Introduction to Wave
Wave is the best-practice operations layer for Kubernetes. Every day it analyzes your cluster, diagnoses the root cause of scaling, sizing, and reliability issues, and recommends the right configuration, then applies it automatically across scaling, sizing, diagnosis, traffic, and nodes. Like a Kubernetes ops expert that lives in your cluster, filling the operational gaps Amazon EKS and OpenShift leave open. SREs and platform engineers ship more reliable clusters at 30-40% lower cost.
Wave Autoscale is now Wave. The technical identifiers (wave-autoscale
namespace, Helm chart, env vars) are unchanged for compatibility, so you'll
still see them in install commands.
What is Wave?
Wave is a Kubernetes platform that both advises and automates the Day 2 operations engineers otherwise hand-tune on top of HPA, VPA, and Prometheus. ML models analyze each workload's pattern to diagnose issues, recommend the right configuration, and (when you let them) apply autonomous scaling, sizing, traffic, and capacity decisions in real time. You stay in control: run any capability in advisory (recommend-only) mode first, then turn on automation once you trust it.
Built for SREs, platform engineers, and DevOps teams running Kubernetes at scale, Wave reclaims the 60-70% of engineering time spent on Day 2 operations and reroutes it to strategic work.
Sub-brands at a glance
Wave Sizing
Continuous, in-place vertical right-sizing.
- Smart Sizing: continuous vertical sizing of Pod requests/limits. In-place patches every 10 minutes vs VPA's evict-and-recreate, and pairs with Autopilot instead of fighting it.
- Realtime Resizing: applies sizing changes to running pods without a restart, using in-place resize where the cluster supports it and falling back to a manifest patch on older clusters.
- Smart Sizing Report: a cluster-wide report of right-sizing recommendations and the cost they reclaim, ready to share with your team.
Wave Diagnosis
ML-powered observability and forecasting.
- Cluster Resource Forecast: 7-30 day cluster capacity forecasts, so procurement and capacity planning happen ahead of exhaustion, not after.
- Memory Leak Detection: predicts OOMKill events hours to days in advance, so you find leaks before the 3 a.m. page.
- CPU Utilization Analysis: workload-level CPU breakdowns with per-label segmentation. Skip the Grafana drill-down; Wave shows which label group is driving cost.
- Pod Scheduling Delay Detection: detects pods stuck in Pending, with a per-workload latency breakdown vs the cluster-wide aggregates Prometheus shows.
- Idle Node Detection: surfaces underutilized nodes for cost reduction. Action-oriented, not "look at this dashboard."
Wave Autoscale
ML-driven horizontal scaling and intelligent replica recommendations.
- Autopilot: ML-driven horizontal scaling per workload, with under-10s response from learned per-workload patterns vs HPA's 30s to 5min reactions to static thresholds.
- Autopilot Scheduler: UI-based scheduled scaling for known events. No more cron jobs editing HPA YAMLs before Black Friday.
- Min/Max Recommendation: data-driven
minReplicas/maxReplicasrecommendations, replacing guesswork with bounds derived from actual workload behavior.
Wave Karpenter
Production-grade Karpenter operations on Amazon EKS.
- Karpenter Dashboard: real-time node monitoring with cost tracking, spot ratio, and NodePool analytics. The visibility Karpenter doesn't ship out-of-the-box.
- Karpenter Node Warmup: proactive node pre-provisioning with image pre-caching, for 10× faster cold starts (31s to 3s).
- Karpenter Spot Workload Placement: per-deployment spot/on-demand split via webhook injection. Capture spot savings without putting baseline pods at risk.
Wave Flow
Priority-based traffic shaping on Istio.
- Wave Flow: WASM-based priority shaping across Istio Sidecar, Ambient, and Envoy Standalone. Critical traffic (checkout, payments) stays protected during surges while bulk traffic gracefully degrades.
- NetFUNNEL Integration: workload-aware virtual waiting room that absorbs traffic surges instead of returning 5xx during peak events.
Additional Features
Storage, alerts, and programmability.
- PV Auto-Expansion: expands PersistentVolumes before they fill. No more 3 a.m. "PV full" pages.
- PV Capacity Forecast: forecasts PV exhaustion 7-30 days out, so you plan capacity instead of reacting.
- PV Cleanup: removes orphaned PersistentVolumes, reclaiming storage costs that nobody had time to chase.
- Unused PV Detection: finds attached-but-idle PVs, spotting waste that PVC counts hide.
- Alerts: configurable webhook alerts for scaling events and anomalies (Slack, email, or custom), without writing yet another exporter.
- Programmable API: RESTful API for everything Wave does, so you can automate Wave from your own pipelines.
- SSO (OIDC): OIDC single sign-on for console access, wired into your existing identity provider.