Retry logic looks harmless in isolation. When a gateway, application service and SDK each apply their own retry policy, a single failed operation can multiply into a dependency outage. That interaction is the starting point for my retry-storm analysis.
