The　world　of　connected　devices　has　led　to　the　rise　of　the　Internet　of　Things　paradigm,　where　applications　rely　on　multiple　devices,　gathering　and　sharing　data　across　highly　heterogeneous　networks.　The　variety　of　possible　mechanisms,　protocols,　and　hardware　has　become　a　hindrance　in　the　development　of　architectures　capable　of　addressing　the　most　common　IoT　use　cases,　while　abstracting　services　from　the　underlying　communication　subsystem.　Moreover,　the　world　is　moving　toward　new　strict　requirements　in　terms　of　timeliness　and　low　latency　in　combination　with　ultra-high　availability　and　reliability.　Thus,　future　IoT　architectures　will　also　have　to　support　the　requirements　of　these　cyber-physical　applications.　In　this　regard,　edge　computing　has　been　presented　as　one　of　the　most　promising　solutions,　relying　on　the　cooperation　of　nodes　by　moving　services　directly　to　end　devices　and　caching　information　locally.　Therefore,　in　this　article,　we　propose　a　modular　and　scalable　architecture　based　on　lightweight　virtualization.　The　provided　modularity,　combined　with　the　orchestration　supplied　by　Docker,　simplifies　management　and　enables　distributed　deployments,　creating　a　highly　dynamic　system.　Moreover,　characteristics　such　as　fault　tolerance　and　system　availability　are　achieved　by　distributing　the　application　logic　across　different　layers,　where　failures　of　devices　and　micro-services　can　be　masked　by　this　natively　redundant　architecture,　with　minimal　impact　on　the　overall　system　performance.　Experimental　results　have　validated　the　implementation　of　the　proposed　architecture　for　on-demand　services　deployment　across　different　architecture　layers.