Introducing Application Metrics: Track the Signal, See the Spike, Jump to the Trace

Why It Matters

By preserving rich context in metrics, teams can diagnose production issues faster and reduce downtime, turning obscure bugs into actionable alerts that tie directly to user‑impacting events.

Key Takeaways

• Application Metrics store full events with user, region, project attributes.
• Metrics integrate with traces, allowing one‑click navigation from spikes to logs.
• Three metric types: counter, distribution, gauge, all customizable via SDK.
• No extra dependencies; enabled with a single line in recent Sentry SDKs.
• Used to locate a rare Session Replay bug affecting seven users.

Pulse Analysis

Observability platforms have long struggled to balance granularity with scalability. Traditional logs provide detail but are noisy, while spans offer structured insight yet often miss outliers due to sampling. High‑cardinality metrics bridge this gap, delivering precise, queryable signals without sacrificing performance. Sentry’s Application Metrics embraces this philosophy by persisting each event with its full attribute set, enabling developers to slice data by user, region, or project—capabilities typically reserved for log‑level tools.

The new feature integrates seamlessly into existing Sentry SDKs; a single method call activates metric collection, eliminating the need for sidecar services or additional libraries. Developers can choose from counters for simple tallies, distributions for value spreads, and gauges for real‑time state tracking. Because every metric carries its trace identifier, a sudden spike becomes a launchpad into the broader debugging workflow, linking directly to related logs, errors, and performance traces. This unified view accelerates root‑cause analysis and supports automated alerting based on business‑critical thresholds.

Sentry illustrated the practical impact by diagnosing a Session Replay failure that manifested only when more than 1,000 video segments loaded. By instrumenting a distribution metric with project and replay IDs, the team identified seven affected sessions, reproduced the issue, and deployed a fix within minutes. For organizations, this translates to faster incident resolution, lower mean‑time‑to‑recovery, and the ability to monitor key product KPIs—such as payment declines or queue depths—with the same fidelity previously reserved for infrastructure telemetry. Adoption is straightforward: enable the feature, pick the most critical signal, attach relevant attributes, and let the system surface actionable insights.