def redeploy_api() -> str:
api_token = connexion.request.form["api_token"]
if auth_token(api_token):
# Service Status Management
tensorflow_model_server_down = True
for proc in psutil.process_iter():
pinfo = proc.as_dict(attrs=['pid', 'name', 'create_time'])
if pinfo['name'] == "tensorflow_model_server":
log("[INFO] TensorFlow model server is up", api_token)
tensorflow_model_server_down = False
if tensorflow_model_server_down:
log("[INFO] Starting a new tensorflow_model_server", api_token)
tf_server = subprocess.Popen([
"tensorflow_model_server "
"--rest_api_port=$REST_API_PORT "
"--model_config_file_poll_wait_seconds=$CONFIG_FILE_POLL_SECONDS "
"--model_config_file=$MODEL_CONFIG_FILE > $TENSORFLOW_SERVER_LOGS_FILE 2>&1"
],
stdout=subprocess.DEVNULL,
shell=True,
preexec_fn=os.setsid)
try:
model_version = int(connexion.request.form["model_version"])
except:
model_version = None
pass
model_name = connexion.request.form["model_name"]
append = connexion.request.form["append"]
model_download_path = os.environ['MODEL_SAVE_PATH'] + "/" + model_name
# Model Versioning Management
# if model_version was not specified, it will be set by default as "the latest model_version number + 1"
if model_version is None:
if not os.path.exists(model_download_path):
deployment_status = "[ERROR] There is no model stored with this name" + model_name + ". please choose one of the saved models"
else:
listOfFile = os.listdir(model_download_path)
model_versions = []
for file in listOfFile:
file_casted = file
try:
file_casted = int(file)
model_versions.append(file_casted)
except:
pass
# check if the model directory is empty or not
if not model_versions:
model_version = 1
else:
model_versions.sort()
model_version = model_versions[-1]
version_path = model_download_path + "/" + str(
model_version)
log(
"[INFO] the version that will be redeployed is : " +
str(model_version), api_token)
else:
version_path = model_download_path + "/" + str(model_version)
if not os.path.isdir(version_path):
deployment_status = "[ERROR] This model version path doesn't exist: " + version_path + ". Please choose an already uploaded model version."
return log(deployment_status, api_token)
# Model Deployment
log(
"[INFO] Redeploying the version " + str(model_version) + " of " +
model_name, api_token)
if append == "true":
deployment_status = "[INFO] " + utils.append_version_model_service_config(
model_name, version_path, model_version)
else:
deployment_status = "[INFO] " + utils.add_version_model_service_config(
model_name, version_path, model_version)
return log(deployment_status, api_token)
else:
return log("[ERROR] Invalid token ", api_token)
def deploy_api(model_name, model_file) -> str:
model_zip_path = os.environ['MODEL_SAVE_PATH'] + str(uuid.uuid4()) + ".zip"
api_token = connexion.request.form["api_token"]
model_name = connexion.request.form["model_name"]
append = connexion.request.form["append"]
model_file = connexion.request.files["model_file"]
model_download_path = os.environ['MODEL_SAVE_PATH'] + "/" + model_name
download_model = True
# if model_version is empty, model_version will be set on None
try:
model_version = int(connexion.request.form["model_version"])
except:
model_version = None
pass
log("[INFO] Calling deploy_api", api_token)
if auth_token(api_token):
# Service Status Management
tensorflow_model_server_down = True
for proc in psutil.process_iter():
pinfo = proc.as_dict(attrs=['pid', 'name', 'create_time'])
if pinfo['name'] == "tensorflow_model_server":
log("[INFO] TensorFlow model server is already up", api_token)
tensorflow_model_server_down = False
if tensorflow_model_server_down:
log("[INFO] Starting a new tensorflow_model_server", api_token)
tf_server = subprocess.Popen([
"tensorflow_model_server "
"--rest_api_port=$REST_API_PORT "
"--model_config_file_poll_wait_seconds=$CONFIG_FILE_POLL_SECONDS "
"--model_config_file=$MODEL_CONFIG_FILE > $TENSORFLOW_SERVER_LOGS_FILE 2>&1"
],
stdout=subprocess.DEVNULL,
shell=True,
preexec_fn=os.setsid)
# Model Versioning Management
# if model_version was not specified, it will be set by default as "the latest model_version number + 1"
if model_version is None:
if not os.path.exists(model_download_path):
os.makedirs(model_download_path)
model_version = 1
else:
listOfFile = os.listdir(model_download_path)
model_versions = []
for file in listOfFile:
file_casted = file
try:
file_casted = int(file)
model_versions.append(file_casted)
except:
pass
# check if the model directory is empty or not
if not model_versions:
model_version = 1
else:
model_versions.sort()
model_version = model_versions[-1] + 1
log("[INFO] new version to be deployed : " + str(model_version),
api_token)
else:
version_path = model_download_path + "/" + str(model_version)
if os.path.isdir(version_path):
download_model = False
log(
"[WARN] This model version already exists. \
The uploaded model version will be ignored. The existing version will be deployed.",
api_token)
# Model Downloading
# if the specified model version doesn't exist in the directory,
# the zip file uploaded by the user will be downloaded
if download_model:
version_path = model_download_path + "/" + str(model_version)
log("[INFO] Downloading the new model in " + str(version_path),
api_token)
model_file.save(model_zip_path)
with zipfile.ZipFile(model_zip_path, "r") as zip_ref:
zip_ref.extractall(version_path)
os.remove(model_zip_path)
# Model Deployment
log(
"[INFO] Deploying the version " + str(model_version) + " of " +
model_name, api_token)
if append == "true":
deployment_status = utils.append_version_model_service_config(
model_name, model_download_path, model_version)
else:
deployment_status = utils.add_version_model_service_config(
model_name, model_download_path, model_version)
# print("The new tensorflow model file was deployed successfully at: ",os.environ['MODELS_PATH'], model_version)
return log("[INFO]" + deployment_status, api_token)
else:
return log("[INFO] Invalid token", api_token)