Now that Count is a supported aggregation function for histograms, it would be nice to swap over the metrics derived from traces to metric semantic conventions defined by OTEL since these aren't subject to sampling.
The main one would be http.server.request.duration which can be used to calculate throughput, error rate, etc. https://opentelemetry.io/docs/specs/semconv/http/http-metrics/#metric-httpserverrequestduration
Likewise, there is a semantic convention http.client.request.duration that could be used for displaying visualizations of outbound requests. Unfortunately, the low-cardinality path attribute is optional, but the Services tab could at least display external error per outbound host.
Now that
Countis a supported aggregation function for histograms, it would be nice to swap over the metrics derived from traces to metric semantic conventions defined by OTEL since these aren't subject to sampling.The main one would be
http.server.request.durationwhich can be used to calculate throughput, error rate, etc. https://opentelemetry.io/docs/specs/semconv/http/http-metrics/#metric-httpserverrequestdurationLikewise, there is a semantic convention
http.client.request.durationthat could be used for displaying visualizations of outbound requests. Unfortunately, the low-cardinality path attribute is optional, but the Services tab could at least display external error per outbound host.