Facebook Configuration Management: Rolling Release with GateKeeper

Facebook advocates early and frequent software releases to obtain rapid feedback, but acknowledges that frequent deployments increase the risk of bugs affecting the site.

The GateKeeper platform enables rolling releases by allowing engineers to control feature activation through online configuration switches, progressing from internal "dogfood" testing to staged exposure of 1%, 10%, and finally 100% of users.

If a problem is detected during a rollout, the offending feature can be disabled instantly via a configuration flag without changing any source code.

GateKeeper’s UI lets users add or remove constraints, adjust probability thresholds, and modify parameters such as specific user IDs, while the underlying logic is stored as JSON in Configerator and compiled into a Boolean decision tree at runtime.

The internal switch logic resembles disjunctive normal form (DNF) with sampling, allowing flexible expression of constraints like employee status or device type, and can be extended with custom key‑value store calls (e.g., laser() ) to integrate external data for feature gating.

Multi‑level switches are also supported, where a top‑level toggle can enable or disable entire groups of subordinate switches.

Overall, GateKeeper provides a highly flexible yet potentially error‑prone mechanism for managing large‑scale feature rollouts, balancing agility with the need for careful configuration governance.