Getting "Connection to ...azure.com timed out. (connect timeout=300)" when doing many calls to begin_analyze_document

Question

Getting "Connection to ...azure.com timed out. (connect timeout=300)" when doing many calls to begin_analyze_document

Mauricio Villegas 5

We are trying out azure document intelligence from python. In particular we are using client.begin_analyze_document() with model id prebuilt-read. When we are doing a load test, a large amount of requests continuously for several minutes, we are getting a timeout error. From the docs we were expecting a 429 error, instead of a timeout, see here. This is problematic because instead of a failure which could be used to control the requests rate and do retries, the calls just take a long time delaying the entire process.

Do we have wrong expectations of getting a 429 error? Or are we misconfiguring something which is why we get the timeout behavior?

We have tried both azure-ai-formrecognizer and the new azure-ai-documentintelligence which both produce the timeouts. And the stack trace we get is attached.

stacktrace-formrecognizer.txt

stacktrace-documentintelligence.txt

Thanks!

Mauricio Villegas 5 Reputation points

2024-07-01T15:53:50.0766667+00:00

The stack trace gets stripped out.
navba-MSFT 27,625 Reputation points Microsoft Employee Moderator

2024-07-02T03:24:40.7+00:00
@Mauricio Villegas Welcome to Microsoft Q&A Forum, Thank you for posting your query here!

.

If the stack trace is getting trimmed, Could you please paste the complete stacktrace in a txt file and attach it here ?

Do you have any proxy / firewall within your org which might be intercepting the request ? If Yes, could you please bypass it ? Or test from a different network and check if you encounter the same issue ?

Please update your document intelligence python package to new and updated version.

How big is the document file size ? Do you see the same issue with smaller sized files too ?

Which api-version are you using ? is it 2024-02-29-preview. ?

Which region is the DI resource deployed to?

Do you see 408 status error code at the time of the error ?

.

Awaiting your reply.
Mauricio Villegas 5 Reputation points

2024-07-02T08:45:22.3233333+00:00

If the stack trace is getting trimmed, Could you please paste the complete stacktrace in a txt file and attach it here ?

Done

Do you have any proxy / firewall within your org which might be intercepting the request ? If Yes, could you please bypass it ? Or test from a different network and check if you encounter the same issue ?

We run it from a kubernetes cluster in azure. I am not aware of any proxy or firewall that would intercept the requests.

Please update your document intelligence python package to new and updated version.

We recently starting doing these tests. We are using the latest version of the packages.

How big is the document file size ? Do you see the same issue with smaller sized files too ?

We submit single page jpeg images. I would think the size is always less than 3MB. I don't think the issue is related to the size of the images. I also did a test from my laptop, starting a high number of requests in parallel (~300), with exactly the same image, many more requests that the supposed 15 per second quota. I did not get any error or rejected requests. What happens is that some requests just take much more longer than the usual ~3-4 seconds. Locally I did not get timeouts because the timeout is 300 seconds (5 minutes) which I did not reach.

Which api-version are you using ? is it 2024-02-29-preview. ?

As I mentioned, we tried both formrecognizer and documentintelligence. The former by default I think uses 2023-07-31 and the later by default uses 2024-02-29-preview.

Which region is the DI resource deployed to?

West Europe

Do you see 408 status error code at the time of the error ?

no
navba-MSFT 27,625 Reputation points Microsoft Employee Moderator

2024-07-02T09:25:49.43+00:00

@Mauricio Villegas I have sent you a private message asking for more details. Please reply once you get a chance.
navba-MSFT 27,625 Reputation points Microsoft Employee Moderator

2024-07-03T09:45:44.6733333+00:00

@Mauricio Villegas I have shared my findings in the Private Message. Please check once you get a chance.
Thai Nguyen Dinh 0 Reputation points

2026-09-24T10:55:22.9233333+00:00

@navba-MSFT I have same problem in the same context, did you resolve this?

Your answer

Mauricio Villegas 5 Reputation points

2024-07-01T15:53:50.0766667+00:00

The stack trace gets stripped out.
navba-MSFT 27,625 Reputation points Microsoft Employee Moderator

2024-07-02T03:24:40.7+00:00

@Mauricio Villegas Welcome to Microsoft Q&A Forum, Thank you for posting your query here!

.

If the stack trace is getting trimmed, Could you please paste the complete stacktrace in a txt file and attach it here ?

Do you have any proxy / firewall within your org which might be intercepting the request ? If Yes, could you please bypass it ? Or test from a different network and check if you encounter the same issue ?

Please update your document intelligence python package to new and updated version.

How big is the document file size ? Do you see the same issue with smaller sized files too ?

Which api-version are you using ? is it 2024-02-29-preview. ?

Which region is the DI resource deployed to?

Do you see 408 status error code at the time of the error ?

.

Awaiting your reply.
Mauricio Villegas 5 Reputation points

2024-07-02T08:45:22.3233333+00:00

If the stack trace is getting trimmed, Could you please paste the complete stacktrace in a txt file and attach it here ?

Done

Do you have any proxy / firewall within your org which might be intercepting the request ? If Yes, could you please bypass it ? Or test from a different network and check if you encounter the same issue ?

We run it from a kubernetes cluster in azure. I am not aware of any proxy or firewall that would intercept the requests.

Please update your document intelligence python package to new and updated version.

We recently starting doing these tests. We are using the latest version of the packages.

How big is the document file size ? Do you see the same issue with smaller sized files too ?

We submit single page jpeg images. I would think the size is always less than 3MB. I don't think the issue is related to the size of the images. I also did a test from my laptop, starting a high number of requests in parallel (~300), with exactly the same image, many more requests that the supposed 15 per second quota. I did not get any error or rejected requests. What happens is that some requests just take much more longer than the usual ~3-4 seconds. Locally I did not get timeouts because the timeout is 300 seconds (5 minutes) which I did not reach.

Which api-version are you using ? is it 2024-02-29-preview. ?

As I mentioned, we tried both formrecognizer and documentintelligence. The former by default I think uses 2023-07-31 and the later by default uses 2024-02-29-preview.

Which region is the DI resource deployed to?

West Europe

Do you see 408 status error code at the time of the error ?

no
navba-MSFT 27,625 Reputation points Microsoft Employee Moderator

2024-07-02T09:25:49.43+00:00

@Mauricio Villegas I have sent you a private message asking for more details. Please reply once you get a chance.
navba-MSFT 27,625 Reputation points Microsoft Employee Moderator

2024-07-03T09:45:44.6733333+00:00

@Mauricio Villegas I have shared my findings in the Private Message. Please check once you get a chance.
Thai Nguyen Dinh 0 Reputation points

2026-09-24T10:55:22.9233333+00:00

@navba-MSFT I have same problem in the same context, did you resolve this?