Guides

ManagedWorkload guide

The ManagedWorkload is Napwake's core resource. Each one names a Deployment or StatefulSet whose lifecycle Napwake manages.

View .md Ask Claude Ask ChatGPT

Name the target.
managedworkload.yaml
```
apiVersion: napwake.dev/v1alpha1
kind: ManagedWorkload
metadata:
  name: my-api
  namespace: staging
spec:
  target:
    kind: Deployment
    name: my-api
```
Only target is required, and the rest takes its defaults. Napwake watches it, tracks its cost and learns its forecast, but pauses it only when you ask, with kubectl napwake pause.

Add an idle policy, and what else you need.

apiVersion: napwake.dev/v1alpha1
kind: ManagedWorkload
metadata:
  name: my-api
  namespace: staging
spec:
  target:
    kind: Deployment
    name: my-api

  idlePolicy:
    idleAfter: 1h
    activity:
      cpuThreshold: 10
      prometheus:
        - promQL: 'sum(rate(nginx_ingress_controller_requests{exported_namespace="staging", exported_service="my-api"}[5m]))'
    autoResume: true

  wake:
    mode: Request
    maxWait: 2m
    page: true

  prediction:
    mode: balanced

  costTracking:
    rates:
      cpuPerHour: "0.031"
      memoryPerHour: "0.004"
      storagePerMonth: "0.08"

  dryRun: false

With an idlePolicy, Napwake pauses it on its own. Each field’s type, default and limits are in the API reference.

Once my-api has had no activity for idleAfter, Napwake pauses it. A request to its Service wakes it.

What each part does

Field	What it does
`target`	The workload, in the same namespace. It can’t change once set, since a paused target would be left at zero. The API server refuses the edit. Create another ManagedWorkload for another workload
`idlePolicy`	When it’s idle, and pauses. Deploys and the `napwake.dev/last-activity` and `napwake.dev/active-until` annotations always count as activity. See Idle detection and Prometheus activity
`dependsOn`	Up to 20 workloads this one needs. They aren’t paused while it’s awake, and waking it wakes them. See Dependencies
`wake`	What wakes it through the doorman while paused. `mode: Request` is any request, and `mode: Ask` only a person’s click. See Wake on request
`prediction`	How soon the forecast acts. Leave it out, or write `prediction: {}`, for the default. See Forecast setting and Forecasting
`costTracking`	Your own `rates`, in place of the nodes’ list price. None may be negative. See Cost tracking
`dryRun`	Measures without pausing, waking the workload if it was paused. See Dry run

Edit the one the label created

Settings with more than one value aren’t annotations. Set them on the label’s ManagedWorkload, such as with kubectl patch, and Napwake keeps them.

The annotations own these. Edits are overwritten	Yours, and kept
`dryRun`, `idlePolicy` (`idleAfter`, `activity.cpuThreshold`, `autoResume`, `pauseEarly`), the workloads `dependsOn` lists, `wake`	`idlePolicy.activity.prometheus`, `waitForReady` on a listed dependency, `costTracking`, and `prediction`, whose `confidence` percentage overrides `napwake.dev/forecast`

Scaled up outside Napwake

A paused workload scaled up by anything else, such as kubectl scale or a deploy, wakes. Whoever did it wants it running. status.lastScaledUp records when, and what did it. For GitOps, see GitOps tools. While it runs, Napwake leaves its replicas to its team or HPA.

One workload per target

Only one ManagedWorkload can manage a Deployment or StatefulSet. A newer one for the same target gets a DuplicateTarget condition naming the other. It does nothing until the older one is gone.