Scale-to-zero can't count to one

Short summary

Scale-to-zero serverless architectures don't fit workloads requiring exactly one continuous, exclusive instance—like auth cores with leadership elections and key management. Auth systems need exactly one running always to prevent double-fired side effects, but serverless platforms only offer zero (idle) or N (under load). The authors abandoned serverless for their auth backplane and achieved cost savings through simpler infrastructure config changes instead.

• •Serverless scale-to-zero optimizes for stateless workloads but fails for systems requiring a continuous exclusive role
• •Auth cores need exactly one always-running instance to maintain leadership leases and prevent double-firing side effects
• •Cost savings came from infrastructure optimization, not platform migration—simpler config changes were sufficient

Generated with AI, which can make mistakes.