If different Docker Engine versions are running simultaneously in a Docker Swarm cluster, this may lead not to an obvious service outage but to a more subtle scenario: partial traffic degradation on individual nodes. In this case, the issue appeared on one of the manager nodes, Traefik started reporting an unavailable status ( health=0 ) for the router-app service, and the cause, according to the working hypothesis, was related to differences in iptables rules and overlay networking between Docker 28.1.1 and 28.2.2. On June 22, 2025, this exact scenario occurred in the production cluster of th