The Gap Between Signals
Metrics tell you how much and traces tell you why, but historically they lived apart. You would see a latency spike on a chart, then guess which traces caused it. Exemplars close that gap.
What an Exemplar Is
An exemplar is a sample data point attached to a metric that includes the trace id of a request that contributed to it. When a latency histogram records a slow bucket, it can stash the trace id of one of those slow requests.
Why It Changes Debugging
- One click drill down: from a spike on a dashboard straight to a concrete trace.
- No guessing: the exemplar is a real request in that bucket, not an estimate.
- Cheap: only a few exemplars per bucket, not one per request.
The workflow becomes notice a spike, click an exemplar, land on the trace, read the critical path. Metrics point and traces explain, joined automatically.
Key idea
Exemplars attach a sample trace id to a metric data point, letting you jump from an aggregate spike directly to a concrete trace that caused it.