Guaranteed Delivery — Enterprise Integration Patterns in practice, part 18 of 24. Teams often say a message is guaranteed once a connector returns success. That statement ignores producer crash windows, broker durability, consumer acknowledgement, retention, regional failure and target-system side effects.
Guaranteed Delivery means the messaging system persists a message so it survives process or machine failure and remains available until a consumer acknowledges it or policy expires it. End-to-end business guarantee requires more: reliable publication, idempotent consumption and a recoverable record of uncertain outcomes.
Guaranteed Delivery: the architecture
Write a failure table for every boundary: before send, after broker accept, during processing, after target commit and before acknowledgement. Run those tests; happy-path screenshots do not prove delivery.
Implementation on Microsoft Azure
- Service Bus provides durable brokered messaging and at-least-once delivery. Use send confirmation and a transactional outbox when database change and publication must remain consistent.
- Choose namespace tier, zone support and disaster-recovery design from the workload objective rather than assuming a managed service removes architecture decisions.
- Set retention, duplicate detection and dead-letter policy to match the actual business replay window.
Implementation on Boomi
- Boomi runtime queues can provide durable decoupling inside the runtime topology when the feature is enabled; document storage and availability assumptions for the chosen runtime.
- When Boomi uses Azure Service Bus, acknowledge only after the process reaches the intended durable outcome. Test connector behavior during runtime restart and lock expiry.
- A clustered runtime improves availability, but connectors and target operations still need idempotency and failure recovery.
Where the pattern earns its keep
The pattern creates value when it makes coupling, reliability or operational ownership explicit. Name the message boundary, define the contract and prove the failure path. Treat Azure resources and Boomi components as deployable implementations of that decision, with versioning and monitoring that follow the same business identifiers.
Stronger durability increases latency, storage and operational complexity. Not every telemetry event needs the same guarantee as a financial command.
Production checklist
- Define the business loss tolerance.
- Close producer publication gaps.
- Persist before acknowledging.
- Align retention with replay promises.
- Test process, node and regional failure.
Measure the outcome
Track end-to-end latency, successful business outcomes, retry attempts, duplicate prevention, queue or process age and unresolved failures. Platform availability is useful, but the real service level is whether the business message reached a correct, recoverable outcome inside the promised time.
Frequently asked questions
Is Guaranteed Delivery an Azure or Boomi feature?
It is an architecture pattern, not a product feature. Azure services and Boomi components implement parts of the pattern. Keeping that distinction clear lets the team compare platforms, test the intended behavior and migrate an implementation without changing the business design.
Can Azure and Boomi implement this pattern together?
Yes. A hybrid design is often strongest when one platform owns each responsibility. For example, Azure can own a durable shared channel while Boomi owns application connectivity and translation. Correlation, security, retry ownership and the recovery path must cross that boundary explicitly.
What should we prove before production?
Prove the unhappy path: dependency outage, duplicate delivery, invalid contract, timeout, restart and controlled replay. Verify the same business identifier is searchable in both platforms and that an operator can recover work without editing payloads or bypassing governance.
Primary references
- www.enterpriseintegrationpatterns.com documentation
- learn.microsoft.com documentation
- help.boomi.com documentation
Continue the series: ← Retry and Redelivery Without Creating a Retry Storm | Control Bus: Operate Integrations as One Platform →





0 Comments