Min/Max Recommendation
Recommends min and max replicas for each Deployment from 7 to 14 days of its traffic, as a Performance set and a Cost set.
At 9 p.m. the peak hits maxReplicas, and the HPA reports ScalingLimited
kubernetes.io, HorizontalPodAutoscaler v2 API and HPA walkthrough. The 9 p.m. peak is illustrative.
minReplicas and maxReplicas are fixed numbers you set when you create the HPA. minReplicas defaults to 1 Pod, and maxReplicas is a hard cap.
When the peak needs more Pods than maxReplicas, the HPA stops at the cap and reports ScalingLimited. When minReplicas is set high to be safe, those Pods run all night.
Pod counts and hours are illustrative. ScalingLimited and the minReplicas default: kubernetes.io. Performance and Cost sets, 7 to 14 days of history: Min/Max Recommendations docs.
How Wave does it
- Engineersets minReplicas and maxReplicas once
- HPA specHPAmin 20 and max 24, every hour of the day
- 9 p.m. peakHPA controllerstops at maxReplicas
- conditionkubectl describe hpaScalingLimited: True
- Wave Agentwith cAdvisor, on every node
- gRPCWave Corekeeps 7 to 14 days of traffic
- historyWave Intelligencelearns the 24-hour cycle
- two setsRecommendation tabPerformance and Cost min/max, with a 24-hour heatmap
Learn the daily cycle
Wave reads 7 to 14 days of each Deployment's traffic and learns its 24-hour pattern (Min/Max Recommendations docs).
Get two sets
A Performance set keeps more headroom for spikes. A Cost set runs leaner. A 24-hour heatmap shows the peak and the quiet hours.
Apply and watch
Copy the values into the Deployment's Autopilot settings. A common choice is Min Cost with Max Performance. Watch scaling for about a week (Min/Max Recommendations docs).