A food delivery company needs live driver ETAs from 20K Kafka events per second and nightly SaaS order dumps for BI. Analysts query curated data in Redshift with sub-second dashboards. Ingestion must be managed without custom servers. Which analytics pipeline supports real-time MSK streaming with batch SaaS ingestion?

The correct design uses Amazon MSK to ingest 20K Kafka events per second with managed Apache Flink for stateful low-latency processing to derive live driver ETAs, uses Amazon AppFlow for serverless SaaS API ingestion of nightly order dumps without managing servers, and consolidates curated datasets in Amazon Redshift for sub-second BI dashboards. The Kinesis-based design with Flink and AppFlow into Redshift suits many streaming workloads but violates source fidelity because the facts require native Kafka protocol, topics and consumer semantics on MSK rather than a Kinesis Data Streams substitute. The MSK with Flink design that syncs SaaS data via Database Migration Service into Redshift suits database replication but fails on ingestion type because DMS performs database CDC and migrations, not SaaS API extraction with OAuth, incremental pulls and object mapping that AppFlow provides. The MSK with Flink and AppFlow design that serves from an S3 lake with Athena reflects a valid lakehouse pattern but fails the serving requirement because Athena over S3 provides flexible ad hoc analysis rather than the provisioned curated warehouse performance needed for consistent sub-second dashboards on Redshift. As an exam caveat, map each service to its source and serving specialty instead of assuming any integration or analytics engine is interchangeable. As a practical check, verify MSK partition throughput and Flink checkpointing for ETA accuracy, AppFlow schedule and throttling for nightly dumps, and Redshift distribution, sort keys and materialized views against dashboard latency.