kpulse

Event-driven Kubernetes monitoring for developers and startups on their first cluster. One Pod, one ConfigMap, ~64 Mi of memory. Slack / email / webhook out of the box. Not a Prometheus replacement. Web: kpulse.io

kpulse watches the cluster, catches the failure modes that wake teams up at night (pod crashes, PVC full, certs expiring, rollouts stuck), and pings the channel of your choice. Outgrow it later by adding Prometheus alongside, not instead.

TL;DR

curl -fsSL https://kpulse.io/install.sh | bash
kubectl -n kpulse edit secret kpulse-secrets   # add SLACK_WEBHOOK_URL
kubectl -n kpulse edit configmap kpulse-config # uncomment slack stanza, set cluster.name
kubectl -n kpulse rollout restart deploy/kpulse

You now have alerts on 12 common cluster failure modes flowing to Slack.

Why kpulse

• Day-1 ready. Install, paste one Slack webhook, you have alerts on the 12 most common failure modes.
• No time-series stack required. No Prometheus, no Grafana, no Alertmanager, no PVCs.
• Sane defaults. All 12 monitors enabled, thresholds tuned to be quiet on a healthy cluster.

Install

curl -fsSL https://kpulse.io/install.sh | bash

Then configure a channel:

kubectl -n kpulse edit configmap kpulse-config   # set cluster.name, enable a channel
kubectl -n kpulse edit secret kpulse-secrets     # add e.g. SLACK_WEBHOOK_URL
kubectl -n kpulse rollout restart deploy/kpulse

Test it:

kubectl -n kpulse port-forward svc/kpulse 8080:8080 &
curl 'http://localhost:8080/test-channel?name=slack'

What it watches out of the box

Monitor	Triggers	Severity
pod_crashes	CrashLoopBackOff, OOMKilled, ImagePullBackOff, etc.	critical
pod_restarts	> 5 restarts in 15 min	warning
warning_events	Warning-type k8s Events (with noisy reasons filtered)	info
pvc_usage	PVC > 80% warn, > 90% crit	warn/crit
node_conditions	DiskPressure, MemoryPressure, PIDPressure, NotReady	critical
node_disk	node rootfs/imagefs > 85% warn, > 92% crit	warn/crit
tls_cert_expiry	TLS Secret expiring in < 14 d warn, < 3 d crit	warn/crit
rollout_stuck	Deployment/StatefulSet rolling for > 15 min	warning
job_failed	Job condition Failed=True	warning
cronjob_missed	> 2 missed schedules	warning
hpa_at_max	HPA pinned at maxReplicas for > 30 min	warning
daemonset_unscheduled	desired != ready for > 10 min	warning

Full details and how to deliberately trigger each one: docs/monitors.md.

Install options

Method	Command
One-line script	curl -fsSL https://kpulse.io/install.sh | bash
Raw manifest	kubectl apply -f https://github.com/dnl555/kpulse/releases/latest/download/kpulse.yaml
Helm	helm install kpulse oci://ghcr.io/dnl555/charts/kpulse --namespace kpulse --create-namespace
Local build	git clone ... && make build image

Contributing

Issues and PRs welcome. The codebase is intentionally small (~1500 LOC Go, 7 packages):

internal/
  alert/        Alert struct + Severity
  config/       ConfigMap parser + Secret resolver
  notifiers/    Slack / SMTP / webhook / Discord / Teams
  engine/       dedupe, routing, digest
  state/        ConfigMap-backed persistence
  monitors/     12 monitors (informer-based + periodic)
  httpsrv/      /healthz, /readyz, /metrics, /test-channel
cmd/kpulse/     main wiring

Run make test to test, make build to build the binary, make image VERSION=dev to build the image.

Channels

Slack, SMTP email, generic webhook, Discord, Microsoft Teams. Pick any subset. Each goes through the same dedupe + digest engine.

See docs/channels.md for configuration and routing.

What kpulse is NOT

• Not a metrics store. No time series, no PromQL, no historical graphs.
• Not a dashboard. No UI in v1.
• Not Alertmanager. No silencing rules, no on-call schedules, no acknowledgements.

If you need any of those, run Prometheus + Grafana + Alertmanager. kpulse covers the gap before you're ready for that stack, and keeps doing the noisy "did Kubernetes break again" work after.

License

MIT. See LICENSE and NOTICE (kwatch is acknowledged as inspiration; no kwatch code is included).