Consumption Logic Apps: Service Bus managed connector actions (Send message, Complete message) time out after 120 s although Service Bus completes the operation, causing duplicates and dead-lettered messages

José Gomes 20 Reputation points
2026-11-18T19:23:44.32+00:00

Multiple Consumption Logic Apps in West Europe that use the Service Bus managed connector intermittently fail on Send message and Complete the message with:

BadRequest – HTTP request failed: the server did not respond within the timeout limit.

The operation itself succeeds on the Service Bus side: messages sent by a “timed out” action are delivered to the queue. Only the response back to the Logic App never arrives within the 120-second limit. The action’s retry policy then sends the same message again, which creates duplicates.
Timed-out completes cause lock loss and dead-lettering.

All Service Bus metrics for the namespace are healthy during the incidents. We believe the delay is in the managed connector layer between the Logic Apps runtime and Service Bus.

Environment

  • Region: West Europe
  • Service Bus namespace: Standard tier, public network access = all networks (no IP restrictions, no private endpoints), IPv4 only
  • Logic Apps: Consumption, about 15 workflows affected, each with its own Service Bus API connection
  • API connection authentication: Logic Apps Managed Identity; all connections show status Ready
  • About 10 queues affected

Timeline

  • 2026-11-18, approx. 08:00–13:50 UTC: main incident; continuous timeouts across many workflows. Traffic before and after this window was processed without errors.
  • Earlier, smaller occurrences: 2026-10-29 and 2026-11-11.
  • No configuration changes were made to the namespace, API connections or workflows before the incident.

Symptoms and evidence

  1. Send message actions time out after about 120 s, but the message is delivered. With retries, the same message (same Logic App run ID and payload) arrived on the queue 2–4 times. (not always. some runs during that window were normal)
  2. Complete the message also times out (sometimes).
  3. Service Bus namespace metrics for the same period:
    • Throttled Requests: 0
    • Server Errors: 0
    • User Errors: about 1,500, spread across 10 queues, only during the incident window. We believe these result from retried completes on messages whose lock was already lost or that were already completed.
    • Resource Health: Available throughout
  4. Many workflows, each with a separate API connection, are affected at the same time. This rules out per-connection limits. Timeline
    • 2026-11-18, approx. 08:00–13:50 UTC: main incident; continuous timeouts across many workflows. Traffic before and after this window was processed without errors.
    • No configuration changes were made to the namespace, API connections or workflows before the incident.
    Symptoms and evidence
    1. Send message actions time out after about 120 s, but the message is delivered. With retries, the same message (same Logic App run ID and payload) arrives on the queue 2–4 times. Some duplicates are enqueued only 1–2 seconds apart, which is shorter than our retry interval. This may indicate retries inside the connector backend.
    2. Complete the message also times out. On the most affected queue, about 750 messages were dead-lettered in one day with MaxDeliveryCountExceeded.
    3. Service Bus namespace metrics for the same period:
      • Throttled Requests: 0
      • Server Errors: 0
      • User Errors: about 1,500, spread across 10 queues, only during the incident window. We believe these result from retried completes on messages whose lock was already lost or that were already completed.
      • Resource Health: Available throughout
    4. Many workflows, each with a separate API connection, are affected at the same time. This rules out per-connection limits.

Questions

  1. Was there an incident or degradation of the Service Bus managed connector (or the Logic Apps connector runtime) in West Europe on 2026-11-18 around 08:00–13:50 UTC, and on 2026-10-29 and 2026-11-11?
  2. Where was the time spent for these operations? Did the connector backend receive the Service Bus response, and why wasn’t it returned within 120 s?
  3. Is there a known issue with managed identity authentication in the Service Bus connector?
  4. What do you recommend to prevent this from recurring for Consumption workflows?

The service bus is the thing we expect to "always" work so this is a critical issue for us.

Azure Logic Apps
Azure Logic Apps

An Azure service that automates the access and use of data across clouds without writing code.

0 comments No comments

1 answer

Sort by: Oldest
  1. Allan Solomon Mejia 11,050 Reputation points
    2026-11-18T20:02:27.15+00:00

    Hello @José Gomes

    Your evidence is consistent with an ambiguous send outcome: Service Bus accepts the operation, but the acknowledgment doesn’t return to the caller before the managed connector times out. Microsoft documents this exact failure class when a message is committed but the acknowledgment is lost; retrying can create duplicates.

    However, the available metrics aren’t sufficient to confirm a West Europe platform incident or identify where the 120 seconds were spent. That requires Microsoft’s managed-connector backend telemetry. Check Azure Service Health > Health history for the affected subscription and UTC periods. Resolved service issues remain visible there for 90 days. If nothing appears, open an Azure support request and provide:

    • Logic App names, workflow run IDs, and correlation IDs
    • API connection resource IDs
    • Service Bus namespace and entity names
    • Exact UTC timestamps for successful and timed-out operations
    • Action inputs/outputs, retry history and tracking identifiers
    • Message IDs and lock tokens for representative examples

    Logic Apps exposes the workflow correlation ID and action retry information in run history, which Microsoft Support can correlate with backend traces.

    Managed identity is a supported authentication method for the Service Bus managed connector. The simultaneous impact across independently created connections makes an isolated RBAC or expired-credential problem less likely, but it doesn’t prove a regional connector incident.

    For recurrence mitigation:

    1. Enable Service Bus duplicate detection and supply a stable MessageId based on the business operation. Standard tier supports duplicate detection. When the same MessageId is sent again within the configured detection window, Service Bus accepts the request but discards the duplicate. The managed connector’s Send message action exposes the MessageId property.
    2. Review the Send message retry policy. Setting it to none can prevent automatic action retries, but it also removes recovery from genuine transient failures and still leaves the original send outcome unknown. Duplicate detection plus a stable ID is therefore safer than relying solely on disabled retries. Logic Apps supports default, none, fixed, and exponential policies when the operation exposes retry settings.
    3. Make downstream processing idempotent. Duplicate detection protects repeated sends only within its configured window; consumers should still reject an already-processed business-operation ID.
    4. Treat the completion failures separately. Under peek-lock, an expired or unsettled lock increases DeliveryCount. Once MaxDeliveryCount is exceeded, Service Bus moves the message to the dead-letter queue with MaxDeliveryCountExceeded. Increasing that value may provide temporary operational headroom, but it doesn’t correct the failed settlement path.
    5. For the most critical workflows, evaluate Logic Apps Standard with the built-in Service Bus connector. Consumption supports only the shared managed connector, while Standard also provides the built-in connector, which Microsoft documents as generally offering better performance and configurable send/receive timeouts.

    The documented timeout controls apply to the Standard built-in connector, so they shouldn’t be presented as a Consumption workaround.

    References:


    Help make this community better for everyone: If this answer helped or resolved your issue, please accept it or upvote it. If not, share more details in a comment so we can continue the discussion and find the right solution. Thank you.

    Was this answer helpful?

    0 comments No comments

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.