[translation] account for nonstandard LRO status strings

[kristapratico]

The service will return "Cancelled" (with two "l"s), "Cancelling", and "ValidationFailed".

Here are the status strings that azure-core polling currently checks to see if the operation has reached a terminal state: https://github.com/Azure/azure-sdk-for-python/blob/master/sdk/core/azure-core/azure/core/polling/base_polling.py#L52-L54

We need to check for these nonstandard statuses or the LRO will run forever.

Options:

1. Simply map the nonstandard statuses to those of the ones checked in the get_status method:

            if status:
                if status == "ValidationFailed":
                    return "Failed"
                elif status in ["Cancelled", "Cancelling"]:
                    return "Canceled"
                return status
            else:
                raise BadResponse("No status found in body")

2. Define a custom polling method that will inherit from LROBasePolling and override the methods checking for a terminal state like here in Text Analytics: https://github.com/Azure/azure-sdk-for-python/blob/master/sdk/textanalytics/azure-ai-textanalytics/azure/ai/textanalytics/_lro.py#L32

I favor option 1, but we might need to re-evaluate if it makes sense with the current error states of the service. E.g. if we map "ValidationFailed" to a status == "Failed", now we will throw an exception instead of just returning the JobStatusDetail when wait_until_done is called by the user.

[mshaban-msft]

i think we should go with the first approach it's much simpler
you can also find all service "status" responses here
https://docs.microsoft.com/en-us/rest/api/cognitiveservices/translator/documenttranslation/getoperationstatus#status
nice catch btw!

[kristapratico]

If we take option 1, then we always throw in the case of a ValidationError. Error cases 4 and 6 on this table probably shouldn't raise an exception since there isn't any error details at the batch level (user would need to check the documents to understand what happened). I'm going to bring this up in our v-team chat.

[kristapratico]

Tentative plan (may/may not make it for beta.1). If it does not make it in time then fallback plan will be to always return JobStatusDetail.

This is the expected behavior when calling wait_until_done

Service terminal state status	Job level error?	Document level error?	SDK behavior
ValidationFailed	Yes	No, returns 0 documents in list	Throw exception with job level error details
Failed	No	Yes, per document	Return JobStatusDetail, user can inquire on each document for error
Succeeded / (maybe partiallySucceeded)	No	No	Return JobStatusDetail
Cancelled/Cancelling	No	No	Return JobStatusDetail

[mshaban-msft]

@kristapratico
can you elaborate on the ValidationFailed, what is this "job level error"?
i mean what type of error should i throw in here:
https://github.com/mohamedshabanofficial/azure-sdk-for-python/blob/fe31acbe58edbbf7eebf87dc4e1896b45fc70af8/sdk/documenttranslation/azure-ai-documenttranslation/azure/ai/documenttranslation/_polling.py#L85

[kristapratico]

"Job level error" will be the error returned on the BatchStatusDetail. We'll use the error from the generated type (instead of JobStatusDetail) since we'll raise the exception before the code reaches the models wrapping methods.

In Python, we throw HttpResponseError from azure.core.exceptions. We can extract out the error code and message from the returned error and then instantiate and raise the exception like this:

error = HttpResponseError(message="({}): {}".format(error["code"], error["message"]))
raise error

Edit: Realized two things here: 1) we should raise the inner error details here if present, and 2) We need to make sure we don't raise an exception if a "Failed" status is returned from the service. Something like this could work in the _polling.py file:

    def _map_nonstandard_statuses(self, status, body):
        # type: (str, dict) -> str
        """Map non-standard statuses.

        :param str status: lro process status.
        """
        if status == "ValidationFailed":
            self.raise_error(body)
        if status == "Failed":
            # We don't throw in a "Failed" case so this just indicates job completed
            return "Succeeded"
        if status in ["Cancelled", "Cancelling"]:
            return "Canceled"
        return status

    def raise_error(self, body):
        error = body["error"]
        if body["error"].get("innerError", None):
            error = body["error"]["innerError"]
        raise HttpResponseError(
            message="({}): {}".format(error["code"], error["message"])
        )

[kristapratico]

closed with #17738