Nov 17, live webinar: ECS to EKS without freezing the roadmap for a quarter
Scheduled

Kubernetes 1.36 upgrade: production clusters

We will upgrade every production cluster managed by Qovery (EKS, GKE, AKS and Scaleway Kapsule) from Kubernetes 1.35 to 1.36. You can upgrade a cluster yourself from the Console from 30 November.

Monday, January 11, 2027

Through the day, cluster by cluster

Applications running at least 2 instances with liveness and readiness probes keep serving traffic. An application with a single instance can go down for a short time while we replace its node.

What we will do

  1. Check that nothing on the cluster uses a Kubernetes API removed in 1.36. If a Helm chart you deployed yourself does, we stop the upgrade of that cluster and tell you.
  2. Upgrade the Kubernetes control plane of the cluster to 1.36, and the charts Qovery manages to compatible versions.
  3. Replace the nodes. Pods move to new nodes running 1.36 before we shut down the old ones.

What to check before the window

Before every upgrade

  • Tag your non-production clusters as non-production. We upgrade them first.
  • Run each application with at least 2 instances and liveness and readiness probes, so it stays up while we replace nodes.
  • Remove any PodDisruptionBudget that allows no disruption (for example on a single-instance application). It blocks node replacement.
  • Redeploy any service you have not redeployed within your cluster's image retention period. Otherwise it may fail to restart.
  • Make sure Helm charts you deployed yourself do not use Kubernetes APIs removed in the new version. If one does, the upgrade stops.

Kubernetes 1.36 and the component updates of 23 November

  • cert-manager 1.21 (from 23 November). Certificates Qovery manages are not affected. Check only what you built yourself.
    • If an issuer you created authenticates with the cert-manager controller's own service account (serviceAccountRef.name: cert-manager, for example Vault or Route53), it stops working. Give it its own service account, Role and RoleBinding.
    • If your own monitoring selects the cert-manager metrics port by name, update it. The port is renamed from tcp-prometheus-servicemonitor to http-metrics (same number, 9402).
  • Users who manage ACME challenges or orders through the Kubernetes edit role lose some permissions on them.
  • Karpenter 1.15 on EKS clusters that use Karpenter (from 23 November). The update does not replace existing nodes.
    • Dashboards or alerts built on Karpenter metrics may need an update. The reason label is now lowercase on karpenter_nodepool_allowed_disruptions and karpenter_nodepool_nodes_consuming_budgets. The consolidation_type label on karpenter_voluntary_disruption_queue_failures_total now reports empty, single or multi. The error label on karpenter_cloudprovider_errors_total is unknown instead of empty.
    • Karpenter node classes you created outside Qovery become NotReady if they set invalid kubelet options.
  • Load balancer allowed IP ranges that start with an address inside the range (for example 10.0.0.1/8) are rewritten to start at the range start (10.0.0.0/8). They cover the same addresses, and Kubernetes 1.36 rejects the old form. Ranges written as IPv4 inside IPv6 (::ffff:10.0.0.1/104) become plain IPv4 (10.0.0.0/8), and any such range shorter than /96 is rejected. Write new or changed ranges in the standard form.

Full timeline

  1. Mon, 23 Nov 2026We announce the upgrades in the changelog, in a Console banner and by email to the admins of every organization with a managed cluster.
  2. Mon, 30 Nov 2026Kubernetes 1.36 becomes the default for new clusters. You can upgrade your own clusters from the Console.
  3. Thu, 10 Dec 2026We send a reminder in your shared Slack channel and post the maintenance on status.qovery.com.
  4. Mon, 14 Dec 2026All non-production clusters are upgraded to 1.36.
  5. Thu, 7 Jan 2027We send a reminder and post the maintenance on status.qovery.com.
  6. Mon, 11 Jan 2027All production clusters are upgraded to 1.36. (this page)
  7. Mon, 18 Jan 2027Kubernetes 1.37 becomes the default on EKS, AKS and Scaleway. You can upgrade your own clusters from the Console.
  8. Thu, 28 Jan 2027We send a reminder and post the maintenance on status.qovery.com.
  9. Mon, 1 Feb 2027Non-production EKS, AKS and Scaleway clusters are upgraded to 1.37.
  10. Thu, 11 Feb 2027We send a reminder and post the maintenance on status.qovery.com.
  11. Mon, 15 Feb 2027Production EKS, AKS and Scaleway clusters are upgraded to 1.37.

What to expect

Applications running at least 2 instances with liveness and readiness probes keep serving traffic. An application with a single instance can go down for a short time while we replace its node.

If you notice anything unusual during the window, tell us in your shared Slack or Microsoft Teams channel and we will look at it straight away. We post live updates on our status page.