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

our organization has about 30+ microservices, and we're struggling with inconsistent retry policies, timeouts, and a complete lack of circuit breaking between them. debugging cascading failures is a nightmare, and mTLS is currently a manual, painful process. a service mesh, specifically istio or linkerd, seems like it could solve all these problems. the promise of centralized policy enforcement, automatic mTLS, traffic management, and observability out of the box is very attractive. however, the operational overhead seems substantial. injecting a sidecar into every pod means increased memory and cpu consumption, added network latency for every hop, and a whole new layer of complexity to debug when things go wrong. the learning curve for istio, in particular, looks steep. for those who've deployed a service mesh, was the operational overhead justified by the benefits? did you see a measurable improvement in reliability or reduced debugging time? what are the hidden costs or common pitfalls that aren't immediately obvious when you start looking at these tools?