How would you test a feature that integrates a third-party service that's often down?

Clarify first

• What is the expected behavior when the third party is unavailable — fail fast, fallback to cached data, or queue and retry?
• Is there a sandbox or mock endpoint provided by the third party?
• Does the integration have a circuit breaker or timeout configured, and what are the values?
• What alerting or observability is in place for third-party failures?

Functional (when healthy)

• End-to-end integration works correctly when the third party responds normally
• Authentication to the third party works, including token rotation and key refresh
• Data flows correctly in both directions; field mapping is accurate
• Webhooks or callbacks from the third party are received and processed correctly

Failure modes

• Third party returns 5xx → does the application degrade gracefully or crash? Is the user shown a meaningful message?
• Third party times out → is there a configured timeout, or does the request hang indefinitely?
• Circuit breaker: after N consecutive failures, does the circuit open and stop hammering the third party?
• Third party returns an unexpected error code or malformed response → application handles it safely

Fallback & queue

• If a fallback exists (cached data, queued retry), does it activate when the third party fails?
• Is the user given a meaningful message instead of a generic error?
• Are queued messages processed in the correct order when the third party recovers?
• No double-processing: if a queued event is retried, is it idempotent?

Recovery

• When the third party becomes available again, does the integration resume automatically?
• Does the circuit breaker close after the third party stabilises (half-open probe requests)?

Observability

• Are third-party errors distinguishable from first-party errors in logs and APM?
• Does a third-party outage trigger an alert before users are affected?

Close: automate using WireMock or similar to simulate 5xx, timeout, and malformed response scenarios. Keep manual for the full observability check — verify that alerts fire and dashboards show the right signals during a simulated outage.