← Overview

Netflix · Cinder / Hollow 2026

Data quality by default

Moving dataset safety checks out of scattered application configuration and into a managed platform capability.

My role
Platform engineering and cross-team rollout
Outcome
437 production namespaces running platform-wide validators
  1. Configure validators
  2. Validate publish cycle
  3. Announce valid version
Validators can block announcement of a bad dataset before consumers pick it up.

The problem

Dataset validators existed, but adoption depended on application teams configuring them. Configuration was scattered across application code and properties, and the platform lacked a reliable inventory of coverage.

The design

I took over the platform side of an existing project to centralize validator configuration. I added authorized configuration APIs, supported additional validator types, and consolidated configuration that had been split across Cinder and HollowService.

Two checks—null primary keys and duplicate primary keys—became defaults for new namespaces. Teams could change managed configuration without redeploying their producer applications.

Rolling out safer defaults

I partnered with PME on migration and adoption. Existing datasets were introduced to checks in warn-only mode before enforcement, so teams could investigate violations without immediately blocking production publishes.

As of late July 2026, 437 namespaces ran the platform-wide validators. The broader campaigns moved 42 applications' circuit-breaker configuration out of code and into the UI.

Background & references

Public documentation for the systems discussed here.