What is broken now, before anyone asks.
Every few minutes SKYLINE reads each scanned Kubernetes environment and keeps a list of what is broken, even where no alert rule exists. Scans call no model and use no analysis.
Workloads and nodes
Crash loops, image pull errors, OOM kills, pods stuck pending or not ready, nodes not ready or under pressure. Every 5 minutes by default.
From Prometheus
Volumes filling up, and error rates and latency compared with the service’s own last 24 hours. Every 15 minutes.
What breaks later
cert-manager certificates near their end, deprecated APIs and version skew. Once a day.
One finding per workload
A Deployment whose pods crash is one finding. It resolves after two clean scans, so a flapping pod does not open and close all day.
From a finding to a fix
- 01
Scanning is already on
For environments with a stored kubeconfig or the in-cluster agent. Change the interval or skip namespaces per environment.
- 02
Open the list
Priority, kind, workload, namespace and how long it has been open.
- 03
Investigate
One click starts an investigation on the finding. Only that uses an analysis.