Azure AI content understanding REST API timing out after 60 seconds

Question

To whom it may concern,

In the last couple of weeks, my team and I have been trying to create and use analyzers created with the Azure AI Content Understanding REST API. We are able to create analyzers no problem but when they have more than 20 fields, we get the following error when trying to use the model to conduct a field extraction

Analysis failed: {'id': '3882b893-9b0f-4c02-98c0-1a679430b0b7', 'status': 'Failed', 'error': {'code': 'InternalServerError', 'message': 'An unexpected error occurred.'}, 'result': {'analyzerId': 'dev-llmripper-cityofchicago-default-00001', 'apiVersion': '2025-05-01-preview', 'createdAt': '2025-07-14T14:07:51Z', 'warnings': [], 'contents': []}}

Last week, we were able to run models that had 50 fields but not 70+. After doing some investigation, it appears that the Internal Server Error occurs anytime the request is being processed for more than 60 seconds. This appears to be a known preview-service bug according to ChatGPT's research:

Evidence	What it shows
Microsoft Q&A thread “Content Understanding Analyzer not working due to internal server error” – several users hit a 500 after ~60 s while the same files succeed in the portal. The accepted answer confirms it’s a backend issue and asks customers to open a support ticket.(Microsoft Learn)	Backend aborts the operation; no timeout value is given.
Microsoft Q&A thread “Content Understanding Analyzer not working due to internal server error” – several users hit a 500 after ~60 s while the same files succeed in the portal. The accepted answer confirms it’s a backend issue and asks customers to open a support ticket.([Microsoft Learn](https://learn.microsoft.com/en-us/answers/questions/2282480/content-understanding-analyzer-not-working-due-to?utm_source=chatgpt.com"Content Understanding Analyzer not working due to internal server ..."))	Backend aborts the operation; no timeout value is given.
Microsoft Q&A thread “Content Understanding API Taking Excessive Time (Timeout Issue)” – API calls sometimes fail after 120 s even when the portal finishes in 5-10 s. MS staff again point to transient backend problems and recommend retry logic.([Microsoft Learn](https://learn.microsoft.com/en-us/answers/questions/2287751/azure-content-understanding-api-taking-excessive-t"Azure Content Understanding API Taking Excessive Time (Timeout Issue) - Custom Analyzer - Microsoft Q&A"))	Different customers observe different cut-offs (60 s, 120 s) → no fixed limit.
Service-limits article – lists file size, pages, audio/video length, max ops/min etc. but contains no “operation duration” field.([Microsoft Learn](https://learn.microsoft.com/en-us/azure/ai-services/content-understanding/service-limits"Service quotas and limits - Content Understanding - Azure AI services	Microsoft Learn"))	If a time limit existed it would appear here, as it does for other Cognitive Services.

How the operation is supposed to work

POST …/analyzers/{id}:analyze is asynchronous and should return in < 1 s with a 202 Accepted and an Operation-Location header.Microsoft Learn

You then poll GET …/analyzerResults/{resultId} (1 call/s or slower). The job can legally stay in Running for many minutes; no official upper bound is documented.Microsoft Learn

If the backend aborts, you receive status:"Failed" with an error object, or occasionally a 500 if it crashes before producing the error payload (the bug you hit).

Why you hit 64 seconds

• Preview instability – Content Understanding (2025-05-01-preview) still runs on an internal infrastructure that sometimes kills long-running jobs; the cut-off differs by region and load (60 – 120 s in reports). How the operation is supposed to work
  1. POST …/analyzers/{id}:analyze is asynchronous and should return in < 1 s with a 202 Accepted and an Operation-Location header.Microsoft Learn
  2. You then poll GET …/analyzerResults/{resultId} (1 call/s or slower). The job can legally stay in Running for many minutes; no official upper bound is documented.Microsoft Learn
  3. If the backend aborts, you receive status:"Failed" with an error object, or occasionally a 500 if it crashes before producing the error payload (the bug you hit).
  Why you hit 64 seconds
  • Preview instability – Content Understanding (2025-05-01-preview) still runs on an internal infrastructure that sometimes kills long-running jobs; the cut-off differs by region and load (60 – 120 s in reports).

It seems the issue is similar to this one: https://learn.microsoft.com/en-us/answers/questions/2282480/content-understanding-analyzer-not-working-due-to?utm_source=chatgpt.com

We've been able to successfully use the REST API when there is less than 20 fields. We've tried adding an exponential retry backoff into the app logic but that also return 500 errors for each operation attempted when the analyzer was more than 20 fields.

Any guidance on solutions or alternative approaches to try would be greatly appreciated:)