A production-style Azure integration operations platform proving that APIs, workflows, messaging, monitoring, incident recovery, and operational documentation can work together as one supportable cloud system.
API requests enter through Azure API Management and trigger a producer Logic App. The workflow reads account data from Dynamics 365 / Dataverse and publishes events to Service Bus. A separate consumer processes the queued messages, keeping the two sides decoupled.
The environment was deliberately pushed through gateway authentication failure, rate limiting, consumer outage, queue backlog, and encoded-message parsing failure. Each fault was isolated, fixed, validated, monitored, and documented.
Six operational layers form the end-to-end integration path.
Controlled ingress, backend routing, and rate limiting.
Receives the API request and publishes account events.
Provides the account records used by the integration.
Decouples producer and consumer and absorbs downstream outages.
Decodes ContentData, parses JSON, and processes queued events.
Monitors gateway failures, workflow failures, queue depth, and dead-letter state.
The strongest evidence comes from faults that were deliberately reproduced, investigated, and recovered.
APIM failed while forwarding to the Logic App backend. A direct PowerShell test returned Accepted, isolating the fault to APIM policy and callback parameters.
BACKEND HEALTHY / POLICY FIXEDThe gateway returned 429 after more than three requests in sixty seconds, proving that the configured rate-limit policy protected the backend.
GATEWAY CONTROL VALIDATEDThe consumer was deliberately disabled while APIM and the producer remained healthy. The queue reached 45 active messages and triggered the backlog alert.
45 ACTIVE / ALERT FIREDLogic App run history showed the Service Bus content was encoded. The parsing step was fixed by decoding ContentData before Parse JSON.
PAYLOAD DECODED / RUN SUCCEEDEDMonitoring covers the gateway, producer, queue, consumer, and dead-letter path so an integration problem can be detected before it becomes invisible operational debt.
Detect failed HTTP-triggered producer executions.
Detect consumer failures including parsing errors.
Detect active messages crossing the operational threshold.
Detect any dead-letter message count greater than zero.
Support policy, routing, and backend-authentication investigation.
Architecture, gateway tests, queue health, alerts, failures, recovery, and runbooks from the working environment.
A safe browser-only reconstruction of the downstream consumer outage used during the real validation.
Mirrors the validated platform incident: producer stays healthy, Service Bus absorbs the outage, monitoring fires, and the queue drains after the consumer is repaired.
Detect → triage → isolate → fix → validate → hand over.
Detect → triage → isolate → fix → validate → hand over.
Detect → triage → isolate → fix → validate → hand over.
Detect → triage → isolate → fix → validate → hand over.
Detect → triage → isolate → fix → validate → hand over.
Backend authentication, callback parameters, rate limiting, and routing policy all affect whether the integration actually works.
The producer remained healthy during consumer failure because messages could wait safely in the queue until recovery.
Consumer run history and dead-letter state still need to be checked after recovery to confirm clean end-to-end processing.
Incident knowledge becomes repeatable when symptoms, triage, remediation, evidence, and validation are documented clearly.