Observability

Definition

Observability is the ability to understand a system’s internal state from the telemetry it emits. Its four core data types are captured by the acronym MELT: Metrics, Events, Logs, and Traces. New Relic’s model is a concrete reference implementation of these ideas.


Core Ideas

MELT data types

  • Metrics — aggregated numeric measurements over time.
  • Events — discrete occurrences with arbitrary attributes; you can report custom events, and turn high-cardinality events into metrics to make them cheaper to query and chart at scale.
  • Logs — timestamped text records.
  • Traces — the path of a request across services.

Distributed tracing

A trace follows one request across service boundaries. It’s composed of spans — each span represents one operation within the trace. Agents emit span events; the backend stores contextual metadata describing the relationships and timing calculations between spans. Make a high-value transaction a key transaction to monitor/alert on it differently.

Alerting model

A layered vocabulary:

  • Threshold — defines what counts as a violation.
  • Condition — a monitored data source plus its thresholds.
  • Policy — a group of one or more conditions.
  • Violation — occurs when a data source crosses a condition’s threshold.
  • Incident / Issue — an issue is a collection of one or more incidents.

Infrastructure & real-user monitoring

  • Infrastructure — CPU, memory, disk, network sampled (e.g. every 5s) into SystemSample events; InfrastructureEvent records state deltas.
  • Browser / RUM — tracks Core Web Vitals: LCP (loading, target <2.5s), FID (interactivity, <100ms), CLS (visual stability, <0.1). JS error tracking also matters for SEO, since crawlers increasingly run JavaScript.

Relationships


References

  • New Relic — MELT data types
  • New Relic — Alerts concepts and workflow
  • New Relic — Distributed tracing details
  • New Relic — PageViewTiming / Core Web Vitals