Application Metrics powered by Spans

[Dhrumil-Sentry]

A few months ago we made the difficult decision to not release our generic Metrics solution which was in Beta. We understand this has been frustrating for many users, but after thorough evaluation, we determined there were strong reasons to take this step:

• Lack of flexibility: Traditional metrics solutions require developers to predefine tags and aggregation rules, which limits their ability to drill down and analyze data retrospectively. This approach hinders effective debugging, as unexpected issues cannot be investigated beyond predefined metrics. We believe that allowing developers to query raw data flexibly is the best way to resolve complex application health issues.
• Low confidence in affordable pricing: Our goal is to offer products at price points that make them accessible to all developers. However, we were not confident in finding a near-term solution that would have achieved this for the generic metrics solution we built.

Our Path Forward

We’re excited to introduce Application Metrics, now available for a limited cohort of customers. It lets you easily aggregate and analyze span data without being restricted by predefined configurations or fixed parameters. Unlike traditional metrics setups, it gives you the flexibility to adjust your queries and groupings as needed.

What Use Cases Can Application Metrics Solve?

Application Metrics supports a variety of use cases, offering improved visibility into application health and making it easier to debug and troubleshoot errors and slowdowns.

1. Track API Latency and Identify Anomalies: Monitor the 95th percentile of time taken for API responses, grouped by customer, to identify if any users are overusing the system.
2. Monitor Payload Sizes in Background Jobs: Track the average payload size in background job processing to prevent slowdowns.
3. Track Token Usage for Third-Party APIs: Ensure that requests to external APIs like OpenAI stay within limits by tracking token usage per request which you slice by your customer segments
4. Monitor LCP (Largest Contentful Paint) for Specific Users or Pages: Monitor LCP for high-value users on the checkout page to identify page load issues that may impact revenue.

How can you use Application Metrics?

Attach attributes/tags to your spans via SDKs

You can add any custom attribute to your spans via our SDKs and use these attributes for filtering and aggregation in our UX. Here’s an example from our React SDK:

const span = Sentry.getActiveSpan();
if (span) {
  span.setAttribute("customerType", "Premium");
  // Or set multiple attributes at once:
  span.setAttributes({
    "payloadSize": 42,
    "cached": true,
  });
}

Use the new Trace Explorer to query and aggregate your spans

To support flexible, real-time queries on application metrics, we’re introducing an upgraded Trace Explorer UX (Go to Explore→Traces) making it easy to:

• Filter and aggregate span data based on attributes like duration, custom fields, and endpoints.
• Visualize high-level metrics or drill down into specific span samples.
• Switch seamlessly between viewing span aggregates and underlying trace details.

Example: You can query the p75 and p90 of LCP for your webpages and aggregate these score by custom attribute customerType

Extrapolating for Sample rates

Since we’re aggregating spans to compute your aggregates, we will automatically extrapolate your aggregations like count, percentiles etc to account for your sampling rates. This will help us show you reasonably accurate aggregates even if you apply client side sample rates

Your feedback is appreciated

We’ve rolled this feature to a limited cohort of customers. Please reach out to [email protected] to get access

We’re actively working on this querying experience so please feel free to share any feedback

• Is the filtering UX easy to use? We want to ensure you quickly find the values to filter by
• Do you need any advance features (equations, multi-queries etc)?

Your input will help us build a better experience, so please drop your comments and questions below. We’re excited to help you unlock the full potential of your span data with Sentry.