Azure Custom Vision and Computer Vision Timeout Period

Hello @Preetham Johnson Vedala ,

Welcome to Microsoft Q&A .Thank you for reaching out to us.

Based on the currently available documentation, there is no published or customer-configurable value that defines how long Azure Custom Vision or Azure AI Vision cloud endpoints remain warm.The managed cloud services do not expose a documented warm-retention timer or an idle-to-cold transition period.

In practice, the term "warm connection" is commonly used to describe one of the following scenarios:

• A slower first request after a period of inactivity.
• Reuse of an existing HTTP/TCP connection versus creation of a new connection.
• Startup latency in self-hosted Vision containers.
• Network-related delays introduced by proxies, gateways, firewalls, or load balancers.

Because these behaviors originate from different layers of the solution, there is no single documented timeout value that applies across all scenarios.

Please check if the following steps help-

If increased latency is consistently observed after periods of inactivity, the following checks may help determine the contributing factor:

1. Compare the latency of the first request after an idle period with subsequent requests.
2. Reuse SDK or HTTP client instances rather than creating a new client for every request to reduce connection establishment overhead and take advantage of connection pooling.
3. Enable diagnostic logging and review Azure Monitor metrics to identify latency trends and determine whether delays occur during connection establishment or request processing.
4. Review network components such as proxies, VPNs, gateways, firewalls, and load balancers that may influence request latency.
5. If Vision containers are being used, review container scaling and replica configuration to determine whether startup latency is contributing to the behavior.

The following references might be helpful , please check them out

• Azure Vision documentation - Quickstarts, Tutorials, API Reference - Foundry Tools | Microsoft Learn
• Custom Vision documentation - Quickstarts, Tutorials, API Reference - Foundry Tools | Microsoft Learn
• Use prediction endpoint to programmatically test images with classifier - Custom Vision - Foundry Tools | Microsoft Learn
• HttpClient guidelines for .NET - .NET | Microsoft Learn
• Call the Image Analysis 3.2 API - Foundry Tools | Microsoft Learn
• Enable diagnostic logging - Foundry Tools | Microsoft Learn
• Retry pattern - Azure Architecture Center | Microsoft Learn

Please let us know if the response was helpful

Thank you

hi Preetham Johnson Vedala & thx for sharing urs issue here at Q&A portal,

there isn't a documented 'warm connection' timeout for either Azure AI Vision (Computer Vision) or Custom Vision. These are managed services, and Microsoft doesn't publish how long backend instances stay warm or when a request will hit a cold path. The warm/cold behavior is an implementation detail and can vary over time. If you're seeing occasional higher latency, it's usually better to measure it rather than assume a fixed timeout. Things like region, model loading, autoscaling, and transient platform conditions can all affect the first request.

If u're asking because you need predictable latency, a common approach is to send a lightweight request periodically and monitor the response times. Just keep in mind Microsoft doesn't guarantee that this will keep a specific backend instance 'warm.'

I'd be interested to know if anyone from the product team can confirm whether there is any documented warm-instance timeout for Azure AI Vision services, but as far as I'm aware, that information isn't publicly documented.

rgds,

Alex

&

If my answer was helpful pls mark it and additional thx if u follow me at Q&A portal

and at my blog https://ctrlaltdel.blog/