Netflix · RAW Hollow 2025–2026
Fast datastore provisioning
Making a new in-memory datastore available quickly enough to provision inline with a developer workflow.
- My role
- Co-development, authorization, API extensions, and lifecycle testing
- Outcome
- ~15 minutes → ~20 seconds to provision a namespace
- Claim a warm namespace
- Secure and validate
- Migrate to dedicated resources
The problem
On-demand infrastructure creation made RAW Hollow provisioning a roughly 15-minute operation. That was too slow for workflows that needed to create a datastore as part of bringing up a runtime.
The design and my contribution
The team added programmatic provisioning and a fast path using pre-warmed namespaces. A namespace is temporarily assigned to a user before moving to dedicated resources in the background.
I implemented authorization for claimed namespaces, validation and status API improvements, client-library extensions, and provisioning observability. I also co-authored target-namespace bootstrap and post-cutover health checks. My teammate owned the warm pool, claiming and cutover mechanics, and migration pipeline.
Making the contract dependable
I owned the routine end-to-end harness that exercises provisioning and teardown. This made the integration more credible for other teams relying on the API rather than driving provisioning manually.
The fast path reduced provisioning to about 20 seconds, meeting the requirement to complete within a minute.
Background & references
Public documentation for the systems discussed here.