ovtools v0.17.0

[0.17.0] 2026-08-21

Added

• A health check names the virtual machines claiming the same MAC address, with the namespaces and networks involved. The cluster raises its own alert when duplicates appear, but says only that they exist. It counts both the address a machine is running with and the one pinned in its configuration, because two machines can ask for the same address while only one of them gets it, and nothing in the running state shows that the other asked. The message says which of the three situations it is: both running with it, one running and one waiting to take it at its next restart, or neither running with it yet.

Changed

• The grouping control on the Fleet tab marks the view you are looking at more quietly. It was painted in the same strong blue the buttons that do something use, so a control that only reports where you are stood out more than the actions around it.

ovtools v0.16.0

[0.16.0] 2026-08-20

Added

• A health check reports Windows virtual machines running without the Hyper-V enlightenments. Without them Windows behaves as though it were on physical hardware and pays a full trip to the hypervisor for work that a single call would settle, which shows up as sluggish menus and slow application launches while the CPU figures look ordinary. Machines that arrive through a migration are the usual case, since the migration creates them without the settings the Red Hat Windows templates carry. It is reported once for the fleet with the count, rather than once per machine, and links to a write-up of how to apply them.

Fixed

• Demo mode no longer shows a stopped virtual machine attached to a node. Its disks inherited the contradiction and reported a node for a machine that is not running anywhere.

ovtools v0.15.0

[0.15.0] 2026-08-20

Added

• The health checks for a virtual machine now appear in its detail panel, so someone inspecting a machine sees its warnings without having to know they live on the Health tab. When the node underneath has a check of its own, the panel says so and links to it, because a node problem belongs to the node rather than to every machine running on it. A machine with nothing to report shows no section at all.
• A health check flags a vCPU change that never reached the guest. The virtual machine reads as though the change was granted and the guest does not have it, so whoever asked and whoever approved both believe it happened. The check separates a change that has not been applied yet from one that cannot be applied at all, because the socket ceiling puts it out of reach.
• The vCPU count in the VMs table invites a closer look where the number needs qualifying: when the machine is waiting for a CPU, or when a change to that count never landed. Resting the pointer on it explains which.
• The overcommit figure on the Nodes tab links to the machines behind it, when there are any waiting for a CPU.

Changed

• vCPU counts are what the guest has, not what the machine was configured to ask for. The two part company when a change cannot be applied to a running machine, and every figure built on the count inherited the wrong one: the vCPUs column, the detail panels, the overcommit ratio per node, the dashboard total and the spreadsheet ratio. What was asked for is kept separately and shown where it differs.
• The node CPU check says what the machines on the node measure instead of concluding it from the node's own pressure. The two are different readings: node pressure counts time when anything at all on the machine was queued, which logging and monitoring reach on their own, while the per-machine reading is what that guest lost. A node under pressure whose machines all read fine no longer raises a warning nobody can act on; the figure stays on the Nodes tab where it belongs.

Fixed

• Demo mode shows health checks against the machines and nodes it actually generates. Every check named a placeholder, so none of them could be tied to anything on screen, and the new detail panel would have looked empty for every machine.
• The vCPU wait figures in demo mode stay put between refreshes instead of being drawn again each time.

ovtools v0.14.0

[0.14.0] 2026-08-19

Added

• OVTools now says which virtual machines are waiting for a CPU. This is the reading VMware administrators know as CPU ready: the share of time a machine's virtual CPUs spend queued for a physical one while the guest believes they are running. It sits next to the vCPU count on the VMs tab, in the detail panel and in the Fleet pane, and it raises a health check past the same thresholds people arriving from VMware already use. Machines with pinned CPUs are left out, because they own their CPUs and never queue for one.
• A health check flags a virtual machine given more vCPUs than the node it runs on has CPUs. A machine configured that way has never had all of its vCPUs running at once. The message shows how the vCPUs are laid out, which usually explains where the count came from, since machines imported from VMware carry that sizing over as it was.
• The labels people put on their virtual machines are on screen: a column in the VMs table, and a row in the detail panel and the Fleet pane. The column starts hidden, because labels are long and most people are not looking for them. Labels that OpenShift Virtualization writes for its own bookkeeping are left out.

Changed

• When a tab is refused for lack of permission, the notice names a role for the one resource that was refused. It used to suggest read access to the whole cluster, which would have handed over every workload in the namespaces the person is not meant to see.
• The read role that shipped with the deployment was taken back out, for the same reason. It granted read access across every namespace, which is the wrong answer for anyone whose access is meant to stop at their own projects.

Fixed

• The Refresh button shows the data it asked for. It used to show the previous figures, and only a second press showed anything new. It now stays busy until the new data arrives and then reloads the tab, without holding up the rest of the interface, and pressing it again while it works joins the refresh already running instead of queueing another.
• The CSV button says a tab has nothing to export, instead of sending the person to an error page outside the application. It affects Dashboard, Topology, Fleet and Console, which are not tables.
• Tabs load faster for a user whose access is limited to a few namespaces. An empty list was being read as a missing one, so every render fetched it again rather than using the answer already held.
• The first load is faster on large clusters. The health checks were listing the virtual machines again when the refresh cycle had already done it. On a cluster of about 800 machines the first load went from 18 to 15 seconds.

ovtools v0.13.0

[0.13.0] 2026-08-18

Added

• OVTools now has a light theme, picked from a settings menu in the header. The default follows the operating system, so a machine set to switch in the evening switches OVTools with it, and an explicit choice is remembered. The login screen, the charts and the topology map all follow whichever theme is in use.
• A read role, ovtools-reader, now ships with the deployment manifest. It covers exactly what OVTools reads on a signed-in user's behalf, so granting it is the answer when someone is told a tab is not permitted. It replaces the older advice to grant cluster-reader, which does not cover the snapshot resources.
• Resting the pointer on the logo now says what it is. The logo has always been a link back to the dashboard.
• The snapshot contents list now shows where the orphaned storage is concentrated, grouped by storage class, by driver and by the namespace the snapshot came from. Each group separates the part that deleting the content will not free, which is the part that has to be reclaimed on the storage system. The breakdown follows whatever filters are applied.

Changed

• The footer now shows only the version.

Fixed

• Status badges, labels and severity marks are readable in both themes. Most of them carried a colour picked when dark was the only theme, which left them faint on the dark background, and hovering a row made them harder to read rather than easier. The row under the pointer was the hardest one in the table.
• Demo mode now raises every health check a real cluster raises. It was showing about a third of them, alongside a set of check types that no cluster produces, so the check type filter offered values that match nothing and hid most of what the tool reports.
• Demo mode now shows the case where two storage classes share a Ceph pool and the usage figure passes 100 percent, which is what the explanation on the Used column exists for.
• The usage figures in the tables colour their text again instead of filling the whole cell with a block of colour, and the storage usage bar is visible again.
• A tab no longer keeps saying you are not permitted to view it after the access has been granted. The refusal was remembered for the rest of the session and it replaced the rows, so the data stayed hidden until the next sign-in.
• The loading indicator no longer shows a dark panel over a light page.
• The panel showing where orphaned storage sits now lines up with the table below it.
• The Logout button is back inside the header actions instead of sitting against Refresh.
• The note under the orphaned storage breakdown now says which columns it means and what the space is that deleting a content will not free. It sits between two tables and read as though it described the wrong one.

ovtools v0.12.2

[0.12.2] 2026-08-18

Fixed

• Demo mode now shows the same snapshot reclaim advice a real cluster gets, including both deletion policies side by side, instead of the older single summary line.
• The link from an orphaned snapshot content to the OpenShift console now opens the right page. It was pointing one level too deep and did not land where it should.
• Console links are built correctly on clusters whose console address is recorded with a trailing slash, which would previously have produced a doubled slash in the link.

ovtools v0.12.1

[0.12.1] 2026-08-07

Added

• The snapshot contents list now totals the space held by the orphaned entries it is showing, following whatever filters are applied, so narrowing to the old ones answers how much that selection would give back.
• The health check now says whether deleting an orphaned snapshot content actually frees anything. When its deletion policy is Retain it does not: the snapshot stays on the storage system and has to be reclaimed there, which is the usual case for snapshots left behind by a backup tool. The snapshot's identifier on the storage system is now available as a column and in the exports, so it can be found on the array.
• An orphaned snapshot content found by the health check now links straight to its page in the OpenShift console, so it can be inspected without hunting for it by name. The link opens the YAML, which is where the driver, volume handle and source snapshot are; the console has no dedicated page for this kind. Only orphans get the link.

Fixed

• The warning about Prometheus returning no data no longer appears right after opening the dashboard when nothing is actually wrong. The check runs while the first load is still in progress and could simply run out of time, which was being reported as a problem with the cluster. The warning now waits until the condition is still there on the following check. The status dot still changes colour immediately, so a real outage is visible straight away.

Added

• When the Prometheus check does report no data, the log now says whether the check failed to complete or genuinely found nothing, which is the first question when someone reports the warning.

ovtools v0.12.0

[0.12.0] 2026-08-07

Added

• Users whose access is limited to a few namespaces now see their own virtual machines, disks, networks, CPU, storage claims, health checks and metrics, instead of an empty dashboard. What each person sees matches what their permissions allow, the same way the OpenShift Virtualization console behaves. Nothing changes for users who can already read the whole cluster.
• Tabs that need cluster-wide access a user does not have, such as Nodes and Datastore, now say so and name the permission to ask an administrator for, instead of showing an empty table with no explanation. A note at the top of the page lists everything the session could not read, so partial numbers are not mistaken for the whole cluster.

Changed

• The dashboard loads several times faster. OVTools was holding its own requests back well below what the cluster allows; on a test cluster the first load went from about 10 seconds to about 2.

Fixed

• Signing in repeatedly no longer leaves earlier sessions refreshing data in the background, which slowly added memory use and cluster traffic over a long-running day.
• Some health warnings could appear for every virtual machine at once when OVTools could not read a cluster-wide setting they depend on. Those checks now stay quiet instead of reporting a problem they cannot actually see.

Notes

• On plain Kubernetes with KubeVirt, a namespace-limited user needs either permission to list namespaces or the new OVTOOLS_NAMESPACES setting naming their namespaces. On OpenShift this works with no configuration.

ovtools v0.11.8

[0.11.8] 2026-08-05

Fixed

• On large clusters OVTools could run out of memory and restart in a loop at startup. The bundled deployment now gives the container more memory and keeps the runtime within that limit, so it starts and stays up on big fleets.

Changed

• Metrics now work for regular (non-admin) users and on clusters that route traffic through an HTTP proxy. When running inside the cluster, OVTools reads metrics from the internal monitoring service directly instead of the external address.

Added

• OVTools now records which monitoring endpoint it uses, and any connection error, in its log, so it is easier to tell why a metrics column shows "-".

Upgrade

• Reapply the OpenShift deployment manifest (oc apply -f deploy/openshift/) to pick up the new memory settings. Updating the image on its own keeps the old limits.

ovtools v0.11.7

[0.11.7] 2026-08-04

Fixed

• Signing in with an "@" in the username (an email-style OpenShift user) now works. The dashboard is no longer empty for those accounts. OpenShift's proxy shortens such names to the part before the "@", and OVTools now recovers the full username so the user's own permissions apply. This needs one extra read permission on the service account, so reapply the OpenShift deployment manifest for it to take effect.