Analyze Traces → Latency and Performance Analysis

How to analyze latency and performance across services, operations, and single traces.

Service level

On Service Overview: the Latency, SL Latency, and Apdex tiles (with % change vs. previous) and the Latency trend chart show when performance changed. Turn on Compare with previous period to confirm.

Operation level

On Web Operations (and Databases / Messaging): sort by P95 / P99 to find tail‑latency offenders; Most Time Consuming ranks by peak latency. The Operations table exposes Min · Median · Average · Max · P95 · P99.

Accuracy: Average, counts, error rate, and Apdex are exact; P95/P99 are estimated from latency histogram buckets — treat them as directional.

Segment level (within an operation)

The operation Segment Breakdown shows which downstream call (grpc, db, external HTTP) dominates latency — often the fastest route to the bottleneck before opening a trace.

Trace level

In Trace Detail: the waterfall (longest bars), the per‑span Exclusive Duration (self time vs. waiting), and the Latency tab (per‑service share) pinpoint where a single request spent its time.

Apdex

Apdex scores requests against a threshold T: satisfied < T, tolerating < 4T. Keep the UI Apdex Threshold aligned with the Trace Proxy Threshold.