Share via

Azure AI Search returning "Could not complete vectorization action. The vectorization endpoint returned status code '400' (BadRequest)."

Nimesh 40 Reputation points
2026-03-27T03:45:35.69+00:00

I have Azure AI search index linked to a Azure Blob storage RAG. I have 2 replicas with 99.9% read SLA. But I get the error message below intermittently.

User's image

I am using embedding model large for my vectorization profile in Microsoft Foundry. I even tried to use managed identity instead of API keys but it doesn't allow me to switch to System managed identity.

User's image

I have a scheduled indexer which runs every 720 minutes. I ran it manually and it was successful. But still I get the same error message.

User's image

I have a developer subscription with $150 renewing every month. I have around $70 left.

Please let me know how this could be resolved.

Azure AI Search
Azure AI Search

An Azure search service with built-in artificial intelligence capabilities that enrich information to help identify and explore relevant content at scale.


Answer accepted by question author
  1. Aditya N 2,550 Reputation points Microsoft External Staff Moderator
    2026-04-01T03:03:06.25+00:00

    Hello @Nimesh

    Thank you for letting us know that you tried for another region. Please feel free to reach out to us in case of issue re-appears. Please comment below in case of any further queries.

    0 comments No comments

2 additional answers

Sort by: Most helpful
  1. Nimesh 40 Reputation points
    2026-04-01T02:59:29.4833333+00:00

    Hi, it seems this issue is intermittent only for the US region, I created a new one in the Central India region, and it works fine now

    0 comments No comments

  2. Q&A Assist
    2026-03-27T03:46:07.04+00:00
    AI generated content. This question contains an answer created with AI from Q&A Assist. Learn more

    The 400 “Could not complete vectorization action” error is returned by the embedding endpoint that Azure AI Search is calling through the vectorizer. Troubleshooting needs to focus on the vectorizer configuration and the embedding model, not the replicas or indexer schedule.

    Use these steps:

    1. Check common vectorizer errors
      • In the index definition, confirm that the vectorizer referenced in vectorSearch.profiles exactly matches a vectorizer entry in vectorizers.
      • Verify that the vector field’s dimensions property matches the output dimension of the text-embedding-3-large deployment. A dimension mismatch will cause a 400 from the embedding endpoint.
      • Ensure the fields parameter in the vector query targets a searchable vector field; otherwise queries can return empty results even if vectorization succeeded.
    2. Validate authentication and roles
      • If using API keys, confirm the key is valid for the Azure OpenAI / Foundry resource.
      • If using managed identity, the search service’s identity must have the Cognitive Services OpenAI User role on the Azure OpenAI resource so the vectorizer can call the model.
      • A 401/403 from the model will surface as a vectorizer error; correct roles or keys and retry.
    3. Check for quota and rate‑limit issues
      • Intermittent failures can be caused by Azure OpenAI quota or rate limits. Review the Azure OpenAI quotas and limits for embedding models and ensure the deployment has sufficient tokens‑per‑minute (TPM) for both indexing and query workloads.
      • If the same embedding deployment is used for both indexing and queries, consider creating separate deployments (one for indexing, one for queries) so each has its own TPM allocation.
    4. Review diagnostic logs
      • Enable diagnostic logging on the search service and run a query that reproduces the error.
      • Use the provided Kusto query to confirm vector queries and inspect AdditionalInfo for details about the vectorization call:
             OperationEvent
             | where TIMESTAMP > ago(30m)
             | where Name == "Query.Search" and AdditionalInfo["QueryMetadata"]["Vectors"] has "TextLength"
        
      • Use the correlation data to cross‑check with Azure OpenAI metrics (requests, errors, throttling) on the model deployment.
    5. Apply best practices to stabilize vectorization
      • Prefer managed identity over API keys in production once the identity and roles are correctly configured.
      • Deploy the embedding model in the same region as the search service to reduce latency and lower the chance of transient failures.
      • Monitor TPM usage and, if necessary, request higher limits or reduce batch sizes.

    If, after aligning dimensions, fixing any auth issues, and confirming quota is not exceeded, the 400 errors persist, capture the failing request’s timestamp and correlation IDs from logs and open a support case so the backend vectorization and model calls can be inspected.


    References:

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.