withsoon

Chapter

3 / 10

On this page

Architecture map

High-level diagram

Trace the shared data platform in one clean architecture map

Keep this section visual: one diagram, one selected-node explanation, and one clear separation between the media path and the data path.

High-level architecture

Starts at 50%. Hover into the diagram to zoom in, then fine-tune with the controls. Hover any component for the interview-ready explanation.

Explore the architecture
50%
HTTPS / gRPCAvro + schema registryreal-time fan-outanalytical fan-outCDC fan-outClient devicesTV apps, mobile, web, game consolesEdge / Open Connect CDNVideo delivery, not part of data pathMicroservices tierPlayback, recs, billing, search, A/B, UICassandra, EVCache, DynamoDB, MySQL-AuroraApache KafkaKeystone transport backbone~1M msg/sec hot topics, 4-6h retentionApache Flink20,000+ jobsenrich, join, aggregate, windowIceberg sinkFlink/Spark streaming writersExactly-once commitsCDC connectorsDebezium / DBLog / DynamoDBBack into KafkaEVCache / CassandraServingpersonalizationElasticsearchObservabilityon-callDruidReal-time OLAPQoE, live metricsS3 data lake + Apache IcebergACID, schema evolution, time travelhidden partitioning, compactionBatch processingSpark on EMR / TitusAirflow / Maestro orchestrationMetadata and governanceMetacat, Genie, lineagePII tags, IAM, table-level ACLsConsumption layerPresto/Trino, Spark SQL, RedshiftBI tools, ML features, A/B, finance

Last reviewed June 2026 Β· By Prasoon Parashar

Numbers are interview assumptions, not real Netflix internal figures.

Was this tab useful?