Common Debugging

Do not add null checks everywhere as the first action. That may stop one crash while converting a violated data contract into silent empty behavior, hiding the root cause and potentially corrupting results.

For the urgent outage, mitigate safely while investigating: capture the failing stack trace and request/context identifiers, identify the affected release and input shape, and use an existing rollback, feature-flag disablement, traffic reduction, or known-good artifact if one is available and approved. Add narrowly scoped, structured observability at the failing boundary—not production debug prints—and preserve the evidence.

State one hypothesis, such as “the upstream response omits `items` after release X.” Test it with a representative production payload or a production-like replay and compare the runtime value/type before the read. If the value is legitimately optional, add a boundary-level validation and explicit empty/error state at that contract boundary. If it is required, fail with a clear handled error and repair the producer, sequencing, or deployment mismatch. Never swallow the exception without a defined recovery path.

After the cause is proven, deploy the smallest fix, then verify the original crash reproduction, valid and malformed inputs, dependent workflows, and error-rate recovery. Remove temporary instrumentation and document the root cause, mitigation, and rollback plan.