Design a Metrics System
Company: Snapchat
Role: Software Engineer
Category: System Design
Difficulty: medium
Interview Round: Onsite
Design a metrics system. Begin by clarifying what kinds of metrics it serves, who queries them, and the required freshness, retention, scale, and tolerance for missing data. State a concrete baseline before describing your architecture; the source provides no fixed workload or service-level targets.
### What a Strong Answer Covers
- A metric data model with time, dimensions, units, and appropriate aggregation semantics.
- Collection, ingestion, storage, and query paths connected to the stated use case.
- Cardinality, retention, and the cost of both ingestion and broad queries.
- Duplicate or late samples, missing data, process resets, and component failures.
- A way to operate and monitor the metrics service itself.
### Follow-up Questions
- Why is averaging the reported latency percentiles from several servers usually incorrect?
- How would an accidental per-request label affect the system, and where would you control it?
Overview: Design a metrics system with clear measurement semantics, scalable ingestion and queries, controlled label cardinality, retention, and visible failure modes.