0

I am able to deploy a Azure Machine learning prediction service in my workspace ws using the syntax

aciconfig = AciWebservice.deploy_configuration(cpu_cores=1, 
                                               memory_gb=8, 
                                               tags={"method" : "some method"}, 
                                               description='Predict something')

and then

service = Webservice.deploy_from_image(deployment_config = aciconfig,
                                       image = image,
                                       name = service_name,
                                       workspace = ws)

as described in the documentation.
However, this exposes a service publicly and this is not really optimal.

What's the easiest way to shield the ACI service? I understand that passing an auth_enabled=True parameter may do the job, but then how can I instruct a client (say, using curl or Postman) to use the service afterwards?

Davide Fiocco
  • 5,350
  • 5
  • 35
  • 72

2 Answers2

2

See here for an example (in C#). When you enable auth, you will need to send the API key in the "Authorization" header in the HTTP request:

client.DefaultRequestHeaders.Authorization = new AuthenticationHeaderValue("Bearer", authKey);

See here how to retrieve the key.

silent
  • 14,494
  • 4
  • 46
  • 86
  • If we have multiple users and/or applications that will consume the API, we shouldn't expose the same primary API key to them all. How to do this? – Martin Thøgersen May 20 '19 at 07:52
  • 1
    You could for instance put Azure API Management in front. https://learn.microsoft.com/en-us/azure/api-management/ – silent May 20 '19 at 07:57
  • Yes, I already looked at Azure API Management, but it was unclear where to pass the primary token. Can you elaborate a bit? Appreciate it. – Martin Thøgersen May 20 '19 at 08:01
  • 1
    See here how to add a HTTP header: https://learn.microsoft.com/en-us/azure/api-management/api-management-transformation-policies#SetHTTPheader So you would do user authorization in API management and then have API mgmt inject the same primary key for all requests to you ML backend service – silent May 20 '19 at 08:05
0

First, retrieve the primary and secondary keys with a (Python) syntax like

service.get_keys()

If you are using curl, the syntax may look like this: 

curl -H "Content-Type:application/json" -H "Authorization: Bearer <authKey>" -X POST -d '{"data": [some data]}' http://<url>:<port>/<method>

where <authKey> is one of the keys retrieved above.

Davide Fiocco
  • 5,350
  • 5
  • 35
  • 72