Edge computing is a distributed computing paradigm that relies on computational resources of end devices in a network to bring benefits such as low bandwidth utilization, responsiveness, scalability and privacy preservation. Applications range from large scale sensor networks to IoT, and concern multiple domains (agriculture, supply chain, medicine. . . ). However, resource usage optimization, a challenge due to the limited capacity of edge devices, is typically handled in a centralized way, which remains an important limitation. In this paper, we propose a decentralized approach that relies on a combination of blockchain and consensus algorithm to monitor network resources and if necessary, migrate applications at run-time. We integrate our solution into an application container platform, thus providing an edge architecture capable of general purpose computation. We validate and evaluate our solution with a proof-of-concept implementation in a national cultural heritage building.