How to deal with 500 errors on Azure Document Intelligence APIs

Question

How to deal with 500 errors on Azure Document Intelligence APIs

Andrei Pruteanu 0

We encountered several errors like this:

File "/home/app/azure/core/polling/base_polling.py", line 803, in run raise HttpResponseError(response=self._pipeline_response.http_response, error=err) from err azure.core.exceptions.HttpResponseError: (InternalServerError) An unexpected error occurred. Code: InternalServerError Message: An unexpected error occurred. Exception Details: (InternalServerError) An unexpected error occurred. Code: InternalServerError Message: An unexpected error occurred. Target: 0

How should we deal with them? Does the API client have a retry mechanism builtin or we have to implement one as a wrapper?
Here is a more extensive log:

Traceback (most recent call last):
  File "/home/app/azure/core/polling/base_polling.py", line 788, in run
    self._poll()
  File "/home/app/azure/core/polling/base_polling.py", line 820, in _poll
    raise OperationFailed("Operation failed or canceled")
azure.core.polling.base_polling.OperationFailed: Operation failed or canceled  The above exception was the direct cause of the following exception:  Traceback (most recent call last):   File "bla1.py", line 647, in run     self.__process()   File "bla2.py", line 209, in __process     self.azure_client.extract()   File "bla.py", line 213, in extract
    azure_response = poller.result()
                     ^^^^^^^^^^^^^^^
  File "/home/app/azure/core/polling/_poller.py", line 254, in result
    self.wait(timeout)
  File "/home/app/azure/core/tracing/decorator.py", line 105, in wrapper_use_tracer
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/home/app/azure/core/polling/_poller.py", line 273, in wait
    raise self._exception  # type: ignore
    ^^^^^^^^^^^^^^^^^^^^^
  File "/home/app/azure/core/polling/_poller.py", line 188, in _start
    self._polling_method.run()
  File "/home/app/azure/core/polling/base_polling.py", line 803, in run
    raise HttpResponseError(response=self._pipeline_response.http_response, error=err) from err
azure.core.exceptions.HttpResponseError: (InternalServerError) An unexpected error occurred.
Code: InternalServerError
Message: An unexpected error occurred.
Exception Details:	(InternalServerError) An unexpected error occurred.
	Code: InternalServerError
	Message: An unexpected error occurred.
	Target: 0
	
	
	
processing.ocr_error

Andrei Pruteanu 0 Reputation points

2025-07-04T10:41:17.2866667+00:00

Does it make sense to move the model in other region at runtime? To me it doesn't.. Our users can tolerate some delay but not tens of minutes to do the move.
Manas Mohanty 6,530 Reputation points Microsoft External Staff Moderator

2025-07-04T15:51:29.04+00:00

Hi Andrei Pruteanu

Copy operation should copy a model and clone it without tempering with original model

Please test with a sample custom model before implementing in target model.

We should have model copied in multiple regions to handle regional outages.

Thank you.
Andrei Pruteanu 0 Reputation points

2025-07-04T21:18:33.0333333+00:00

Hi Manas,

Thank you for the answer! Since our clients are within EU where GDPR rules are very strict, copying data and models in regions outside of EU for redundancy is usually not an option. What do you recommend in that case? Are these outages affecting multiple regions inside EU at once?

best,
Andrei
Manas Mohanty 6,530 Reputation points Microsoft External Staff Moderator

2025-07-07T04:51:13.4666667+00:00

Hi Andrei Pruteanu

Sorry for the delay in response over weekend.

Found few products groups ticket on other customers. They seem be to be mitigated.

Does the issue still persist for you.

Thank you.
Andrei Pruteanu 0 Reputation points

2025-07-07T06:37:16.42+00:00

Hi @Manas Mohanty ,

Still haven't got an answer to my last 2 questions:
[..]
Since our clients are within EU where GDPR rules are very strict, copying data and models in regions outside of EU for redundancy is usually not an option. What do you recommend in that case?
Are these outages affecting multiple regions inside EU at once?
[..]
Best,
Andrei
Manas Mohanty 6,530 Reputation points Microsoft External Staff Moderator

2025-07-07T07:17:29.0033333+00:00

Hey Andrei Pruteanu

Sorry for skipping a question of yours by mistake.

Thank you for emphasizing that all European regions were impacted same time and you have stricter GDPR policies.

But we should have models copied to non-European model in case of wider natural disasters impacting European region as failover mechanism and part of disaster recovery.

https://learn.microsoft.com/en-us/azure/ai-services/document-intelligence/how-to-guides/disaster-recovery?view=doc-intel-4.0.0

Models will have only 24 hours retention to temporarily data used while processing and deleted immediately be deleting by the respective document analysis operation through rest api.

Document intelligence and Speech services has commitment tier option which gives more stability, discounted price and stability compared to pay as you go.

https://learn.microsoft.com/en-us/azure/ai-services/commitment-tier

Validation

The Custom model feature allows customers to build custom models from training data stored in customer’s Azure blob storage locations

The interim outputs after analysis and labeling are stored in the same location.

The trained custom models are stored in Azure storage in the same region and logically isolated with their Azure subscription and API credentials.

For all features, the input data and results are deleted within 24 hours and not used for any other purpose. For customer trained models, the customers can delete their models and associated metadata at any time by using the API.

Reference - https://learn.microsoft.com/en-us/legal/cognitive-services/document-intelligence/data-privacy-security#data-stored-by-document-intelligence

Thank you
Andrei Pruteanu 0 Reputation points

2025-07-07T11:51:22.2266667+00:00

Hi,

As I stated initially, the data (including models) cannot leave the EU by law. Does the multi-region redundancy cover that case (e.g. if models are deployed both in Ireland and France for redundancy)?

best,
Andrei
Manas Mohanty 6,530 Reputation points Microsoft External Staff Moderator

2025-07-09T07:34:47.01+00:00

Yes Andrei Pruteanu

Deploying multiple resources in Multiple European regions would still be GDPR compliant.

Reference- https://servicetrust.microsoft.com/viewpage/GDPR

I have converted my 1st comment to answer if you would like to acknowledge our efforts.

Thank you

1 answer

Your answer

Andrei Pruteanu 0 Reputation points

2025-07-04T10:41:17.2866667+00:00

Does it make sense to move the model in other region at runtime? To me it doesn't.. Our users can tolerate some delay but not tens of minutes to do the move.
Manas Mohanty 6,530 Reputation points Microsoft External Staff Moderator

2025-07-04T15:51:29.04+00:00

Hi Andrei Pruteanu

Copy operation should copy a model and clone it without tempering with original model

Please test with a sample custom model before implementing in target model.

We should have model copied in multiple regions to handle regional outages.

Thank you.
Andrei Pruteanu 0 Reputation points

2025-07-04T21:18:33.0333333+00:00

Hi Manas,

Thank you for the answer! Since our clients are within EU where GDPR rules are very strict, copying data and models in regions outside of EU for redundancy is usually not an option. What do you recommend in that case? Are these outages affecting multiple regions inside EU at once?

best,
Andrei
Manas Mohanty 6,530 Reputation points Microsoft External Staff Moderator

2025-07-07T04:51:13.4666667+00:00

Hi Andrei Pruteanu

Sorry for the delay in response over weekend.

Found few products groups ticket on other customers. They seem be to be mitigated.

Does the issue still persist for you.

Thank you.
Andrei Pruteanu 0 Reputation points

2025-07-07T06:37:16.42+00:00

Hi @Manas Mohanty ,

Still haven't got an answer to my last 2 questions:
[..]
Since our clients are within EU where GDPR rules are very strict, copying data and models in regions outside of EU for redundancy is usually not an option. What do you recommend in that case?
Are these outages affecting multiple regions inside EU at once?
[..]
Best,
Andrei
Manas Mohanty 6,530 Reputation points Microsoft External Staff Moderator

2025-07-07T07:17:29.0033333+00:00

Hey Andrei Pruteanu

Sorry for skipping a question of yours by mistake.

Thank you for emphasizing that all European regions were impacted same time and you have stricter GDPR policies.

But we should have models copied to non-European model in case of wider natural disasters impacting European region as failover mechanism and part of disaster recovery.

https://learn.microsoft.com/en-us/azure/ai-services/document-intelligence/how-to-guides/disaster-recovery?view=doc-intel-4.0.0

Models will have only 24 hours retention to temporarily data used while processing and deleted immediately be deleting by the respective document analysis operation through rest api.

Document intelligence and Speech services has commitment tier option which gives more stability, discounted price and stability compared to pay as you go.

https://learn.microsoft.com/en-us/azure/ai-services/commitment-tier

Validation

The Custom model feature allows customers to build custom models from training data stored in customer’s Azure blob storage locations

The interim outputs after analysis and labeling are stored in the same location.

The trained custom models are stored in Azure storage in the same region and logically isolated with their Azure subscription and API credentials.

For all features, the input data and results are deleted within 24 hours and not used for any other purpose. For customer trained models, the customers can delete their models and associated metadata at any time by using the API.

Reference - https://learn.microsoft.com/en-us/legal/cognitive-services/document-intelligence/data-privacy-security#data-stored-by-document-intelligence

Thank you
Andrei Pruteanu 0 Reputation points

2025-07-07T11:51:22.2266667+00:00

Hi,

As I stated initially, the data (including models) cannot leave the EU by law. Does the multi-region redundancy cover that case (e.g. if models are deployed both in Ireland and France for redundancy)?

best,
Andrei
Manas Mohanty 6,530 Reputation points Microsoft External Staff Moderator

2025-07-09T07:34:47.01+00:00

Yes Andrei Pruteanu

Deploying multiple resources in Multiple European regions would still be GDPR compliant.

Reference- https://servicetrust.microsoft.com/viewpage/GDPR

I have converted my 1st comment to answer if you would like to acknowledge our efforts.

Thank you

Answer 1

Hi Andrei Pruteanu

You can switch to other available region in case of 500 internal server error.

Regional support for all Azure services

If it is a custom model, you can use copy authorization to copy the model to another document intelligence resource (different region)

You might need to implement custom retry logic, but in case of 500 error, it is better to switch to another resource, Retries are more fruitful in case of 429 rate limit errors

    for retry_attemp in range(retry_count):
        try:
            results = await poller.result()
            success = True
        except Exception as ex:
            logger.error(f"Attempt {retry_attemp}: GET Exception {ex}")
            await asyncio.sleep(retry_sleep)
            token = poller.continuation_token()
            poller = await client.begin_analyze_document_from_url(
                None, None, continuation_token=token
            )
            continue

Reference for retry logic- https://github.com/Azure/azure-sdk-for-python/issues/35952#issuecomment-2168864017

Other areas are to check are

Permission (Cognitive Services user)
Change in network configuration (Please check you have subnet/join permission on virtual network if document intelligence source moved to virtual network)
Please check from metrics tab if you have exhausted your quota. Service limits, Troubleshoot latency issues
Any policy related blocks can be checked in activity logs or application insights.

Hope it helps

Thank you.

Share via

How to deal with 500 errors on Azure Document Intelligence APIs

1 answer

Your answer