InternalServerError with in-page segmentation and child-analyzer routing

Max Gonzaga 0 Reputation points
2026-11-21T23:32:10.8233333+00:00

Service: Azure AI Content Understanding

API version: 2026-06-01-preview

Completion model: gpt-6.1-sol

Studio project: a6878060-fa88-45ab-9d90-be3b84e7436a

Analyzing a single-page PDF containing several documents fails with:

InternalServerError - An unexpected error occurred.

Configuration:

  • enableSegment: true
  • allowInPageSegments: true
  • segmentPerPage: false

Categories route to custom child analyzers.

Reproduction results:

  • In-page segmentation + routing: fails.
  • Same configuration with enableLayout=true: fails.
  • In-page segmentation without child routes: succeeds.
  • Routing with in-page segmentation disabled: succeeds.
  • Child analyzers processing the same PDF directly: succeed.

Studio request ID:

2f51ed99-52c9-4d62-a950-65e2f41821ee

REST reproduction request ID:

1e39b46d-45e6-477c-bb3d-39dfcb006177

Azure Content Understanding in Foundry Tools
0 comments No comments

1 answer

Sort by: Most helpful
  1. AI answer

    2026-11-21T23:33:12.9166667+00:00

    Open an Azure support request for a reproducible service-side failure in the documented 2026-06-01-preview sub-page routing path, and use two-stage classification and extraction as the temporary workaround.

    The configuration itself is valid:

    {
      "config": {
        "enableSegment": true,
        "allowInPageSegments": true,
        "segmentPerPage": false,
        "contentCategories": {
          "CategoryA": {
            "description": "Content belonging to category A.",
            "analyzerId": "custom-child-analyzer-a"
          }
        }
      },
      "models": {
        "completion": "gpt-6.1-sol"
      }
    }
    

    For this feature, enableSegment and allowInPageSegments must both be true, while segmentPerPage must be false or omitted. Setting an analyzerId on a category is the documented mechanism for automatically forwarding that segment’s source range to the child analyzer.^1^

    Your matrix isolates the failure to the interaction between:

    • in-page segmentation; and
    • automatic child-analyzer invocation.

    The PDF, child analyzers, classification, and routing without sub-page segmentation all succeed independently. Enabling layout also does not change this boundary. The preview classifier already uses layout-based features automatically, so that setting is not a required repair for this scenario.^2^

    gpt-6.1-sol is supported with 2026-06-01-preview, so the selected completion-model family is not, by itself, an unsupported combination.^3^

    Temporary workaround

    Run the workflow in two stages:

    1. Remove each category’s analyzerId, retaining in-page segmentation.
    2. Analyze the PDF and collect each returned segment’s:
      • category
      • segmentId
      • source
      • span
      • confidence
    3. Map the returned category to the intended child analyzer in application code.
    4. Invoke that child analyzer separately, passing the segment’s source expression as its range.

    The preview response exposes source specifically for targeted sub-analyzer extraction, using a polygon expression such as:

    {
      "segmentId": "segment1",
      "category": "CategoryA",
      "confidence": 0.98,
      "source": "D(1,1.32,1.49,3.13,1.49,3.13,3.86,1.32,3.86)"
    }
    

    This reproduces the intended routing behavior without exercising the failing automatic handoff.^1^

    If production reliability is required, retain this fallback or use page-boundary routing where acceptable. Sub-page segmentation and API version 2026-06-01-preview are public-preview capabilities without an SLA and are not recommended for production workloads.^1^

    Evidence to attach to the support request

    Create the request from Azure portal → Help + support → Create a support request and include:

    • Service: Azure AI Content Understanding
    • API version: 2026-06-01-preview
    • Completion model and deployed model version
    • Azure region and Content Understanding resource name
    • Analyzer IDs for the parent classifier and affected children
    • UTC timestamps for both failures
    • Studio request ID: [GUID]
    • REST request ID: [GUID]
    • Sanitized parent-analyzer JSON
    • Sanitized analyze request and complete failed operation response
    • The four-result isolation matrix from the question
    • A minimal or securely supplied reproduction PDF

    The portal allows logs or a compressed diagnostic package to be attached. Select Yes for Advanced diagnostic information if organizational policy permits Microsoft Support to collect relevant resource diagnostics.^4^

    For 2026-06-01-preview, available diagnostics are returned automatically when polling the Operation-Location; preserve the final response body and headers. The x-ms-diagnostics header is optional for this preview API.^5^

    Do not include subscription keys, document secrets, or confidential document content in the ticket’s text fields.


    References

    1. Classify and route your data using Content Understanding (programming-language-rest)
    2. Content Understanding classification/segmentation
    3. Azure Content Understanding in Foundry Tools service quotas and limits
    4. Create an Azure support request
    5. Retrieve diagnostics from Azure Content Understanding in Foundry Tools
    AI-generated content may be incorrect. Read our transparency notes for more information.

    Was this answer helpful?

Your answer

Answers can be marked as 'Accepted' by the question author and 'Recommended' by moderators, which helps users know the answer solved the author's problem.