Comparisons
Two tools, two questions. Neither column wins — what decides is which question you actually have in front of you.
Metrics vs Traces
Metrics tell you *that* something is wrong across all requests; traces tell you *where* the time went in one request. Neither substitutes for the other, and teams that buy only one spend years compensating for the gap.
Trends, rates, alerting and "is this normal?" across every request
Why any individual request was slow, and anything not pre-aggregated
Cheap per request; cost explodes with label cardinality
"Are we slower than yesterday, and for which endpoint?"
Locating latency across service boundaries in a specific slow request
Aggregate behaviour, unless you sample carefully and reason statistically
Expensive per request; usually sampled, which risks missing the tail
"Which span consumed 1.8 of these 2.4 seconds?"
| Dimension | Metrics | Traces |
|---|---|---|
| Granularity | Aggregated across all requests | One request, end to end |
| Cost driver | Cardinality of labels | Volume and sampling rate |
| Finds the tail | Shows that a tail exists | Shows what is in it — if sampling kept it |
| Alerting | The natural fit | Rarely alerted on directly |
| Cross-service | Per-service, joined by hand | Joined by construction |