azureml Submitting deployment to compute taking very long

Question

Azureml is stuck on submitting deployment to compute for a very long time. how can I speed this up? is it because of cpu and memory or due to other reasons?

log output 

Running
2022-08-23 14:51:11+00:00 Creating Container Registry if not exists.
2022-08-23 14:51:11+00:00 Registering the environment.
2022-08-23 14:51:13+00:00 Use the existing image.
2022-08-23 14:51:13+00:00 Generating deployment configuration.
2022-08-23 14:51:24+00:00 Submitting deployment to compute..

code

#Define the deployment configuration
aciconfig = AciWebservice.deploy_configuration(
    cpu_cores = 1,
    memory_gb = 1,
    dns_name_label = os.environ['ACI_DNS_NAME_LABEL']
)

env = Environment.from_conda_specification("env", "../Environments/score_env.yml")

inf_conf = InferenceConfig(entry_script="score.py",environment=env)

#deploy successful  models as a web service 
webservice_name = os.environ['WEB_SERVICE_NAME']
retries = 2
while retries > 0:
    try:
        service = Model.deploy(ws, webservice_name,models_latest,inf_conf,aciconfig, overwrite=True)
        service.wait_for_deployment(True)
        print("Webservice updated")
        break

    except:
        print(service.get_logs())
        retries -= 1
        if retries == 0:
            raise


    

Answer 1

While choosing the compute cluster, we need to choose the high availability as well as the proper VCPU count and the plan.

Reasons for taking long time:

• Far region
• Less number of availability zones
• Less count of VCPU cores.
• Service plan

Follow the reproduced procedure.

Create a resource

Choose the proper compute resource, in this case using KBS Follow the below procedure for the configuration settings.

Choose the “DEV/TEST” for better results

Make it as using the high number of AZ. We can use 2 also based on the region.

Make it as 99.95% available and Standard B4ms plan for 4VCPU’s