Deckhouse Flant Edition
Changes on Rock Solid
1.65.11.66.202.10 18:03
1.64.21.65.125.09 15:01
1.63.51.64.218.09 12:38
1.63.41.63.517.09 15:24
1.63.31.63.415.09 19:24
1.63.21.63.315.09 12:59
1.62.21.63.211.09 15:23
1.66.2
AlphaBetaEarly AccessStableRock Solid- Dates
- released unknown detected 02.10.2026 18:03
- Requirements
- no requirements
- Stage
- General Availability
- On release channels
- Alpha — since 02.10 · Beta — since 02.10 · Early Access — since 02.10 · Stable — since 02.10 · Rock Solid — since 02.10.2026 18:03
- Before
- never
Compare with
none
What changed1.65.11.66.2
New features 16
- The VerticalPodAutoscaler form has been reworked: the controller kind is picked from three (Deployment, StatefulSet, DaemonSet), the name from the existing resources of that kind in the namespace, and the API version is filled in automatically. Eviction requirements, controlled values, the scope, and the choice of recommender were added, and the recommendation boundaries are collected in a single table. An arbitrary controller kind is set on the YAML tab; quick container buttons and per-container consumption charts were removed from the form.duplicate not checked
- VerticalPodAutoscaler recommendations are visible without opening YAML: the "Overview" tab shows the resource states, the target workload with a link to the controller, the controller pods, a table of container recommendations with the current request and the size of the change, and the "How it will be applied" section with the update mode and the pod disruption risk.duplicate not checked
- An image opens for viewing and editing right from the list: the form appears in the column next to it, and the list stays in place.duplicate not checked
- Cluster images in the system part are shown by a generated list — with filters, sorting, and export.duplicate not checked
- Virtual machines that cannot migrate no longer hold a node in maintenance: permission to restart them is given by a checkbox in the reboot and drain dialogs, or by a button on the "State" tab. The platform used to wait for that permission with no way to grant it from the console.duplicate not checked
- A drain of a node that is already running can be cancelled.duplicate not checked
- You can see what is really happening to a node and whether it needs your attention: the state follows the actual course of maintenance rather than the request that was filed — "Preparing to reboot", "Rebooting", "Being recreated", "Finishing maintenance", "Approve VM restart" — and the badge tooltip explains it.duplicate not checked
- A service with health checks can be created and edited in the console without writing the manifest by hand: type, ports, workload selection by labels, and HTTP, TCP, or PostgreSQL probes are filled in as form fields.duplicate not checked
- You can now see which pods stand behind a service: the service card has tabs with endpoints — whether a pod is ready to accept traffic and which node it runs on. The separate menu sections listing endpoints are gone; endpoints open from the service card.duplicate not checked
- Managed service resources are set with a slider: the number of cores and the amount of memory are set that way, and the storage class now comes with an explanation of what it gives.duplicate not checked
Fixes 27
- Fixed the layout of the Security card on the namespace overview: a long list of scores no longer overflows the card, and on narrow screens the card now spans the full row width.duplicate not checked
- Opening a pod, Deployment, StatefulSet, DaemonSet, Job or CronJob page on a cluster with a large number of events no longer loads kube-apiserver: object events are now requested only when the "Events" tab is opened, not right when the page loads.duplicate not checked
- The "Events" tab of a pod and controllers opens faster and no longer shows an event counter: the events subscription no longer scans every event in the namespace.duplicate not checked
- Deleting a virtual machine no longer requires stopping it first: the dialog warns you when the machine is running, and the delete button shows the operation in progress while deletion waits for finalizers.duplicate not checked
- The image status looks the same in the image list, in the image form, and in the disk table of a virtual machine: a freshly created image shows "Creating" rather than "Unknown", and for a failed one the tooltip next to the status names the reason.duplicate not checked
- The API and assistant ports of the console no longer answer an arbitrary client in the cluster: requests go through `kube-rbac-proxy`, which checks the rights of the caller. The metrics and diagnostics ports are closed: without the right to the `prometheus-metrics` subresource the metrics are not served. Console publishing proves itself with a client certificate, and traffic from it to the console goes over TLS — both on the nginx-ingress path and on the ALB path.duplicate not checked
- The "By nodes" and "By workloads" vulnerability report lists no longer break when no vulnerabilities were found, and the "Specification" tab of a report says so instead of showing an empty screen.duplicate not checked
- The status of a locked local user explains in a tooltip who locked the account and until when, and a permanent lock is named in words. An unlocked user shows the "Active" status again on platforms where the cluster does not fill in the lock block.duplicate not checked
- The local user form accepts the same addresses as the cluster does: an address with a percent sign used to pass, while an address with a digit in the top-level domain did not.duplicate not checked
- The reboot button no longer leads to a second stop of the node: it is disabled while maintenance is running, and in a group with `approvalMode: Manual` the request now carries the disruption approval, so the node is not stopped again just to confirm it.duplicate not checked