The latest articles on DEV Community by Victor Coisne (@vcoisne). https://dev.to/vcoisne https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F105479%2F8e4ac2c3-8ca4-4999-b3c4-1dc10b03779b.jpeg https://dev.to/vcoisne en Victor Coisne Wed, 28 Oct 2026 21:02:49 +0000 https://dev.to/plakar/free-backup-for-kubevirt-vms-and-openshift-33j6 https://dev.to/plakar/free-backup-for-kubevirt-vms-and-openshift-33j6 <p><em>Originally published on the <a href="https://plakar.link/kubevirt-devto" rel="noopener noreferrer">Plakar blog</a>.</em></p> <p><strong>TL;DR:</strong> Plakar's Kubernetes integration now backs up KubeVirt VMs: every disk plus the VM manifest, in one snapshot, consistent across disks. On OpenShift, a new <code>openshift</code> package keeps restores from doing things you didn't ask for. Both are free.</p> <p>OpenShift Virtualization is Red Hat's packaging of upstream KubeVirt. A VM is a Kubernetes object, its disks are PVCs, so our Kubernetes integration should have covered them already. It didn't: it rejected <code>VolumeMode: Block</code> outright, and most VM disks are block-mode. That was the actual blocker.</p> <h2> We don't freeze VMs. KubeVirt does. </h2> <p>A disk snapshotted mid-write may not boot. A VM with three disks snapshotted a few seconds apart is worse. KubeVirt already solves this: its <code>VirtualMachineSnapshot</code> freezes the guest through the qemu guest agent and coordinates every disk. So we drive it instead of reimplementing it.</p> <p>The flow: Plakar requests a <code>VirtualMachineSnapshot</code>, a short-lived pod streams each disk out as raw blocks, and Plakar dedups, compresses and encrypts it into the same snapshot as the VM manifest. Restore goes the other way: fresh PVCs, data written back, manifest applied, into the original namespace or a new one.</p> <p>No guest agent, no freeze: Plakar gives up after 10 minutes instead of waiting forever. Install the agent.</p> <h2> OpenShift: restore what you backed up, nothing more </h2> <p>Backup stays a complete record. Filtering happens on restore, because some objects come with side effects: restoring a <code>BuildConfig</code> re-triggers a build, a <code>Route</code> reassigns a host, <code>SecurityContextConstraints</code> and OAuth tokens belong to the cluster. The <code>openshift</code> package skips them. Its helper pod sets its security context explicitly, and it matches OpenShift's default restricted SCC: OpenShift doesn't let pods add capabilities anyway.</p> <h2> Try it </h2> <p>Pre-built packages need a free Plakar account:<br> </p> <div class="highlight js-code-highlight"> <pre class="highlight shell"><code><span class="nv">$ </span>plakar login <span class="nv">$ </span>plakar pkg add k8s <span class="nv">$ </span>plakar backup k8s+vm:/vms/my-vm </code></pre> </div> <p>On OpenShift:<br> </p> <div class="highlight js-code-highlight"> <pre class="highlight shell"><code><span class="nv">$ </span>plakar pkg add openshift <span class="nv">$ </span>plakar backup openshift+vm:/vms/my-vm </code></pre> </div> <p>RBAC roles and options are in the <a href="https://plakar.io/docs/community/main/integrations/kubernetes/" rel="noopener noreferrer">Kubernetes docs</a>.</p> <h2> How it ships </h2> <p>KubeVirt support lives in the Kubernetes integration, which is <a href="https://github.com/PlakarKorp/integrations/tree/main/k8s" rel="noopener noreferrer">open source</a> (ISC). The <code>openshift</code> package is not open source, but it is free for everyone in the Plakar community: create a free account, run <code>plakar login</code>, install it.</p> <p>Running VMs on Kubernetes? Break it and tell us on <a href="https://discord.gg/uuegtnF2Q5" rel="noopener noreferrer">Discord</a>.</p> kubernetes kubevirt devops backup