Problem context and objectives
Lead systems engineer · Metric Shard Atlas
Design storage for a metrics platform ingesting timestamped points from millions of hosts and serving recent dashboards.
The design must survive ingest 5M points/s and retention 30 days. Every box must earn its place by satisfying a requirement or containing a failure.
How you win
- 1Trace one user action from request to durable outcome
- 2Connect every component to a stated requirement
- 3Explain the losing tradeoff, not only the chosen technology
Rules and pressure
- ingest: 5M points/s
- retention: 30 days
- Horizontal write scale
- Recent range queries
New words in this mission
Open a term for a plain-language explanation.partition or shard+
One slice of a larger dataset or workload. Splitting work raises capacity, but cross-slice operations become harder.
Live architecture trace
Metric Shard Atlas: request to outcome
Host agents sends “batch” to Ingest tier. A compound key controls hotspots and retains time-bucket locality for bounded range scans.
Design storage for a metrics platform ingesting timestamped points from millions of hosts and serving recent dashboards. Follow one real user action through the complete architecture; a component is useful only when you can explain the request, state, or failure it handles.