SigNoz catches 150% latency regression with one OTel attribute

The author deployed two synthetic versions of a test service to a local Kind cluster and measured latency with SigNoz. Version v1.0.0 generated 60 spans at a 100 ms duration, while version v2.0.0 generated 60 spans at a 250 ms duration, both at a rate of one span per second [DevTo]. After forwarding the OpenTelemetry Collector’s OTLP endpoint (port 4317) to SigNoz, the p99 latency chart showed two distinct lines: one hovering around 100 ms and the other around 250 ms – a clear 150% regression. The service.version resource attribute allowed SigNoz to group the data by version in its Query Builder. By setting service.version="v1.0.0" for the healthy build and service.version="v2.0.0" for the regressed build, the dashboard displayed both curves side-by-side without any custom aggregation logic [DevTo]. This approach enables version-level visibility without extra pipelines, immediate regression detection, and is extensible to error-rate and rollout monitoring. For instance, adding a status_code filter can surface a hidden 5xx increase, or combining service.version with k8s.pod.name can differentiate pods during a rolling update [DevTo].