A ``new wave'' of systems research is focusing on the dual synergistic areas of reduced total-cost-of-ownership and managed/hosted online services. Many groups have proposed visions for self-managing, self-adapting, or self-healing systems; we have presented an architecture and implemented prototype that realizes some of those behaviors in a state-storage subsystem for online services. We have also attempted to illuminate one approach to self-management in the context of this work: make recovery so fast and cheap that false positives during failure detection become less important, thereby allowing the use of powerful, self-adapting, application-generic failure detection techniques such as statistical-anomaly analysis. We hope that SSM will both prove useful as a building block for future online services and encourage others working on self-managing systems to explore similar recovery-friendly designs.