Approaches presented herein enable deploying a distributed machine learning framework in an edge computing environment. More specifically, a status of a connection between a computing system and an edge node of a plurality of edge nodes is monitored. At least one server node and a group of worker nodes from the plurality of edge nodes are identified based on the status. A path for distributing the training data to the worker nodes is determined based on the status. The training data from the edge node to the worker nodes is distributed via the path.