Design a telemetry pipeline that reconciles inconsistent metric names

A large fleet of client software already in the field sends telemetry to a data infrastructure team. Each record carries a telemetry_name saying what it measures (a request latency, an error count, a feature-usage event), a value, a timestamp and some context. Older client versions are still out there and report the same measurement under different or inconsistent names, and you cannot update them.

Design the system that ingests these records, maps every variant name onto one canonical metric, stores the results and lets the team query them. Cover where and when names get reconciled (at ingestion or at query time), how the mapping is kept and changed without losing history, what happens to names nobody has mapped yet, and how the pipeline scales and stays reliable as the fleet grows.