Understanding and Validating Database System Administration

A large number of enterprises need their commodity database systems to remain available at all times. Although administrator mistakes are a significant source of unavailability and cost in these systems, no study to date has sought to quantify the frequency of mistakes in the field, understand the context in which they occur, or develop system support to deal with them explicitly. In this paper, we first characterize the typical administrator tasks, testing environments, and mistakes using results from an extensive survey we have conducted of 51 experienced administrators. Given the results of this survey, we next propose system support to validate administrator actions before they are made visible to users. Our prototype implementation creates a validation environment that is an extension of a replicated database system, where administrator actions can be validated using real workloads. The prototype implements three forms of validation, including a novel form in which the behavior of a database replica can be validated even without an example of correct behavior for comparison. Our results show that the prototype can detect the major classes of administrator mistakes.