Metrics

Send counters, gauges, and distributions from your code to track application health and drill down into related traces, logs, and errors.

With Sentry Metrics, you can send counters, gauges, and distributions from your code to track application health signals like email.sent, checkout.failed, or queue.depth. What makes Sentry's Metrics unique is that every metric event is trace-connected, meaning you can click directly into the related traces, logs, and errors when something looks off.

Unlike traditional metrics tools that only show if something's changed, Sentry's trace-connected metrics let you understand why it changed. When a checkout.failed counter spikes, you can click into a sample and see the exact trace, spans, logs, and errors that produced that metric event.

Not everything in your application generates an error or requires full tracing. Metrics are perfect for tracking:

  • Business events: checkout.started, checkout.failed, payment.declined, refund.issued
  • Application health: email.sent, email.failed, job.processed, job.retried
  • Resource utilization: queue.depth, pool.in_use, cache.hit_rate
  • User behavior: search.zero_results, feature.enabled, configuration.loaded

To get started with Metrics, navigate to the Getting Started page and select your SDK from the list.

Sentry supports three types of metrics:

Counters track how many times something happened. Each increment adds to a cumulative total.

Aggregation methods: sum, per_second, and per_minute.

Use cases: events that occur over time

  • email.sent - Track total emails sent
  • checkout.failed - Track failed checkout attempts
  • api.request - Track API calls

Gauges set a specific value at a point in time, like a snapshot. The latest value replaces the previous one.

Aggregation methods: min, max, avg, per_second, and per_minute.

Use cases: current state or level

  • queue.depth - Current number of items in queue
  • pool.in_use - Active connections in a pool
  • memory.usage - Current memory consumption

Distributions record numeric values to compute statistical aggregates.

Aggregation methods: p50, p75, p95, p99, avg, sum, min, max, count, per_second, and per_minute.

Use cases: values that vary and need statistical analysis

  • cart.amount_usd - Purchase amounts for revenue tracking
  • query.duration_ms - Query latency for performance monitoring
  • file.size_bytes - File sizes for storage analysis

The Metrics Explorer provides two main views:

The Aggregates tab shows trends and totals across any attributes you've added to your metrics. You can:

  • View time-series charts showing metric trends
  • Group by custom attributes (e.g., sum(metrics_value) grouped by email_type)
  • Apply filters to drill down into specific segments
  • Compare multiple time periods

Use this view to spot trends, identify anomalies, and understand aggregate behavior across your application.

The Samples tab shows individual metric events with direct links to their traces. This is where trace-connected metrics shine:

  • See each individual metric event as it occurred
  • Click any sample to open the full trace waterfall
  • View related spans, logs, and errors from the same moment
  • Understand the exact context that produced each metric

When you notice a spike or dip in the Aggregates view, switch to Samples to drill into specific occurrences and find the root cause.

Sentry's metrics enable a seamless debugging workflow:

  1. Spot the anomaly: Notice a spike in your Aggregates view (e.g., checkout.failed increased 3x)
  2. Open samples: Switch to the Samples tab to see individual metric events
  3. Pick a trace: Click on a sample to open the full trace that produced it
  4. Find the cause: Examine spans for slow queries, check logs for error messages, view related exceptions
  5. Fix the issue: With full context in hand, identify and resolve the root cause

Instead of "metric alert → guesswork," debugging becomes a single flow: Metric spike → sample → trace → root cause.

Each metric event automatically includes:

FieldDescription
nameThe metric identifier (e.g., checkout.failed)
kindThe metric type (counter, gauge, or distribution)
valueThe numeric value
trace_idLinks to the active trace when the metric was recorded
span_idLinks to the specific span within the trace
unit (optional)Unit of measurement (e.g., usd, ms, bytes)
custom attributesAny key-value pairs you add for grouping and filtering

You'll soon be able to:

  • Create alert rules based on metric queries
  • Add metric visualizations to dashboards
  • Set up notifications when metrics cross thresholds
  • Save common queries for quick access

Metrics are best for application and code-based health signals — numeric indicators of whether your application logic is behaving as expected.

Good candidates for metrics:

  • Business KPIs tied to code execution
  • Application health indicators
  • Resource utilization tracking
  • User action counts
  • Success/failure rates

Not ideal for metrics:

  • Infrastructure metrics (CPU, memory, disk) - use APM tools
  • Log aggregation - use Logs
  • Full request tracing - use Traces

Your Sentry metrics can be leveraged with AI agents and tooling for debugging and automated analysis:

  • Sentry CLI: Provides command-line access to your metrics data for feeding into AI tools and scripts
  • Sentry MCP Server: Enables natural language queries and deep integration with AI tools like Claude, Cursor, and VS Code using the Model Context Protocol
  • Seer: Sentry's AI debugging agent can use metrics alongside traces, logs, and errors to provide intelligent issue analysis

We'd love to hear your feedback:

  • Are the SDK APIs easy enough to send your desired metrics?
  • Any feedback about the Metrics Explorer UX - what could make this better?
  • Any default metrics you'd like Sentry to send automatically?

Share your thoughts on GitHub or email us at feedback-metrics@sentry.io.

Was this helpful?
Help improve this content
Our documentation is open source and available on GitHub. Your contributions are welcome, whether fixing a typo (drat!) or suggesting an update ("yeah, this would be better").