Skip to main content
A development or test cluster that runs all night and all weekend costs money while nobody uses it. You can stop it to release its compute, start it again when you need it, and let power schedules do both on a timetable, for example stop at 19:00 on weekdays and start again at 07:00. Stop, start and power schedules are available for clusters Ankra provisions on Hetzner, OVHcloud, UpCloud, DigitalOcean, Scaleway, AWS EC2, Ankra Cloud, Proxmox VE and HPE Morpheus. A cluster you connected to Ankra cannot be stopped from Ankra: the page says “This cluster was connected to Ankra, so Ankra does not own its servers. Stop and start it where it runs.” Managed Kubernetes clusters that support it stop and start at the provider, without schedules; see Managed Kubernetes.

Before you start

  • Stopping and starting needs clusters.operate on the cluster (Operator, Admin or Owner). Without it the page says “Your role cannot stop or start this cluster. That needs the clusters.operate permission.”
  • Adding, changing and removing schedules needs clusters.write (Member, Operator, Admin or Owner). Without it the page says “Your role can see these schedules but cannot change them.”
  • Starting creates servers, so the organisation’s billing must allow it. Otherwise the start is refused: “The organisation’s billing does not allow starting servers right now. Check the plan and payment method, then try again.” See Billing.
See Roles and access.

Stop and start a cluster

Go to Infrastructure → Clusters → <your cluster> → Settings → Stop, start and schedules (/next/clusters/<cluster-id>/settings/power). Stop or start now shows the cluster’s State and Provider, and offers Stop cluster while the cluster runs and Start cluster while it is stopped.
1

Stop the cluster

Click Stop cluster. On providers that capture the cluster state, leave Keep cluster state (recommended) ticked. Type the cluster’s name and click Stop <cluster name>.Check: the page reports the stop with a Follow the stop link to its operation, and the state reads stopped once it finishes.
2

Start the cluster

Click Start cluster. Where the provider offers it, choose under What to start: Everything (the control plane and the worker nodes) or Only the control plane (workloads cannot run until worker nodes are added). Click Start cluster.Check: the page says Ankra is starting the cluster, with a Follow the start link. “Starting takes about 6 minutes, and stopping is offered again once it is running.”

What a stop does

It depends on the provider: The state snapshot. On the providers that delete servers, Ankra first captures an encrypted etcd snapshot and stores it in your organisation’s oldest ready backup vault, or in an Ankra-managed default vault when you have none. The servers are deleted only once the snapshot is stored. The capture runs as a Cluster State Snapshot operation; if it fails, the cluster stays running and nothing is torn down. Untick Keep cluster state (recommended) (CLI --preserve-state=false) to tear down without a snapshot: Secrets, ConfigMaps, custom resources and the bindings between claims and volumes are then lost, and only what Ankra deploys from its own configuration comes back. A forced stop (CLI --force) never captures. When the cluster is not reachable, the box cannot be ticked and the dialog says data kept only on the servers’ disks is lost. Pause instead of tearing down. On k3s clusters on Hetzner, UpCloud and DigitalOcean, ankra cluster hetzner stop my-cluster --mode pause (or the upcloud or digitalocean command) powers the servers off and keeps them with their disks, so etcd, local data and node identities survive, and a start powers them back on. These providers bill a powered-off server in full, so a pause saves no compute cost. Pause takes no snapshot and cannot be combined with --force. The portal has no pause choice; use the CLI. A stop never deletes the cloud block volumes behind PersistentVolumes, forced or not; at a cloud provider they keep billing while the cluster is stopped. On Proxmox VE the kept disks stay on your own host. HPE Morpheus clusters have no cloud volumes: a disk goes with its server, so data kept only on the servers’ disks is lost at a stop. Only deleting the cluster removes the kept volumes; see Terminate or remove.

What a start does

On the providers that delete servers, a start creates fresh servers with new IP addresses and installs Kubernetes. When a snapshot was captured, the first control plane resets etcd from it, so your Secrets, ConfigMaps, custom resources and persistent volume claims come back, and StatefulSets reattach their existing cloud disks. A kubeadm control plane also gets its certificate authorities back, so existing tokens and kubeconfigs stay valid. To start a fresh cluster instead, use ankra cluster hetzner start my-cluster --restore-state=false (or the command for your provider); the portal always restores a captured snapshot. Stacks and add-ons configured in Ankra are reconciled either way. The snapshot is kept for as long as the cluster stays stopped. Once the cluster runs again, a restored snapshot is kept for 7 more days, and one that was never restored for 90 days from its capture. The scheduled snapshots Ankra takes while the cluster runs follow their own retention window. On Scaleway, AWS EC2 and Ankra Cloud, and after a pause, a start powers the existing servers back on; see AWS clusters.

Schedule stops and starts

The Schedules section on the same page lists the cluster’s schedules. A cluster holds at most 20.

Park a cluster overnight

1

Open Park overnight

Under Schedules, click Park overnight.
2

Choose the days and times

Choose Days (Monday to Friday or Every day), Stop at and Start at (19:00 and 07:00 by default) and the Timezone (your browser’s by default).
3

Add the pair

Click Add both schedules.Check: the table lists Stop the cluster and Start the cluster, each with its Next time.
If the stop schedule is added and the start schedule is not, the dialog says so; click Add the start schedule to add only the missing one.

Add a schedule

1

Open the dialog

Under Schedules, click Add a schedule.
2

Choose the action

Under What should happen, pick Stop or Start. Each schedule does one thing; add a pair to stop and start the cluster.
3

Choose when

Under When, choose On the same days every week and set Days (Every day, Monday to Friday or Saturday and Sunday), Time of day and Timezone. To use a cron expression instead, open Use a cron line instead and enter a five-field Cron line, such as 0 19 * * 1-5; it replaces the days and time.Or choose Once, at a date and time and set Date and time.
4

Save

Click Add schedule.Check: the new row shows the schedule under When and its next run under Next time.
A stop schedule added in the portal stops the cluster the way Stop cluster does. To schedule a stop that only removes the worker servers (scale_to_zero), use the CLI.
Ankra AI, an MCP client or a service-account token can schedule or switch on a scale_to_zero stop only on a cluster a person has confirmed is not production; otherwise the write is refused and says why. A person confirms this under Settings → General with Confirm it is <purpose> or Change what it’s used for; see Cluster settings. Schedules you create yourself in the portal, or with the CLI or API on your own token, are not checked.

Change, pause or remove a schedule

Each row shows the action, When, Next time and Last time with the last outcome (Ran, Skipped or Failed). The row’s options menu offers:
  • Change… opens Change the schedule. Change the fields and click Save schedule. A stop schedule keeps its stop mode.
  • Pause keeps the schedule but it never fires; Resume arms it for its next occurrence. A paused schedule reads Paused under Next time.
  • Remove… asks “Remove this schedule?”; click Remove.
Use the row’s options menu as above.

How schedules run

  • A scheduled stop behaves like Stop cluster, with the mode and state choice the schedule carries.
  • A stop is skipped if the cluster is already stopped or being deprovisioned. A start fires only when the cluster is stopped.
  • If the cluster is busy or the action fails transiently, the schedule retries every 5 minutes for up to 24 hours, then records a failure. A repeated schedule still fires at its next occurrence.
  • A one-off schedule disarms itself after it fires and stays in the list as a record. A repeated schedule re-arms for its next occurrence.
  • Pausing or changing a schedule always wins over a retry in flight.

Verify

  1. Open Settings → Stop, start and schedules. After a stop, Start cluster is offered; after a start, Stop cluster is offered.
  2. Open Operations on the cluster. A stop on a provider that captures state shows a Cluster State Snapshot run.
  3. After a scheduled run, the schedule’s Last time shows when it ran and its outcome.

What can go wrong

Next steps

  • Cloud Cost - see what your clusters cost and what parking them saves.
  • Cluster Settings - every settings page of a cluster, including deleting it.