A retail chain ingests clickstreams and POS events at 40K per second for fraud alerts within seconds, plus nightly inventory joins over terabytes and Salesforce contact sync for dashboards. Exactly-once stream handling and SQL warehousing are needed. Which analytics pipeline meets real-time and batch needs?

The correct pipeline succeeds because MSK durably buffers 40K events per second with partitioned replay, Flink applies stateful processing with checkpointed exactly-once guarantees to raise fraud alerts within seconds, Redshift provides managed SQL warehousing for terabyte-scale nightly inventory joins and dashboard serving, and AppFlow supplies native SaaS sync for Salesforce contacts. The Kinesis Data Streams with Lambda design handles throughput, warehousing, and CRM sync elegantly but violates exactly-once semantics because Lambda event-source mappings are at-least-once and retries duplicate fraud notifications without idempotent sinks. The MSK with Flink and Redshift design using Database Migration Service for CRM violates SaaS integration principles because DMS performs database CDC replication, whereas Salesforce requires API-based extraction with OAuth and field mapping provided by AppFlow. The MSK with Flink and DynamoDB Accelerator design preserves streaming and CRM sync but violates warehousing principles because DAX accelerates key-value lookups and cannot execute complex SQL joins over terabytes economically. An exam caveat is to map each requirement to a service capability rather than choosing familiar services, especially distinguishing streaming delivery guarantees and SaaS versus database movement. A practical check is to load-test partition throughput, verify Flink checkpointing with transactional sinks, explain nightly join plans in Redshift, and validate Salesforce AppFlow connections and incremental schedules.