deploy model and expose model as web service via azure machine learning + azuremlsdk in R

Question

I am trying to follow this post to deploy a "model" in Azure.

A code snipet is as follows and the model, which is simply a function adding 2 numbers, seems to register fine. I don't even use the model to isolate the problem after 1000s of attempts as this scoring code shows:

library(jsonlite)

init <- function()
{
  message("hello world")
  
  function(data)
  {
    vars <- as.data.frame(fromJSON(data))
    prediction <- 2
    toJSON(prediction)
  }
}

Should be fine shouldn't it? Any way I run this code snippet:

r_env <- r_environment(name = "basic_env")
inference_config <- inference_config(
  entry_script = "score.R",
  source_directory = ".",
  environment = r_env)

aci_config <- aci_webservice_deployment_config(cpu_cores = 1, memory_gb = 0.5)

aci_service <- deploy_model(ws, 
                            'xxxxx', 
                            list(model), 
                            inference_config, 
                            aci_config)

wait_for_deployment(aci_service, show_output = TRUE)

Which produces this (after a looooong time):

Running.....................................................................
Failed
Service deployment polling reached non-successful terminal state, current service state: Failed
Operation ID: 14c35064-7ff4-46aa-9bfa-ab8a63218a2c
More information can be found using '.get_logs()'
Error:
{
  "code": "AciDeploymentFailed",
  "statusCode": 400,
  "message": "Aci Deployment failed with exception: Error in entry script, RuntimeError: Error in file(filename, \"r\", encoding = encoding) : , please run print(service.get_logs()) to get details.",
  "details": [
    {
      "code": "CrashLoopBackOff",
      "message": "Error in entry script, RuntimeError: Error in file(filename, \"r\", encoding = encoding) : , please run print(service.get_logs()) to get details."
    }
  ]
}

It does not tell me much. Not sure how to debug this further? How can I run this:

print(service.get_logs())

and where please? Guess this is a Python artifact? Any other input very much welcome.

PS:

At this point in time, I have my suspicion that the above R entry file definition is not what is expected these days. Looking at the Python equivalent taken from here:

import json

def init():
    print("This is init")

def run(data):
    test = json.loads(data)
    print(f"received data {test}")
    return f"test is {test}"

Would something like this not be more suitable (tried it without success).

library(jsonlite)

init <- function()
{
    message("hello world")
}

init <- function()
{
    return(42)
}

Answer 1

Great to see people putting the R SDK through it's paces!

The vignette you're using is obviously a great way to get started. It seems you're almost all the way through without a hitch.

Deployment is always tricky, and I'm not expert myself. I'd point you to this guide on troubleshooting deployment locally. Similar functionality exists for the R SDK, namely: local_webservice_deployment_config().

So I think you change your example to this:

deployment_config <- local_webservice_deployment_config(port = 8890)

Once you know the service is working locally, the issue you're having with the ACI webservice becomes a lot easier to narrow down.