A redundant path does not automatically create a resilient service. Recovery depends on the interaction between topology, convergence, endpoint behaviour, and the process tolerance for interruption.

Define the failure budget

Before choosing a protocol, ask how much packet loss and recovery time each process can tolerate. A supervisory application and a motion-control cell do not share the same failure budget.

Test the complete chain

Validate power loss, link loss, device failure, upstream routing loss, and restoration. Record both network convergence and application recovery; the second number is usually the one operations cares about.

Good resilience is measurable, rehearsed, and visible—not merely drawn as a second line on the diagram.