(model_name='resnet',
version=versions[x])
response = stub.GetModelMetadata(request, 10)
input_metadata, output_metadata = model_metadata_response(
response=response)
print(output_metadata)
assert model_name == response.model_spec.name
assert expected_input_metadata == input_metadata
assert expected_output_metadata == output_metadata
def test_get_model_metadata_rest(self, download_two_models,
start_server_multi_model):
"""
<b>Description</b>
Execute inference request using REST API interface hosting multiple
models
<b>input data</b>
- directory with 2 models in IR format
- docker image
<b>fixtures used</b>
- model downloader
- input data downloader
- service launching
<b>Expected results</b>
- response contains proper response about model metadata for both
models set in config file:
model resnet_v1_50, pnasnet_large
- both served models handles appropriate input formats
"""
print("Downloaded model files:", download_two_models)
urls = [
'http://localhost:5561/v1/models/resnet/metadata',
'http://localhost:5561/v1/models/resnet/versions/1/metadata'
]
expected_outputs_metadata = \
[{'resnet_v2_50/predictions/Reshape_1':
{'dtype': 1, 'shape': [1, 1001]}},
{'resnet_v1_50/predictions/Reshape_1':
{'dtype': 1, 'shape': [1, 1000]}}
]
for x in range(len(urls)):
print("Getting info about resnet model version:".format(urls[x]))
model_name = 'resnet'
expected_input_metadata = {
'input': {
'dtype': 1,
'shape': [1, 3, 224, 224]
}
}
expected_output_metadata = expected_outputs_metadata[x]
response = get_model_metadata_response_rest(urls[x])
input_metadata, output_metadata = model_metadata_response(
response=response)
print(output_metadata)
assert model_name == response.model_spec.name
assert expected_input_metadata == input_metadata
assert expected_output_metadata == output_metadata
def test_get_model_metadata(self, start_server_single_model_from_gc,
create_channel_for_port_single_server):
stub = create_channel_for_port_single_server
model_name = 'resnet'
out_name = 'resnet_v1_50/predictions/Reshape_1'
expected_input_metadata = {'input': {'dtype': 1,
'shape': [1, 3, 224, 224]}}
expected_output_metadata = {out_name: {'dtype': 1,
'shape': [1, 1000]}}
request = get_model_metadata(model_name='resnet')
response = stub.GetModelMetadata(request, 10)
input_metadata, output_metadata = model_metadata_response(
response=response)
print(output_metadata)
assert model_name == response.model_spec.name
assert expected_input_metadata == input_metadata
assert expected_output_metadata == output_metadata
def test_get_model_metadata_rest(self, resnet_8_batch_model_downloader,
start_server_batch_model):
print("Downloaded model files:", resnet_8_batch_model_downloader)
model_name = 'resnet'
out_name = 'resnet_v1_50/predictions/Reshape_1'
expected_input_metadata = {
'input': {
'dtype': 1,
'shape': [8, 3, 224, 224]
}
}
expected_output_metadata = {out_name: {'dtype': 1, 'shape': [8, 1000]}}
rest_url = 'http://localhost:5557/v1/models/resnet/metadata'
response = get_model_metadata_response_rest(rest_url)
input_metadata, output_metadata = model_metadata_response(
response=response)
print(output_metadata)
assert model_name == response.model_spec.name
assert expected_input_metadata == input_metadata
assert expected_output_metadata == output_metadata
def test_get_model_metadata(self, resnet_2_out_model_downloader,
create_channel_for_port_mapping_server,
start_server_with_mapping):
print("Downloaded model files:", resnet_2_out_model_downloader)
result = start_server_with_mapping
print("docker starting status:", result)
time.sleep(30) # Waiting for inference service to load models
assert result == 0, "docker container was not started successfully"
stub = create_channel_for_port_mapping_server
model_name = 'resnet_2_out'
expected_input_metadata = {
'new_key': {
'dtype': 1,
'shape': [1, 3, 224, 224]
}
}
expected_output_metadata = {
'mask': {
'dtype': 1,
'shape': [1, 2048, 7, 7]
},
'output': {
'dtype': 1,
'shape': [1, 2048, 7, 7]
}
}
request = get_model_metadata(model_name=model_name)
response = stub.GetModelMetadata(request, 10)
print("response", response)
input_metadata, output_metadata = model_metadata_response(
response=response)
assert model_name == response.model_spec.name
assert expected_input_metadata == input_metadata
assert expected_output_metadata == output_metadata
def test_get_model_metadata_rest(self, model_version_policy_models,
start_server_model_ver_policy, model_name,
throw_error):
"""
<b>Description</b>
Execute GetModelMetadata request using gRPC interface
hosting multiple models
<b>input data</b>
- directory with 2 models in IR format
- docker image
<b>fixtures used</b>
- model downloader
- input data downloader
- service launching
<b>Expected results</b>
- response contains proper response about model metadata for both
models set in config file:
model resnet_v1_50, pnasnet_large
- both served models handles appropriate input formats
"""
print("Downloaded model files:", model_version_policy_models)
print("Getting info about resnet model")
versions = [1, 2, 3]
expected_outputs_metadata = [{
'resnet_v1_50/predictions/Reshape_1': {
'dtype': 1,
'shape': [1, 1000]
}
}, {
'resnet_v2_50/predictions/Reshape_1': {
'dtype': 1,
'shape': [1, 1001]
}
}, {
'mask': {
'dtype': 1,
'shape': [1, 2048, 7, 7]
},
'output': {
'dtype': 1,
'shape': [1, 2048, 7, 7]
}
}]
expected_inputs_metadata = [{
'input': {
'dtype': 1,
'shape': [1, 3, 224, 224]
}
}, {
'input': {
'dtype': 1,
'shape': [1, 3, 224, 224]
}
}, {
'new_key': {
'dtype': 1,
'shape': [1, 3, 224, 224]
}
}]
for x in range(len(versions)):
print("Getting info about resnet model version:".format(
versions[x]))
expected_input_metadata = expected_inputs_metadata[x]
expected_output_metadata = expected_outputs_metadata[x]
rest_url = 'http://localhost:5560/v1/models/{}/' \
'versions/{}/metadata'.format(model_name, versions[x])
result = requests.get(rest_url)
print(result.text)
if not throw_error[x]:
output_json = result.text
metadata_pb = get_model_metadata_pb2.\
GetModelMetadataResponse()
response = Parse(output_json,
metadata_pb,
ignore_unknown_fields=False)
input_metadata, output_metadata = model_metadata_response(
response=response)
print(output_metadata)
assert model_name == response.model_spec.name
assert expected_input_metadata == input_metadata
assert expected_output_metadata == output_metadata
else:
assert 404 == result.status_code
def test_get_model_metadata(self, model_version_policy_models,
start_server_model_ver_policy,
create_channel_for_model_ver_pol_server,
model_name, throw_error):
"""
<b>Description</b>
Execute GetModelMetadata request using gRPC interface
hosting multiple models
<b>input data</b>
- directory with 2 models in IR format
- docker image
<b>fixtures used</b>
- model downloader
- input data downloader
- service launching
<b>Expected results</b>
- response contains proper response about model metadata for both
models set in config file:
model resnet_v1_50, pnasnet_large
- both served models handles appropriate input formats
"""
print("Downloaded model files:", model_version_policy_models)
# Connect to grpc service
stub = create_channel_for_model_ver_pol_server
print("Getting info about resnet model")
versions = [1, 2, 3]
expected_outputs_metadata = [{
'resnet_v1_50/predictions/Reshape_1': {
'dtype': 1,
'shape': [1, 1000]
}
}, {
'resnet_v2_50/predictions/Reshape_1': {
'dtype': 1,
'shape': [1, 1001]
}
}, {
'mask': {
'dtype': 1,
'shape': [1, 2048, 7, 7]
},
'output': {
'dtype': 1,
'shape': [1, 2048, 7, 7]
}
}]
expected_inputs_metadata = [{
'input': {
'dtype': 1,
'shape': [1, 3, 224, 224]
}
}, {
'input': {
'dtype': 1,
'shape': [1, 3, 224, 224]
}
}, {
'new_key': {
'dtype': 1,
'shape': [1, 3, 224, 224]
}
}]
for x in range(len(versions)):
print("Getting info about resnet model version:".format(
versions[x]))
expected_input_metadata = expected_inputs_metadata[x]
expected_output_metadata = expected_outputs_metadata[x]
request = get_model_metadata(model_name=model_name,
version=versions[x])
if not throw_error[x]:
response = stub.GetModelMetadata(request, 10)
input_metadata, output_metadata = model_metadata_response(
response=response)
print(output_metadata)
assert model_name == response.model_spec.name
assert expected_input_metadata == input_metadata
assert expected_output_metadata == output_metadata
else:
with pytest.raises(AbortionError):
response = stub.GetModelMetadata(request, 10)
def test_specific_version_rest(self, download_two_model_versions,
resnet_2_out_model_downloader, get_test_dir,
start_server_update_flow_specific):
resnet_v1, resnet_v2 = download_two_model_versions
resnet_2_out = resnet_2_out_model_downloader
dir = get_test_dir + '/saved_models/' + 'update/'
resnet_v1_copy_dir = copy_model(resnet_v1, 1, dir)
resnet_2_out_copy_dir = copy_model(resnet_2_out, 4, dir)
time.sleep(8)
# Available versions: 1, 4
print("Getting info about resnet model")
model_name = 'resnet'
out_name_v1 = 'resnet_v1_50/predictions/Reshape_1'
expected_input_metadata_v1 = {
'input': {
'dtype': 1,
'shape': [1, 3, 224, 224]
}
}
expected_output_metadata_v1 = {
out_name_v1: {
'dtype': 1,
'shape': [1, 1000]
}
}
rest_url_latest = 'http://localhost:5563/v1/models/resnet/' \
'versions/1/metadata'
response = get_model_metadata_response_rest(rest_url_latest)
input_metadata, output_metadata = model_metadata_response(
response=response)
print(output_metadata)
assert model_name == response.model_spec.name
assert expected_input_metadata_v1 == input_metadata
assert expected_output_metadata_v1 == output_metadata
rest_url = 'http://localhost:5563/v1/models/resnet/metadata'
response_latest = get_model_metadata_response_rest(rest_url)
print("response", response_latest)
input_metadata_latest, output_metadata_latest = \
model_metadata_response(response=response_latest)
rest_url_v4 = 'http://localhost:5563/v1/models/resnet/' \
'versions/1/metadata'
response_v4 = get_model_metadata_response_rest(rest_url_v4)
print("response", response_v4)
input_metadata_v4, output_metadata_v4 = model_metadata_response(
response=response_latest)
assert response_v4.model_spec.name == response_latest.model_spec.name
assert input_metadata_v4 == input_metadata_latest
assert output_metadata_v4 == output_metadata_latest
# Model status check
rest_status_url = 'http://localhost:5563/v1/models/resnet'
status_response = get_model_status_response_rest(rest_status_url)
versions_statuses = status_response.model_version_status
assert len(versions_statuses) == 2
for version_status in versions_statuses:
assert version_status.version in [1, 4]
assert version_status.state == ModelVersionState.AVAILABLE
assert version_status.status.error_code == ErrorCode.OK
assert version_status.status.error_message == _ERROR_MESSAGE[
ModelVersionState.AVAILABLE][ErrorCode.OK]
###
shutil.rmtree(resnet_2_out_copy_dir)
resnet_v2_copy_dir = copy_model(resnet_v2, 3, dir)
time.sleep(10)
# Available versions: 1, 3
rest_url = 'http://localhost:5563/v1/models/resnet/metadata'
response_latest = get_model_metadata_response_rest(rest_url)
print("response", response_latest)
input_metadata_latest, output_metadata_latest = \
model_metadata_response(response=response_latest)
rest_url_v3 = 'http://localhost:5563/v1/models/resnet/' \
'versions/3/metadata'
response_v3 = get_model_metadata_response_rest(rest_url_v3)
print("response", response_v3)
input_metadata_v3, output_metadata_v3 = model_metadata_response(
response=response_v3)
assert response_v3.model_spec.name == response_latest.model_spec.name
assert input_metadata_v3 == input_metadata_latest
assert output_metadata_v3 == output_metadata_latest
# Model status check
rest_status_url = 'http://localhost:5563/v1/models/resnet'
status_response = get_model_status_response_rest(rest_status_url)
versions_statuses = status_response.model_version_status
assert len(versions_statuses) == 3
for version_status in versions_statuses:
assert version_status.version in [1, 3, 4]
if version_status.version == 4:
assert version_status.state == ModelVersionState.END
assert version_status.status.error_code == ErrorCode.OK
assert version_status.status.error_message == _ERROR_MESSAGE[
ModelVersionState.END][ErrorCode.OK]
elif version_status.version == 1 or version_status.version == 3:
assert version_status.state == ModelVersionState.AVAILABLE
assert version_status.status.error_code == ErrorCode.OK
assert version_status.status.error_message == _ERROR_MESSAGE[
ModelVersionState.AVAILABLE][ErrorCode.OK]
###
# Available versions: 1, 3, 4
resnet_2_out_copy_dir = copy_model(resnet_2_out, 4, dir)
time.sleep(10)
rest_url_v1 = 'http://localhost:5563/v1/models/resnet/' \
'versions/1/metadata'
response_v1 = get_model_metadata_response_rest(rest_url_v1)
input_metadata_v1, output_metadata_v1 = model_metadata_response(
response=response_v1)
assert model_name == response_v1.model_spec.name
assert expected_input_metadata_v1 == input_metadata_v1
assert expected_output_metadata_v1 == output_metadata_v1
out_name_v3 = 'resnet_v2_50/predictions/Reshape_1'
expected_input_metadata_v3 = {
'input': {
'dtype': 1,
'shape': [1, 3, 224, 224]
}
}
expected_output_metadata_v3 = {
out_name_v3: {
'dtype': 1,
'shape': [1, 1001]
}
}
rest_url_v3 = 'http://localhost:5563/v1/models/resnet/' \
'versions/3/metadata'
response_v3 = get_model_metadata_response_rest(rest_url_v3)
input_metadata_v3, output_metadata_v3 = model_metadata_response(
response=response_v3)
assert model_name == response_v3.model_spec.name
assert expected_input_metadata_v3 == input_metadata_v3
assert expected_output_metadata_v3 == output_metadata_v3
expected_input_metadata_v4 = {
'input': {
'dtype': 1,
'shape': [1, 3, 224, 224]
}
}
expected_output_metadata_v4 = {
'res5c_branch2c1': {
'dtype': 1,
'shape': [1, 2048, 7, 7]
},
'res5c_branch2c2': {
'dtype': 1,
'shape': [1, 2048, 7, 7]
}
}
rest_url = 'http://localhost:5563/v1/models/resnet/metadata'
response_v4 = get_model_metadata_response_rest(rest_url)
input_metadata_v4, output_metadata_v4 = model_metadata_response(
response=response_v4)
assert model_name == response_v4.model_spec.name
assert expected_input_metadata_v4 == input_metadata_v4
assert expected_output_metadata_v4 == output_metadata_v4
# Model status check
rest_status_url = 'http://localhost:5563/v1/models/resnet'
status_response = get_model_status_response_rest(rest_status_url)
versions_statuses = status_response.model_version_status
assert len(versions_statuses) == 3
for version_status in versions_statuses:
assert version_status.version in [1, 3, 4]
assert version_status.state == ModelVersionState.AVAILABLE
assert version_status.status.error_code == ErrorCode.OK
assert version_status.status.error_message == _ERROR_MESSAGE[
ModelVersionState.AVAILABLE][ErrorCode.OK]
###
shutil.rmtree(resnet_v2_copy_dir)
shutil.rmtree(resnet_v1_copy_dir)
shutil.rmtree(resnet_2_out_copy_dir)
time.sleep(10)
def test_get_model_metadata(self, download_two_models,
start_server_multi_model,
create_channel_for_port_multi_server):
"""
<b>Description</b>
Execute inference request using gRPC interface hosting multiple models
<b>input data</b>
- directory with 2 models in IR format
- docker image
<b>fixtures used</b>
- model downloader
- input data downloader
- service launching
<b>Expected results</b>
- response contains proper response about model metadata for both
models set in config file:
model resnet_v1_50, pnasnet_large
- both served models handles appropriate input formats
"""
print("Downloaded model files:", download_two_models)
# Starting docker with ie-serving
result = start_server_multi_model
print("docker starting multi model server status:", result)
time.sleep(30) # Waiting for inference service to load models
assert result == 0, "docker container was not started successfully"
# Connect to grpc service
stub = create_channel_for_port_multi_server
print("Getting info about resnet model")
model_name = 'resnet_V1_50'
out_name = 'resnet_v1_50/predictions/Reshape_1'
expected_input_metadata = {
'input': {
'dtype': 1,
'shape': [1, 3, 224, 224]
}
}
expected_output_metadata = {out_name: {'dtype': 1, 'shape': [1, 1, 1]}}
request = get_model_metadata(model_name='resnet_V1_50')
response = stub.GetModelMetadata(request, 10)
input_metadata, output_metadata = model_metadata_response(
response=response)
print(output_metadata)
assert model_name == response.model_spec.name
assert expected_input_metadata == input_metadata
assert expected_output_metadata == output_metadata
model_name = 'pnasnet_large'
out_name = 'final_layer/predictions'
request = get_model_metadata(model_name='pnasnet_large')
response = stub.GetModelMetadata(request, 10)
input_metadata, output_metadata = model_metadata_response(
response=response)
expected_input_metadata = {
'input': {
'dtype': 1,
'shape': [1, 3, 331, 331]
}
}
expected_output_metadata = {out_name: {'dtype': 1, 'shape': [1, 1, 1]}}
print(output_metadata)
assert model_name == response.model_spec.name
assert expected_input_metadata == input_metadata
assert expected_output_metadata == output_metadata
def test_latest_version_rest(self, download_two_model_versions,
get_test_dir,
start_server_update_flow_latest):
resnet_v1, resnet_v2 = download_two_model_versions
dir = get_test_dir + '/saved_models/' + 'update/'
resnet_v1_copy_dir = copy_model(resnet_v1, 1, dir)
time.sleep(8)
print("Getting info about resnet model")
model_name = 'resnet'
out_name = 'resnet_v1_50/predictions/Reshape_1'
expected_input_metadata = {
'input': {
'dtype': 1,
'shape': [1, 3, 224, 224]
}
}
expected_output_metadata = {out_name: {'dtype': 1, 'shape': [1, 1000]}}
rest_url = 'http://localhost:5562/v1/models/resnet/metadata'
response = get_model_metadata_response_rest(rest_url)
input_metadata, output_metadata = model_metadata_response(
response=response)
print(output_metadata)
assert model_name == response.model_spec.name
assert expected_input_metadata == input_metadata
assert expected_output_metadata == output_metadata
# Model status check before update
rest_status_url = 'http://localhost:5562/v1/models/resnet'
status_response = get_model_status_response_rest(rest_status_url)
versions_statuses = status_response.model_version_status
version_status = versions_statuses[0]
assert len(versions_statuses) == 1
assert version_status.version == 1
assert version_status.state == ModelVersionState.AVAILABLE
assert version_status.status.error_code == ErrorCode.OK
assert version_status.status.error_message == _ERROR_MESSAGE[
ModelVersionState.AVAILABLE][ErrorCode.OK]
###
shutil.rmtree(resnet_v1_copy_dir)
resnet_v2_copy_dir = copy_model(resnet_v2, 2, dir)
time.sleep(10)
out_name = 'resnet_v2_50/predictions/Reshape_1'
expected_input_metadata = {
'input': {
'dtype': 1,
'shape': [1, 3, 224, 224]
}
}
expected_output_metadata = {out_name: {'dtype': 1, 'shape': [1, 1001]}}
response = get_model_metadata_response_rest(rest_url)
input_metadata, output_metadata = model_metadata_response(
response=response)
print(output_metadata)
assert model_name == response.model_spec.name
assert expected_input_metadata == input_metadata
assert expected_output_metadata == output_metadata
# Model status check after update
status_response = get_model_status_response_rest(rest_status_url)
versions_statuses = status_response.model_version_status
assert len(versions_statuses) == 2
for version_status in versions_statuses:
assert version_status.version in [1, 2]
if version_status.version == 1:
assert version_status.state == ModelVersionState.END
assert version_status.status.error_code == ErrorCode.OK
assert version_status.status.error_message == _ERROR_MESSAGE[
ModelVersionState.END][ErrorCode.OK]
elif version_status.version == 2:
assert version_status.state == ModelVersionState.AVAILABLE
assert version_status.status.error_code == ErrorCode.OK
assert version_status.status.error_message == _ERROR_MESSAGE[
ModelVersionState.AVAILABLE][ErrorCode.OK]
###
shutil.rmtree(resnet_v2_copy_dir)
time.sleep(10)
def test_update_rest_grpc(self, download_two_model_versions,
resnet_2_out_model_downloader, get_test_dir,
start_server_update_flow_specific,
create_channel_for_update_flow_specific):
resnet_v1, resnet_v2 = download_two_model_versions
resnet_2_out = resnet_2_out_model_downloader
dir = get_test_dir + '/saved_models/' + 'update/'
stub = create_channel_for_update_flow_specific
resnet_v1_copy_dir = copy_model(resnet_v1, 1, dir)
resnet_2_out_copy_dir = copy_model(resnet_2_out, 4, dir)
time.sleep(8)
# Available versions: 1, 4
print("Getting info about resnet model")
model_name = 'resnet'
out_name_v1 = 'resnet_v1_50/predictions/Reshape_1'
expected_input_metadata_v1 = {
'input': {
'dtype': 1,
'shape': [1, 3, 224, 224]
}
}
expected_output_metadata_v1 = {
out_name_v1: {
'dtype': 1,
'shape': [1, 1000]
}
}
request = get_model_metadata(model_name=model_name, version=1)
response = stub.GetModelMetadata(request, 10)
input_metadata, output_metadata = model_metadata_response(
response=response)
print(output_metadata)
assert model_name == response.model_spec.name
assert expected_input_metadata_v1 == input_metadata
assert expected_output_metadata_v1 == output_metadata
rest_url = 'http://localhost:5563/v1/models/resnet/metadata'
response_latest = get_model_metadata_response_rest(rest_url)
print("response", response_latest)
input_metadata_latest, output_metadata_latest = \
model_metadata_response(response=response_latest)
request_v4 = get_model_metadata(model_name=model_name, version=4)
response_v4 = stub.GetModelMetadata(request_v4, 10)
print("response", response_v4)
input_metadata_v4, output_metadata_v4 = model_metadata_response(
response=response_latest)
assert response_v4.model_spec.name == response_latest.model_spec.name
assert input_metadata_v4 == input_metadata_latest
assert output_metadata_v4 == output_metadata_latest
shutil.rmtree(resnet_2_out_copy_dir)
resnet_v2_copy_dir = copy_model(resnet_v2, 3, dir)
time.sleep(3)
# Available versions: 1, 3
request_latest = get_model_metadata(model_name=model_name)
response_latest = stub.GetModelMetadata(request_latest, 10)
print("response", response_latest)
input_metadata_latest, output_metadata_latest = \
model_metadata_response(response=response_latest)
rest_url = 'http://localhost:5563/v1/models/resnet/versions/3/metadata'
response_v3 = get_model_metadata_response_rest(rest_url)
print("response", response_v3)
input_metadata_v3, output_metadata_v3 = model_metadata_response(
response=response_v3)
assert response_v3.model_spec.name == response_latest.model_spec.name
assert input_metadata_v3 == input_metadata_latest
assert output_metadata_v3 == output_metadata_latest
# Available versions: 1, 3, 4
resnet_2_out_copy_dir = copy_model(resnet_2_out, 4, dir)
time.sleep(3)
rest_url = 'http://localhost:5563/v1/models/resnet/versions/1/metadata'
response_v1 = get_model_metadata_response_rest(rest_url)
input_metadata_v1, output_metadata_v1 = model_metadata_response(
response=response_v1)
assert model_name == response.model_spec.name
assert expected_input_metadata_v1 == input_metadata_v1
assert expected_output_metadata_v1 == output_metadata_v1
out_name_v3 = 'resnet_v2_50/predictions/Reshape_1'
expected_input_metadata_v3 = {
'input': {
'dtype': 1,
'shape': [1, 3, 224, 224]
}
}
expected_output_metadata_v3 = {
out_name_v3: {
'dtype': 1,
'shape': [1, 1001]
}
}
request_v3 = get_model_metadata(model_name=model_name, version=3)
response_v3 = stub.GetModelMetadata(request_v3, 10)
input_metadata_v3, output_metadata_v3 = model_metadata_response(
response=response_v3)
assert model_name == response.model_spec.name
assert expected_input_metadata_v3 == input_metadata_v3
assert expected_output_metadata_v3 == output_metadata_v3
expected_input_metadata_v4 = {
'input': {
'dtype': 1,
'shape': [1, 3, 224, 224]
}
}
expected_output_metadata_v4 = {
'res5c_branch2c1': {
'dtype': 1,
'shape': [1, 2048, 7, 7]
},
'res5c_branch2c2': {
'dtype': 1,
'shape': [1, 2048, 7, 7]
}
}
rest_url = 'http://localhost:5563/v1/models/resnet/versions/4/metadata'
response_v4 = get_model_metadata_response_rest(rest_url)
input_metadata_v4, output_metadata_v4 = model_metadata_response(
response=response_v4)
assert model_name == response_v4.model_spec.name
assert expected_input_metadata_v4 == input_metadata_v4
assert expected_output_metadata_v4 == output_metadata_v4
shutil.rmtree(resnet_v2_copy_dir)
shutil.rmtree(resnet_v1_copy_dir)
shutil.rmtree(resnet_2_out_copy_dir)
def test_get_model_metadata_rest(self, download_two_models,
start_server_multi_model):
"""
<b>Description</b>
Execute inference request using REST API interface hosting multiple
models
<b>input data</b>
- directory with 2 models in IR format
- docker image
<b>fixtures used</b>
- model downloader
- input data downloader
- service launching
<b>Expected results</b>
- response contains proper response about model metadata for both
models set in config file:
model resnet_v1_50, pnasnet_large
- both served models handles appropriate input formats
"""
print("Downloaded model files:", download_two_models)
print("Getting info about resnet model")
model_name = 'resnet_V1_50'
out_name = 'resnet_v1_50/predictions/Reshape_1'
expected_input_metadata = {
'input': {
'dtype': 1,
'shape': [1, 3, 224, 224]
}
}
expected_output_metadata = {out_name: {'dtype': 1, 'shape': [1, 1000]}}
rest_url = 'http://localhost:5561/v1/models/resnet_V1_50/metadata'
response = get_model_metadata_response_rest(rest_url)
input_metadata, output_metadata = model_metadata_response(
response=response)
print(output_metadata)
assert model_name == response.model_spec.name
assert expected_input_metadata == input_metadata
assert expected_output_metadata == output_metadata
model_name = 'pnasnet_large'
out_name = 'final_layer/predictions'
rest_url = 'http://localhost:5561/v1/models/pnasnet_large/metadata'
response = get_model_metadata_response_rest(rest_url)
input_metadata, output_metadata = model_metadata_response(
response=response)
expected_input_metadata = {
'input': {
'dtype': 1,
'shape': [4, 3, 331, 331]
}
}
expected_output_metadata = {out_name: {'dtype': 1, 'shape': [4, 1001]}}
print(output_metadata)
assert model_name == response.model_spec.name
assert expected_input_metadata == input_metadata
assert expected_output_metadata == output_metadata