The present disclosure relates to techniques for managing client computing systems, such as a client distributed-computing system. In one embodiment, a desired state of the client distributed-computing system and a current state of the client distributed-computing system are received. Measurement data generated by a plurality of components of the client distributed-computing system is received. The measurement data is processed using one or more analytical or statistical techniques to generate distilled data. Based on one or more policies and the desired state, the distilled data and the current state are analyzed to determine one or more actions predicted to converge the current state of the client distributed-computing system towards the desired state. One or more control commands corresponding to the one or more actions are provided to the client distributed-computing system. The one or more control commands cause the client distributed-computing system to perform the one or more actions.