Coordinated atomic (CA) action is a widely used approach to structuring concurrent activities in distributed embedded applications. Recovery from failures is one of its major concerns. Despite the intuitive simplicity, reasoning about failure recovery in CA actions is rather complex because of the interaction between various system components. This paper proposes an approach based on timed CSP for modelling CA actions and reasoning about different forms of interaction between system components, addressing issues such as failure modes, means of fault-tolerance and fail-safe mechanisms