withsoon
withsoon/System Design/Netflix Data EngineeringMODELING

Storage / Lakehouse

Show S3 layout, Iceberg responsibilities, replayability, compaction, schema evolution, and freshness by layer.

View Backend Track

Chapter

8 / 17

On this page

Bronze / Silver / Gold

MODELINGChapter 88 min

Make Bronze, Silver, and Gold feel like a living operating model.

Bronze is immutable raw truth, Silver is cleaned and trusted, and Gold is curated for business use, so replay and correction stay possible without corrupting official metrics.

MODELING8 minMake Bronze, Silver, and Gold feel like a living operating model.

Quick jumps

Click to jump

Monthly 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

Bronze / Silver / Gold

Make the medallion layers feel operational

The lakehouse must explain immutability, trust, and official reporting visually.

Medallion layer explorer

Bronze

Inputs

  • Raw JSON / Avro events
  • CDC envelopes
  • Partner files
  • CDN / service logs

Rules

  • Do not mutate source truth
  • Partition by ingestion date / hour
  • Hot retention for replay, cold retention for long-term audit
  • Restricted raw access because payloads may still carry sensitive fields
Storage layout

Explain storage paths, versions, and retention

Users should see why Iceberg snapshots and S3 layout matter for replay and recovery.

Event backbone

AWS

MSK

Open source

Kafka + Schema Registry

Why here

Durable fan-out, replay, and independent consumers.

Streaming compute

AWS

Kinesis Data Analytics / EMR on EKS

Open source

Flink

Why here

Stateful stream processing, watermarks, and session logic.

Batch compute

AWS

EMR / Glue

Open source

Spark + dbt

Why here

Large joins, reconciliation, and official daily truth.

Lakehouse format

AWS

S3 + Glue Catalog

Open source

Iceberg

Why here

Schema evolution, snapshots, time travel, and controlled backfills.

Serving

AWS

Redshift / Athena

Open source

Trino / Pinot

Why here

Separate BI, ad hoc, and real-time OLAP workloads.
Say this

Use a clean lakehouse framing line

Keep Bronze, Silver, and Gold responsibilities clear.

Say This In Interview

Bronze is immutable raw truth, Silver is cleaned and trusted, and Gold is curated for business use, so replay and correction stay possible without corrupting official metrics.

Last reviewed June 2026 Β· By Prasoon Parashar

Numbers are interview assumptions, not real Netflix internal figures.

Was this tab useful?