HTTP Timeout issues when hitting Azure Foundry "mistral-document-ai-2512" inferencing endpoint

Question

When hitting the Foundry resource for "mistral-document-ai-2512" I am getting significant latency and networking timeouts.

This started around 9/22 at 11:30 pm EST and has remained consistent for the last 2 days.

Anyone experiencing something similar?

Answer 1

Yes—similar symptoms can occur with Azure Document Intelligence in Foundry Tools when there is a regional service issue or elevated latency in the service path.

For this scenario, the supported troubleshooting path is:

1. Check regional service status first. Review Azure status and verify whether Azure Document Intelligence shows any current impact in the resource’s host region under Geography → Products And Services → AI + Machine Learning → Azure Document Intelligence.
2. In the Azure portal, open the Document Intelligence resource and review Overview → Monitoring for Request latency over the affected time range.
3. Go to Monitoring → Metrics, select + Add metric, keep Scope and Metric Namespace unchanged, then add the Latency metric and review the aggregation that best matches the issue window.
4. If the workflow uses Azure Blob Storage, check storage latency because larger transfers can increase end-to-end time. In the storage account, open Monitoring → Insights and review both E2E latency and Server latency.
5. Check request size. Larger files and parallel processing can increase processing time. A useful normalization is latency per page. Sustained periods longer than one hour where latency per page is consistently above 15 seconds indicate an issue worth addressing.
6. If requests were submitted successfully, avoid immediately retrying failed polling patterns. The Document Intelligence API is asynchronous, results can be available for up to 24 hours, and the request ID from the POST operation can be used to retrieve results later.
7. Set an alert for recurrence: Monitoring → Alerts → Create alert rule, then choose Latency as the signal.

Important service behavior to factor in:

• Foundry Tools do not provide an SLA for latency.
• Latency varies with document size, page count, and page content.
• Variability is expected in this multitenant, asynchronous service, but multi-day sustained degradation should be investigated through monitoring and regional status checks.

If the issue remains unresolved after those checks, the documented escalation path is to contact Microsoft and include:

• model name
• version
• subscription ID
• resource ID
• timestamp and issue description
• request IDs
• logs
• sample files
• JSON output/analyze results
• training set if the issue is related to custom neural model training

---

• Troubleshooting latency issues in Azure Document Intelligence in Foundry Tools
• Troubleshooting latency issues in Azure Document Intelligence in Foundry Tools
• Troubleshooting latency issues in Azure Document Intelligence in Foundry Tools
• Troubleshooting latency issues in Azure Document Intelligence in Foundry Tools