Wave Diagnosis
A monitoring dashboard shows you metrics. Wave Diagnosis answers the questions those metrics are supposed to help you answer: what is failing right now, where you are paying for capacity nothing uses, and what will run short in the next few weeks. It analyzes your cluster continuously and turns the raw signals into direct answers, so you do not have to read charts to find the problem.
Three questions Wave Diagnosis answers
Every check is grouped under the operational question it answers, not by the metric it happens to read.
Failures: "What is wrong right now?"
Something is degraded or crashing, and you need to know what and why before it spreads.
- Pod/Container Failures: the newest diagnosis. Detects abnormal pod and container terminations (OOMKilled, CrashLoopBackOff, silent crashes, evictions, preemptions) automatically from the data the agent already pushes, and captures the logs, YAML, and metric snapshots you need for root cause.
- Memory Leak Detection: predicts OOMKill events hours to days in advance, so you catch leaks before the 3 a.m. page.
- Pod Scheduling Delay: detects pods stuck in Pending with a per-workload latency breakdown, instead of the cluster-wide aggregates Prometheus shows.
- Workload CPU Utilization: classifies every workload by CPU pressure and breaks usage down per label, so you can see which group is throttling or driving cost without a Grafana drill-down.
Waste: "Where is money leaking?"
You are paying for capacity that no workload is using.
- Idle Node Detection: surfaces nodes running only DaemonSets, with no application pods, along with the cost they waste, so you can consolidate or scale them down.
- Unused PV Detection: finds attached-but-idle PersistentVolumes, the waste that PVC counts hide.
Capacity: "What will soon run short?"
A resource is trending toward exhaustion and you want lead time, not a 3 a.m. page.
- Cluster Resource Forecast: forecasts CPU, memory, and pod capacity exhaustion 7 to 30 days out, so procurement and capacity planning happen ahead of the wall, not after it.
- PV Capacity Forecast: forecasts PersistentVolume exhaustion 7 to 30 days out, so you expand storage on a schedule instead of during an outage.
From dashboards to answers
A monitoring dashboard hands you raw metrics and leaves the interpretation to you: open the dashboard, spot a trend, correlate it across workloads and nodes, decide whether it matters, and work out what to do. That loop depends on someone remembering to look and knowing what to look for.
Wave Diagnosis runs that loop for you, continuously. It analyzes every workload, node, and volume, detects the pattern, ranks it by severity (critical, high, medium), and states the recommended action. You get the answer, not the homework.