New Relic APM and Browser correlation for pre-optimization bottleneck measurement

Optimizing Correlation for Latency Attribution

To preserve correlation visibility while managing costs, you must align the sampling rates of the Browser agent and the APM agent. The most effective configuration for pre-optimization measurement is to temporarily increase Browser sampling to 100% for a specific user segment or a controlled test window, while maintaining a standard sampling rate on the APM side.

Sampling and Retention Configuration

Correlation relies on Distributed Tracing, where the Browser agent injects a trace ID into the HTTP headers of requests. To ensure visibility without unbounded cost:

• Browser Sampling: Set to a higher percentage (or 100% for a targeted test) during the measurement phase. Since Browser data is typically lower volume than APM transactions, this is the most surgical way to ensure every backend trace has a corresponding frontend trigger.
• APM Sampling: Keep at default levels unless you are dealing with extremely high-throughput services. Distributed tracing usually captures the full span if the head (Browser) initiates the trace.
• Retention: Be aware that trace data has a shorter retention window than aggregated metrics. Perform your bottleneck analysis within the trace retention period (typically 7–30 days) to avoid missing the granular spans needed for attribution.

Impact of Mismatched Sampling

When sampling is mismatched—specifically when the Browser agent samples at a lower rate than the APM agent—you encounter fragmented traces. In this scenario, the APM agent records the server-side transaction, but because the Browser agent did not sample the initiating page view, no trace ID was passed. This results in "orphaned" APM transactions that lack client-side context, making it impossible to determine if the delay occurred during the network transit or the initial browser request.

Ensuring Consistent Latency Attribution

To accurately attribute latency to the server versus the client, follow these configuration and verification steps:

1. Enable Distributed Tracing: Ensure the "Distributed Tracing" toggle is active in both the Browser application settings and the APM agent configuration.
2. SPA Router Configuration: If using a Single Page Application (SPA), manually instrument router transitions. Automatic instrumentation often misses virtual page changes, which breaks the link between the user's perceived "page load" and the subsequent API calls.
3. Network Gap Analysis: Use the following logic to isolate the bottleneck:
   • Client-side: Time from user action to request dispatch (Browser).
   • Network: The gap between the Browser's request timestamp and the APM transaction start time (TTFB).
   • Server-side: The duration of the APM transaction itself.

Verification: Open the browser's Network tab and verify the presence of traceparent or newrelic headers in the request headers of your API calls. If these are missing, correlation will fail regardless of sampling settings.

One diagnostic detail needed: Are you using a custom API gateway or load balancer that might be stripping custom HTTP headers?