SNMP vs streaming telemetry: the crash that explains the hybrid era
Streaming telemetry is where observability is heading, and for good reason. Per-interface, sub-second, structured data that SNMP polling could never give you. Everyone is moving that way.
But here is what we see across real networks: almost nobody has fully migrated. Most operators run a hybrid, some telemetry where it works, SNMP everywhere it still matters, and a long list of “we will finish the migration next quarter” that never quite arrives.
Part of that is inertia. Part of it is something more practical.
We were integrating a batch of new routers for a service provider, a full nationwide core refresh. Modern stack, IS-IS, BGP, MPLS with a mix of LDP and RSVP. Solid hardware, current generation. Everything stable.
Alongside it, the customer was trialing a move from SNMPv2 to streaming telemetry. A clean proof of concept on one router, gNMIc into Prometheus, Grafana on top, metrics only. Read-only observability, the safest thing you can try.
A few weeks later that router crashed and did a switchover. No clear cause. Then it happened again. The only difference between that box and the 20-plus identical routers around it was one thing: gNMI telemetry was enabled. They disabled it, and months later, not a single crash since.
None of this is an argument against telemetry. It is a reminder that a read-only observability feature still runs code on the box, and on newer protocol implementations that code is far less battle-tested than the SNMP stack the industry has been hammering on for 20 years. New protocols carry new bugs. That is normal. It is also exactly why “just migrate everything to telemetry” is easier said than done.
The honest trade-off:
SNMP: mature, boring, predictable. Coarse polling intervals, clunky MIBs, but it has been in production forever and rarely surprises you.
Streaming telemetry: rich, fast, structured, built for automation and modern observability. Younger, less uniformly implemented across vendors and platforms, and occasionally capable of things like the above.
The future is clearly telemetry. The present, for most networks, is a careful hybrid.
So we are curious where you actually are. Still happy with your SNMP monitoring? Fully migrated to streaming telemetry? Or stuck in the same hybrid middle as almost everyone else, waiting for the platforms to mature before you commit the core?
