Getting 503 at Azure ML studio timegen-1

Question

Getting 503 at Azure ML studio timegen-1

Somyadeep Shrivastava 20 Microsoft Employee

Im getting 503, while calling forecast api or validation for TimeGen-1**
Failed to test model**: Request failed with status code 503

Jerald Felix 7,910 Reputation points

2025-10-22T02:58:14.29+00:00
Hello Somyadeep Shrivastava,

Greetings!

503 errors with TimeGEN-1 are usually one of two things:

1. The endpoint is cold or overloaded

If you just deployed the model or haven't used it in a while, the endpoint might be spinning up. 503 basically means "service unavailable right now" - the backend is either starting up or getting hammered with requests. Give it a minute and try again.

2. There's an actual outage or capacity issue

Check the Azure status page to see if there are any incidents in your region. TimeGEN-1 is still relatively new and sometimes specific regions run into capacity constraints.

Quick things to try:

Wait 30-60 seconds and retry your request

If you're testing, add a simple retry loop with exponential backoff

Check if your endpoint is even running - go to the endpoint details in Azure ML Studio and make sure the deployment status shows "Healthy"

Try switching to a different region if you can (sometimes one region has issues while others are fine)

If it keeps happening:

You might be hitting rate limits or there could be something wrong with your deployment. Can you share:

Which region you're deployed in?

Is this a brand new deployment or was it working before?

Are you seeing this on every request or intermittently?

Also, 503s during model testing aren't uncommon - the validation endpoint might be separate from the forecast endpoint and could have different availability. If one works and the other doesn't, that's actually useful info for troubleshooting.

Let me know what you find!

Best Regards,

Jerald Felix
Somyadeep Shrivastava 20 Reputation points Microsoft Employee

2025-10-22T03:17:03.06+00:00

Thanks for your response

the region we deployed in was eastus2 region

It was a old one like a week back, then i deleted and created a new one today after I got 503s, same issue
it's been consistent since noon

moreover, in terms of overloaded, leaving aside our source code, even the small electricity data set from nixtla throwing the same on nixtla_client_forecast() call and validation of api comes False
Jerald Felix 7,910 Reputation points

2025-10-22T04:02:07.9233333+00:00
Hi Somyadeep,

Thank you for the additional details. Since you're experiencing persistent 503 errors in the eastus2 region even with a fresh deployment and the API validation is failing, this appears to be a regional service availability issue.

Given that:

The issue persists across both old and new deployments

Even minimal workloads (small electricity dataset) are failing

The nixtla_client_forecast() validation returns false

This has been consistent since noon today

I recommend:

Try a different region - Deploy the TimeGEN-1 model in another region (like eastus, westus2, or westeurope) to see if this is specific to eastus2

Check Azure Service Health - Visit the Azure Portal > Service Health to see if there are any known incidents affecting Azure ML in eastus2

Open a support ticket - Since this appears to be a service-level issue rather than a configuration problem, you may need direct Azure support to investigate the backend service status

Could you try deploying in an alternative region and let me know if the issue persists? This will help determine if it's region-specific or account-wide.

Best regards

Answer accepted by question author

0 additional answers

Your answer

Jerald Felix 7,910 Reputation points

2025-10-22T02:58:14.29+00:00

Hello Somyadeep Shrivastava,

Greetings!

503 errors with TimeGEN-1 are usually one of two things:

1. The endpoint is cold or overloaded

If you just deployed the model or haven't used it in a while, the endpoint might be spinning up. 503 basically means "service unavailable right now" - the backend is either starting up or getting hammered with requests. Give it a minute and try again.

2. There's an actual outage or capacity issue

Check the Azure status page to see if there are any incidents in your region. TimeGEN-1 is still relatively new and sometimes specific regions run into capacity constraints.

Quick things to try:

Wait 30-60 seconds and retry your request

If you're testing, add a simple retry loop with exponential backoff

Check if your endpoint is even running - go to the endpoint details in Azure ML Studio and make sure the deployment status shows "Healthy"

Try switching to a different region if you can (sometimes one region has issues while others are fine)

If it keeps happening:

You might be hitting rate limits or there could be something wrong with your deployment. Can you share:

Which region you're deployed in?

Is this a brand new deployment or was it working before?

Are you seeing this on every request or intermittently?

Also, 503s during model testing aren't uncommon - the validation endpoint might be separate from the forecast endpoint and could have different availability. If one works and the other doesn't, that's actually useful info for troubleshooting.

Let me know what you find!

Best Regards,

Jerald Felix
Somyadeep Shrivastava 20 Reputation points Microsoft Employee

2025-10-22T03:17:03.06+00:00

Thanks for your response

the region we deployed in was eastus2 region

It was a old one like a week back, then i deleted and created a new one today after I got 503s, same issue
it's been consistent since noon

moreover, in terms of overloaded, leaving aside our source code, even the small electricity data set from nixtla throwing the same on nixtla_client_forecast() call and validation of api comes False
Jerald Felix 7,910 Reputation points

2025-10-22T04:02:07.9233333+00:00

Hi Somyadeep,

Thank you for the additional details. Since you're experiencing persistent 503 errors in the eastus2 region even with a fresh deployment and the API validation is failing, this appears to be a regional service availability issue.

Given that:

The issue persists across both old and new deployments

Even minimal workloads (small electricity dataset) are failing

The nixtla_client_forecast() validation returns false

This has been consistent since noon today

I recommend:

Try a different region - Deploy the TimeGEN-1 model in another region (like eastus, westus2, or westeurope) to see if this is specific to eastus2

Check Azure Service Health - Visit the Azure Portal > Service Health to see if there are any known incidents affecting Azure ML in eastus2

Open a support ticket - Since this appears to be a service-level issue rather than a configuration problem, you may need direct Azure support to investigate the backend service status

Could you try deploying in an alternative region and let me know if the issue persists? This will help determine if it's region-specific or account-wide.

Best regards

Answer 1

Hi Somyadeep,
You can deploy TimeGEN-1, and it will appear as deployed, but the serverless API is currently unavailable in all Azure regions, attempting to use it will result in this error. Please check again later for availability.
As you already have a Marketplace subscription for it, that is why you are not seeing the following message:
Serverless API is not available now for this model on any of the azure regions, please check again later.

*If you try it with an account that does not have a Marketplace subscription for TimeGEN-1, then you will see the above message.

Feel free to accept this as an answer.
Thankyou for reaching out to The Microsoft QNA Portal.*

Share via

Getting 503 at Azure ML studio timegen-1

0 additional answers

Your answer