Cloud-native applications consist of highly specialized and decoupled services that can be deployed, scaled and managed independently. Maintaining such applications available is a complex task for operators, because software defects and other kinds of faults can be challenging to diagnose and repair to quickly resume operations. Autonomic service operation is therefore a promising approach. However, there are risks associated to guaranteeing safe autonomic actuation, which must be managed. This paper discusses the challenges identified in the context of the development of a platform for autonomic service operation and describe the software architecture of the platform. Results show mean times to detect, diagnose and repair failures in the order of tens of seconds.