Architecture
Walk from event emitters to validation, Kafka, streaming, Bronze/Silver/Gold, warehouse, features, replay, and governance.
On this page
Layered map
Show the full Netflix data journey as one layered system map.
My architecture is layered: emit, validate, publish to Kafka, process in streaming and batch, store in Bronze/Silver/Gold, and serve BI plus features with governance around every layer.
Quick jumps
Click to jumpMonthly users
200M-250M
Docs use both 200M+ and 250M assumptions
Daily active users
80M
Used for heartbeat and Kafka math
Peak concurrency
15M
Useful for QoE and live ops workloads
Interactive layered architecture
Click any node to see its inputs, outputs, failure modes, and interview framing. Highlight paths with the overlay buttons.
Interactive architecture canvas
Click nodes to inspect. Use overlay buttons to trace replay, governance, and cost paths.
Node detail drawer
Kafka / MSK
What it does
Why it exists
Connected path
Quick jumps
Click to select a nodeLegend
Emit → validate → publish → stream → store → aggregate → serve
Walk through this sequence aloud before naming specific tools.
Defend the major architectural choices
Know why each design decision was made — interviewers probe these directly.
Decision
Kafka vs Kinesis
Kafka when you want rich ecosystem, strong replay patterns, and open-source portability.
Decision
Flink vs Spark Streaming
Flink for complex low-latency event-time stateful logic like sessionization and watch-time.
Decision
Spark batch vs dbt
Use Spark for heavy cleansing and big transforms; use dbt for SQL-first Silver-to-Gold modeling where warehouse/lakehouse engines are strong.
Decision
Iceberg vs Delta vs Hudi
Iceberg is a strong default here because of open multi-engine access, snapshots, deletes, and table evolution.
Decision
Redshift vs Athena / Trino
Warehouse for curated BI and finance; Athena/Trino for open ad hoc lake queries.
Decision
Redis vs DynamoDB for online features
Redis for ultra-hot counters and ephemeral features, DynamoDB for durable keyed feature serving.
Narrate the architecture in one flow
Cover layers in order: producers, ingestion, Kafka, processing, lakehouse, serving, governance.
Say This In Interview
My architecture is layered: emit, validate, publish to Kafka, process in streaming and batch, store in Bronze/Silver/Gold, and serve BI plus features with governance around every layer.
30-second answer
FastMy architecture is layered: emit, validate, publish to Kafka, process in streaming and batch, store in Bronze/Silver/Gold, and serve BI plus features with governance around every layer.
2-minute answer
BalancedWalk from event emitters to validation, Kafka, streaming, Bronze/Silver/Gold, warehouse, features, replay, and governance. Focus on PIPELINES, 10 min, Show the full Netflix data journey as one layered system map..
Deep answer
DeepMy architecture is layered: emit, validate, publish to Kafka, process in streaming and batch, store in Bronze/Silver/Gold, and serve BI plus features with governance around every layer. Then go layer by layer through PIPELINES, 10 min, Show the full Netflix data journey as one layered system map., including trade-offs, failure handling, and why this shape is right for the consumer SLAs.
Last reviewed June 2026 · By Prasoon Parashar
Numbers are interview assumptions, not real Netflix internal figures.