kpulsev1GitHub →
v1.0 · MIT · ~1500 LOC Go

Kubernetes alerts
that just work.

Event-driven monitoring for developers and startups on their first cluster. One Pod, one ConfigMap, ~64Mi of memory. Slack, email, and webhooks out of the box.

~ install kpulse
curl -fsSL https://kpulse.io/install.sh | bash
QuickstartSee what it watches
1
Pod
~64Mi
Memory
12
Monitors
5
Channels
// what you actually get

An alert you can act on, in seconds.

No dashboard to open, no query to write. kpulse delivers the namespace, object, reason, and the last log lines straight to your channel of choice.

Example kpulse critical alert: OOMKilled on api-7d9f/server with container logs
Example kpulse resolved alert: PVC checkout/data-api-0 back below threshold
// and when it's over

Auto-recovery alerts, so you can close the tab.

Every alert gets a matching [OK] RESOLVED message the moment kpulse sees the condition clear. No stale incidents, no manual acks, no "is this still firing?" pings at 2am.

  • • Same channel, same thread context — fire and recover, side by side.
  • • Includes the final value vs threshold so you know it really went back to normal.
  • • Works for every monitor: PVC usage, pod crashes, job failures, cert expiry, and more.
// straight into slack

Right where your team already lives.

Compact, scannable Slack messages with severity, namespace, object, and reason — color-coded so a glance is enough. Resolved alerts land in the same channel, so your incident history reads itself.

  • • One-line summaries that fit your channel without taking it over.
  • • Dedupes noisy events so you don't get the same alert 40 times.
  • • Works with any Slack webhook — no app install, no OAuth dance.
Example kpulse alerts in a Slack channel: BackOff, CrashLoopBackOff, and matching RESOLVED messages

// why kpulse

Prometheus is great — until you just want to know when a pod crashes.

Day-1 ready

Install, paste one Slack webhook, you have alerts on the 12 most common failure modes. No tuning weekend required.

No time-series stack

No Prometheus, no Grafana, no Alertmanager, no PVCs. kpulse listens to the Kubernetes API and sends.

Sane defaults

All 12 monitors on. Thresholds tuned to be silent on a healthy cluster. Dedupe and digest built-in.

// philosophy
Your monitoring stack shouldn't be bigger than your app.

kpulse is the lightweight alternative when all you need is to know the moment something breaks.

// 12 monitors out of the box

The failures that wake teams up.

Each fires on a tuned threshold. Each can be silenced or rerouted via ConfigMap.

docs →
pod_crashes
critical

CrashLoopBackOff, OOMKilled, ImagePullBackOff

pod_restarts
warning

> 5 restarts in 15 min

warning_events
info

Warning k8s Events, noise filtered

pvc_usage
warn/crit

PVC > 80% warn, > 90% crit

node_conditions
critical

DiskPressure, MemoryPressure, NotReady

node_disk
warn/crit

rootfs/imagefs > 85% warn, > 92% crit

tls_cert_expiry
warn/crit

TLS Secret < 14d warn, < 3d crit

rollout_stuck
warning

Deployment/StatefulSet rolling > 15 min

job_failed
warning

Job condition Failed=True

cronjob_missed
warning

> 2 missed schedules

hpa_at_max
warning

HPA pinned at maxReplicas > 30 min

daemonset_unscheduled
warning

desired != ready > 10 min

// install

Three commands. One alert pipeline.

  1. 1

    Install

    $ install
    curl -fsSL https://kpulse.io/install.sh | bash
  2. 2

    Configure a channel

    $ configure
    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
  3. 3

    Test it

    $ test
    kubectl -n kpulse port-forward svc/kpulse 8080:8080 &
curl 'http://localhost:8080/test-channel?name=slack'
Raw manifest
kubectl apply -f https://github.com/dnl555/kpulse/releases/latest/download/kpulse.yaml
Helm
helm install kpulse ./charts/kpulse
Local build
git clone … && make build image

// channels

Ping the channel
of your choice.

Pick any subset. Every alert goes through the same dedupe and digest engine before it leaves the cluster.

Slack
Webhook URL, threaded digests.
SMTP Email
Plain SMTP, no provider lock-in.
Webhook
JSON POST to any URL.
Discord
Native embed format.
MS Teams
Adaptive card payloads.

// optional UI

Or open a
browser.

v0.3.0 ships a small read-only UI embedded in the binary. Off by default; flip ui.enabled: true and port-forward to see exactly what's firing, what cleared, and per-monitor counters since startup.

  • ~25 KB embedded, zero CDN deps, works air-gapped
  • Vanilla HTML/CSS/JS (no React/Tailwind/Alpine bundles)
  • Read-only beyond Reset-dedupe + Send-test buttons
  • Same JSON API at /api/v1/* for scripts
kpulse UI dashboard with active alerts and recent activity

// dashboard: severity counters + active + recent in one view

kpulse UI alerts page with active and recent panels

// alerts: full active set + recent ring buffer

kpulse UI monitors page with knobs and per-monitor counters

// monitors: all 12 with live fire/resolve counters

kpulse IS

  • Day-1 alerts on common failures
  • Lightweight: 1 Pod, ~64Mi RAM
  • Slack / SMTP / Webhook / Discord / Teams
  • Sensible thresholds out of the box
  • Dedupe + digest engine

kpulse is NOT

  • Not a metrics store
  • No PromQL or time-series
  • No silencing schedules or on-call rotation
  • Not for teams already running a full observability stack
  • No multi-cluster aggregation (one Pod per cluster)

// faq

Questions.

Does kpulse replace Prometheus?

No — it's an alternative. Prometheus is great until you realize you just wanted to know when a pod is crashing. If you need metrics, PromQL, and on-call schedules, run Prometheus. If you just want instant failure alerts, kpulse is the faster, lighter route.

How much does it cost to run?

One Pod, ~64Mi memory, ~10m CPU. On most clusters, the overhead is invisible.

What permissions does it need?

Read access to Pods, Events, PVCs, Nodes, Deployments, StatefulSets, Jobs, CronJobs, HPAs, DaemonSets, and Secrets (for TLS expiry checks).

Can I silence specific alerts?

Not via silencing rules. You can disable monitors entirely or tune thresholds in the ConfigMap. For full silencing, route to Alertmanager.

Is it open source?

Yes. MIT licensed. ~1500 LOC of Go in 7 packages. Issues and PRs welcome.

Stop discovering crashes
from your users.

One command. One Slack webhook. Twelve fewer pages at 3am.

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