Service mesh (Istio/Linkerd): is the operational overhead justified?

we're at a point with about 30+ microservices where we're seeing inconsistent retry logic, manual timeouts everywhere, and no standardized circuit breaking or mtls. a service mesh like istio or linkerd seems like the obvious solution to these problems, centralizing all that boilerplate at the infrastructure layer. but the operational overhead is a major concern. we're talking about additional memory, cpu, and potential latency overhead per pod, plus the complexity of managing the mesh itself. for a team that's already stretched, is the promise of consistent reliability patterns and observability worth the steep learning curve and resource costs of a service mesh? what's the tipping point where it becomes a clear win?