You can't operate what you can't see. This is a suite of tools that answers, on demand and at fleet scale: which clusters are unhealthy, which services are down, and what changed — before users or pagers find out.
With hundreds of clusters across hundreds of accounts, "is everything okay?" was unanswerable. Dashboards existed per cluster, but nothing swept the whole fleet and produced a single, actionable answer. Incidents were discovered by users; post-upgrade regressions were discovered late.
cluster-health sweeps every cluster in every account concurrently → node conditions, workload status, control-plane health → aggregated CSV: cluster, issue, severity, action k8lens async endpoint prober for 1000s of services → aiohttp fan-out, bounded concurrency → Prometheus metrics for trending, hardened deployment sanity-check post-change validation framework → nodes, workloads, addons, CRDs after any upgrade → pass/fail per check, machine-readable output
Each tool works standalone; together they cover the three questions that matter: steady-state health (cluster-health), user-facing availability (k8lens), and did-my-change-break-anything (sanity-check).
Async where it counts. Endpoint probing uses asyncio + aiohttp with bounded concurrency — thousands of HTTP checks complete in the time a sequential script would spend on dozens.
Native Kubernetes client, not shelling out. Structured API responses instead of parsing kubectl text means checks are precise and version-proof.
Reports built for action. Output is aggregated, deduplicated, and severity-ranked CSV/HTML — designed to be triaged in minutes, not scrolled for hours.
Metrics as a first-class output. k8lens exports Prometheus metrics, so one-off checks double as a continuous monitoring source with alerting and history.
"Is the fleet healthy?" became a ten-minute question with a definitive answer. Post-upgrade regressions surface immediately in sanity reports, and endpoint trends live in Prometheus instead of someone's memory.