Netflix shifts to versioned caching at scale for global metadata consistency
Netflix engineers detail the company's internal shift from demand-filled caching to versioned caching architectures, specifically highlighting the Hollow and OODM platforms. These systems leverage immutable snapshots and atomic cutovers to manage petabyte-scale metadata and improve consistency across their global streaming infrastructure.
Key Takeaways
- Hollow architecture compresses 20 TB datasets down to 0.5 GB to fit in local JVM heaps
- OODM platform manages petabyte-scale data movement by decoupling preparation from serving
- Netflix serves over 1 billion requests per second (RPS) via 120,000 Hollow client nodes
- Proposed Valkey native primitives include atomic logical database swaps and binary blob installs
Why It Matters
The shift from demand-filled to versioned caching addresses the persistent challenge of 'thundering herds' and partial data states in massive distributed systems. For the streaming industry, this move secures referential integrity across global catalogs, ensuring billion-scale request volumes don't trigger cascading outages during metadata updates. By proposing native versioning support in the Valkey ecosystem, Netflix is signaling a standard for high-performance, immutable infra that reduces operational overhead by eliminating complex invalidation logic. Watch for a formal Valkey RFC or plugin release that implements these atomic swap primitives.
Additional Context
The technical pivot toward versioned caching coincides with Netflix's broader strategy to formalize data interoperability. In June 2025, per Netflix Tech Blog, the company unveiled its Unified Data Architecture (UDA), a system designed to project business entities consistently across APIs and pipelines. This move aims to eliminate terminology drift and broken references that occur when microservices redefine core concepts independently. The UDA and versioned caching platforms like OODM work in tandem to ensure that as data moves from offline compute to online serving, its semantic meaning remains unchanged.
Operationally, Netflix continues to manage one of the world's most aggressive caching environments. According to industry reports from September 2025, the company's EVCache infrastructure handles over 400 million operations per second across 22,000 server instances, managing roughly 14.3 petabytes of data. While traditional demand-filled patterns remain vital for transient data like session information, the engineers' recent presentation at the Unlocked Conference in July 2026 highlights a strategic preference for versioned immutability for catalog and recommendation facts.
In the broader open-source ecosystem, the proposal to integrate versioning primitives into Valkey reflects the growing influence of the project since it was forked from Redis. In July 2026, Valkey released version 9.1, focusing on security and high-performance threading, per Linux Foundation reports. Netflix's interest in native versioning primitives—such as atomic clusterwide swaps—could potentially shift Valkey's role from a simple cache to a more sophisticated source-of-truth for stateful streaming services.
Read full article at youtube.com
Enjoy our coverage?
Add StreamingMeme as a preferred source on Google to see more of our streaming news at the top of your Search results.
Add as preferred source