The knowledge graph uses a curated library of base rules to analyze incoming time-series data and provide default insights.
These insights map to the standard categories (Saturation, Amend, Anomaly, Failure, and Error).
For category definitions, severity levels, and UI ring behavior, refer to Insights categories.
In terms of implementation, the knowledge graph operates with two main sets of rules that generate insights: base rules and framework-specific rules. Base rules focus on common request and resource metrics across different platforms and frameworks.
The following table summarizes base insights raised by the base rules.
Metric type
Insight
Category
Request
RequestRateAnomaly
Anomaly
LatencyAverageAnomaly
Anomaly
LatencyAverageBreach
Error
LatencyP99ErrorBuildup
Error
ErrorRatioAnomaly
Anomaly
InboundClientErrorAnomaly
Anomaly
ErrorRatioBreach
Error
ErrorBuildup
Error
Logging
LoggerRateAnomaly
Anomaly
ErrorLogRateBreach
Error
Resource
Saturation
Saturation
ResourceRateAnomaly
Anomaly
ResourceRateBreach
Error
ResourceMayExhaust
Error
Default KPI-based insights
The following examples show how common KPIs map to default insights and how those insights are detected.
KPI
KPI metric
KPI metric description
Insight
Insight description
Request rate
asserts:request:rate5m
Request rate over the last 5 minutes.
RequestRateAnomaly
Uses a 26‑hour baseline and a standard‑deviation model to detect spikes or drops. Triggers when the metric stays outside the bounds for 5 minutes.
Latency (average)
asserts:latency:average
Average latency computed as rate(time_takencount[5m]) / request_rate_5m.
LatencyAverageBreach
Uses the last 7 days to set a threshold at the 95th percentile. Triggers when the threshold is exceeded for 5 minutes.
Latency (average)
asserts:latency:average
Average latency computed as rate(time_takencount[5m]) / request_rate_5m.
LatencyAverageAnomaly
Uses a 26‑hour baseline and a standard‑deviation model to detect spikes or drops. Triggers when the metric stays outside the bounds for 5 minutes.
Latency (p99)
asserts:latency:p99
99th percentile of the latency histogram over a 5‑minute window.
LatencyP99ErrorBuildup
Uses the last 7 days to set a 95th‑percentile threshold. Tracks how many minutes exceed the threshold in a 1‑hour window and triggers when the count passes a limit.
Error ratio
asserts:error:ratio
Error count in the last 5 minutes divided by total request count in the last 5 minutes.
ErrorRatioBreach
Uses the last 24 hours to set a 75th‑percentile threshold. Triggers when the error ratio exceeds that threshold.
