Polling for document intelligence batch analysis result is broken

[krishan-patel001]

• package: azure-ai-documentintelligence:
• version: 1.0.2:
• os: macOS Tahoe 26.5.2:
• Python Version 3.11.11:

Describe the bug
I am setting up a batch request to azure document intelligence using the async interface like the following and saving the continuation token.

batch_request = AnalyzeBatchDocumentsRequest(
    azure_blob_source=azure_blob_source,
    result_container_url=container_url_with_sas,
    result_prefix=result_prefix,
    overwrite_existing=True,  
)

poller = await self._document_intelligence_client.begin_analyze_batch_documents("prebuilt-layout", batch_request, locale="en")

continuation_token = poller.continuation_token()

Then in a queue I poll the job using the previously saved continuation token and print out all the statuses.

poller = await self._document_intelligence_client.get_analyze_batch_result(continuation_token)
print(f"Poller done: {poller.done()}")
print(f"Poller status: {poller.status()}")

result: AnalyzeBatchResult = await poller.result()
print(f"Poller succeeded count: {result.succeeded_count}")
print(f"Poller failed count: {result.failed_count}")
print(f"Poller skipped count: {result.skipped_count}")

However even when the result is present, and therefore the analysis job is completed, the done() still returns False and status() still says InProgress.

Poller done: False
Poller status: InProgress
Poller succeeded count: 5
Poller failed count: 0
Poller skipped count: 0

To Reproduce
Steps to reproduce the behavior:

1. Create a begin_analyze_batch_documents request
2. Save the continuation token
3. Run poller.done() and poller.status(), poller.done() will be false and poller.status() will be InProgress even after the images have finished transcribing. poller.result() will return the successful files even when the other two method claim it isn't complete.

Expected behavior
I expect to be able to kick off a batch process and poll it for being done and once it is complete then get the result rather than having to call the blocking result() method and waiting.

Screenshots
If applicable, add screenshots to help explain your problem.

Additional context
Add any other context about the problem here.

[github-actions]

Thank you for your feedback. Tagging and routing to the team member(s) best able to assist.

//cc: @bojunehsu @yungshinlintw

[github-actions]

🎯 Agentic Issue Triage

Summary: When reconstructing an async poller from a continuation token via get_analyze_batch_result(), the poller incorrectly reports done() = False and status() = InProgress even after the batch document analysis job has completed — yet result() succeeds and returns data.

📋 Issue Details

Package: azure-ai-documentintelligence 1.0.2
Affected API: AsyncDocumentIntelligenceClient.get_analyze_batch_result(), poller.done(), poller.status()
Scenarios:

• User kicks off batch analysis with begin_analyze_batch_documents(), saves continuation_token, then polls status in a separate context using get_analyze_batch_result(continuation_token).
• Despite poller.result() returning a fully populated AnalyzeBatchResult (with succeeded_count > 0), poller.done() returns False and poller.status() returns "InProgress".

Root ask: The author needs done() and status() to accurately reflect the true terminal state of the batch job, enabling non-blocking polling rather than relying solely on the blocking result() call.

🔎 Debugging / Reproduction Notes

The inconsistency — result() succeeds while done() == False — strongly suggests the reconstructed poller's state machine is not transitioning to a terminal state from the polling response.

Suggested investigation steps:

1. Inspect get_analyze_batch_result(): verify how the continuation token is deserialized and whether the initial polling state is set correctly vs. begin_analyze_batch_documents().
2. Capture the raw HTTP polling response when the batch is complete and check the service's status field value (e.g., "Succeeded" vs. "succeeded"). Compare against the SDK's terminal-state detection logic.
3. Compare the _polling_method used in get_analyze_batch_result() vs. begin_analyze_batch_documents() — a mismatch in the terminal-state set could explain why done() never transitions to True.

🏷️ Label Confidence

• Category: Client — azure-ai-documentintelligence is a data-plane SDK; the issue involves async polling behavior in the client library, not ARM/management-plane.
• Service: Document Intelligence — the package is explicitly azure-ai-documentintelligence (under /sdk/documentintelligence/).
• Confidence: High — package and service are unambiguously identified; multiple quality reference issues (e.g., Memory Usage Spike When Analyzing 5MB PDF with Document Intelligence SDK #38972, The Enum name should be DocumentContentFormat #39061, Link to list of regions where compatible version is availbel does not go to a list of regions. #38946) confirm the Client + Document Intelligence label combination.

👥 Owner Routing

• Matched CODEOWNERS entry: # ServiceLabel: %Document Intelligence (line 230) — service label matches Document Intelligence.
• AzureSdkOwners: @bojunehsu, @yungshinlintw
• ServiceOwners: @bojunehsu, @yungshinlintw
• Routing action: Assigned @bojunehsu (randomly selected from two AzureSdkOwners); added needs-team-attention label. ServiceOwners were not pinged because AzureSdkOwners take priority.
• Scan notes: Scanned bottom-to-top; Durable Task Scheduler entry (line 233) did not match; Document Intelligence block (lines 229–231) was the first matching entry.

Generated by Agentic Triage for issue #47972 · 42.9 AIC · ⌖ 9.79 AIC · ⊞ 9.6K · ◷