Feature Toggle Dynamics in Large-Scale Systems: Prevalence, Growth, Lifespan, and Benchmarking

A new study by researcher Xhevahire Tërnava provides the first large-scale, longitudinal analysis of how feature toggles—code flags used for gradual rollouts and A/B testing—evolve and accumulate as technical debt in major software systems. The research analyzed over 4,000 toggle events across two massive codebases: Kubernetes (10 million lines of code over 8.5 years) and GitLab (5 million lines over 5 years). The findings reveal a consistent pattern where the removal of toggles significantly lags behind their creation, with a 35% lag in Kubernetes and a 13% lag in GitLab, leading to a steadily growing inventory of dormant code.

Beyond the growth rate, the study uncovered stark differences in toggle lifespan between the two organizational contexts. Toggles in Kubernetes persisted for a median of 734 days, nearly four times longer than the 185-day median lifespan in GitLab. Alarmingly, a small but persistent fraction of toggles (1.33% in Kubernetes, 0.73% in GitLab) exceeded all previously observed removal durations, effectively becoming permanent fixtures in the codebase. To help engineering teams combat this creeping debt, Tërnava proposes a practical benchmarking framework built on five key metrics, complete with empirically derived threshold zones, allowing teams to objectively assess and compare their toggle hygiene against industry data.