In this paper we present an end-to-end solution to define and execute reliable dynamic reconfigurations of open component-based systems while guaranteeing their continuity of service. It uses a multi-stage approach in order to deal with the different kinds of possible errors in the most appropriate way; in particular, the goal is to detect errors as early as possible to minimize their impact on the target system. Reconfigurations are expressed in a restricted, domain-specific language in order to allow different levels of static and dynamic validation, thus detecting errors before executing the reconfiguration where possible. For errors that can not be detected early (including software and hardware faults), a runtime environment provides transactional semantics to the reconfigurations.