Cheap Recovery: A Key to Self-Managing State

Abstract

Cluster hash tables (CHTs) are a key persistent-storage component of many large-scale Internet services due to their high performance and scalability. We show that a correctly-designed CHT can also be as easy to manage as a farm of stateless servers. Specifically, we trade away some consistency to obtain reboot-based recovery that is simple, maintains full data availability, and only has modest impact on performance. This simplifies management in two ways. First, it simplifies failure detection by lowering the cost of acting on false positives, allowing us to use simple but aggressive statistical techniques to quickly detect potential failures and node degradations; even when a false alarm is raised or when rebooting will not fix the problem, attempting recovery by rebooting is relatively non-intrusive to system availability and performance. Second, it allows us to re-cast online repartitioning as failure plus recovery, simplifying dynamic scaling and capacity planning. These properties make it possible for the system to be continuously self-adjusting, a key property of self-managing, autonomic systems.