Depending on when you’re evaluating it’s a per request overhead. You might/provably have multiple flags per request. Compared to a lambda invocation that pushes a config file to every container if it changes, it’s expensive.