Skip to main content

Release Notes

tip

Looking for breaking changes that might impact your Palette upgrade? Visit the Find Breaking Changes page for a filtered view of relevant updates.

To view release notes for a specific Palette version, use the version selector below.

November 15, 2026 - Release 4.10.44​

Security Notices​

Palette Enterprise​

Breaking Changes​

  • To improve security, Palette now enforces new permission and credential checks on the registry APIs. Automation that calls the following endpoints must use a principal that holds the required permission. Otherwise, the request fails with an authorization error.

    APIRequired permission
    GET /v1/registries/oci/{uid}packRegistry.get on the registry, and read permission on the cluster if the request sets clusterUid
    GET /v1/registries/metadatapackRegistry.list
    POST /v1/registries/pack/validate with a saved password or tokenpackRegistry.update on the registry that registryUid identifies
    POST /v1/registries/helm/validate with a saved password or tokenpackRegistry.update on the registry that registryUid identifies
    POST /v1/registries/oci/basic/validate with a saved password or tokenpackRegistry.update on the registry that registryUid identifies

    Palette uses the saved password or token of a registry only with the saved endpoint of that registry. If you change the endpoint of a registry, re-enter its password or token. Otherwise, the validate and update requests fail with an HTTP 400 response and the RegistryMaskedCredentialEndpointMismatch error code.

    Palette also no longer validates a saved private Helm registry with its saved credentials, even for a principal with packRegistry.update. The POST /v1/registries/helm/validate request fails with an HTTP 400 response and the HelmRegistryValidationForbidden error code. Enter the credentials in the request instead.

    Registry API responses now return an empty string for a password, token, or secret key that was never set, instead of a masked value. A masked value indicates that the secret is set.

  • To improve security, Palette now validates the property names in the filters and orderBy query parameters of Palette API list endpoints. A request fails with an HTTP 400 response and the FilterValidationFailure error code if a property name meets any of the following conditions:

    • It is empty or longer than 256 characters.

    • It contains characters other than letters, numbers, underscores (_), hyphens (-), forward slashes (/), and periods (.).

    • It contains an empty segment, such as metadata..name, or a segment that begins with a hyphen.

    • It references a protected field, such as a field that stores a password, key, or token, or a field nested under a protected field.

    Filters and sort keys on other properties, such as metadata.name, are unaffected. Update any automation that filters or sorts on a property that meets these conditions.

  • To improve security, system API responses now return stored secrets, such as passwords, access keys, and client secrets, as ********. A secret that was never set still returns an empty string. Automation that reads a secret from the system APIs must take it from its source instead, such as the identity provider, cloud storage account, or registry. To keep a stored secret when you update a system configuration, send ******** in that field of the PUT request.
  • To improve security, Palette now restricts the following user and tenant API requests. Update any automation that sends them.

    • GET /v1/users and GET /v1/tenants no longer accept status.login.mode or status.isActive in the filters or orderBy query parameters. Requests that use these fields fail with an HTTP 400 response and the FilterValidationFailure error code.

    • PATCH /v1/users/{uid} accepts only the add, replace, remove, and test operations on the /spec/firstName, /spec/lastName, /spec/emailId, and /spec/roles paths. Requests to patch other paths fail with an HTTP 400 response and the UserPatchPathNotAllowed error code. The move and copy operations fail with an HTTP 400 response and the UserPatchOpNotAllowed error code.

    • On self-hosted Palette and Palette VerteX, GET /v1/users/{uid}/activationlink now requires the user.create permission to read another user's activation link. Requests from users without this permission fail with an HTTP 403 response. Reading your own activation link is unchanged. To read other users' activation links, use an account that has the user.create permission.

  • Palette now rejects PUT /v1/clusterTemplates/{uid}/profiles requests whose scope does not match the scope of the cluster template. For example, a request with a ProjectUid header targeting a tenant-scoped cluster template returns an HTTP 403 response and the ResOperationForbidden error code. Omit the ProjectUid header for a tenant-scoped template, and set it to the UID of the project that owns a project-scoped template. Refer to Modify Cluster Templates.
  • Palette now rejects Kubernetes upgrades that skip a minor version or change the major version. Upgrade one minor version at a time, as the Kubernetes version skew policy requires, for example from 1.33 to 1.34 and then to 1.35. The check applies to the following changes:

    • Cluster updates, including pending and scheduled cluster template updates.

    • Updates to a cluster profile that clusters use. The profile cannot move more than one minor version ahead of the oldest Kubernetes version running on those clusters.

    • Cluster profile changes in a cluster template, and attaching a cluster to a cluster template.

    Rejected requests from the Palette UI, API, Terraform, and Crossplane fail with an HTTP 400 response and the K8sUpgradeNotSupported error code, and the cluster stays on its current version. Patch upgrades and imported clusters are not affected. Refer to Update a Cluster for more information.

Upgrade Notes​

  • User action required: The AKS-managed CSI drivers on AKS clusters deployed with a service principal cloud account now use the AKS control plane identity instead of the cloud account's service principal. This applies to the Azure Files, Azure Disk, and Azure Blob CSI drivers. If these clusters use storage outside the AKS node resource group, grant the AKS control plane identity access to that storage. This includes StorageClasses with a custom resourceGroup parameter, static persistent volumes in other resource groups, and disk encryption sets. Otherwise, provisioning or attaching these volumes fails with HTTP 403 errors. On self-hosted Palette and Palette VerteX, grant this access before you upgrade.

    After the upgrade, Palette restarts the enabled CSI drivers on each of these clusters, which takes a few minutes per cluster. Pods that start or move during the restart wait until the drivers are back, and running pods keep their volumes. If Palette cannot restart the drivers, it raises a warning event on the cluster. Refer to AKS CSI Drivers Fail to Restart After Palette Upgrade for more information.

Features​

  • Users who sign in to a Palette tenant with a password can now add a passkey as a second authentication factor from their My Profile > Security tab. After a user adds a passkey, Palette asks for it each time the user signs in with a password, in any tenant the user belongs to. Tenant administrators can also require every user who signs in to the tenant with a password to add a passkey within a configurable grace period (14 days by default). Palette does not ask for a passkey for SSO sign-ins or for API requests authenticated with API keys. For more information, refer to Manage Passkeys and Passkey Enforcement.
  • You can now deploy the Headlamp dashboard to Palette Virtual Clusters that use the Kubernetes distribution, in Proxied or Direct access mode. Headlamp authenticates through the OIDC configuration of the virtual cluster or its host cluster, or through Palette OIDC if neither has one. For more information, refer to Headlamp.
  • Palette now provides system macros that resolve to the OIDC configuration of a cluster, including the issuer URL, client ID, client secret, username and groups claims, and extra scopes. Reference them in pack values, for example {{.spectro.system.oidc.issuerurl}}, to place OIDC settings in the keys and format that your Kubernetes distribution requires. For more information, refer to OIDC Macros.
  • You can now authenticate Azure cloud accounts with a user-assigned managed identity instead of a service principal and client secret. Managed identity cloud accounts store no secret and work through a self-hosted Private Cloud Gateway (PCG) deployed in Azure. You can also assign managed identities to Azure IaaS and AKS clusters, so that the clusters access Azure without static credentials. The Spectro Cloud Terraform provider also supports managed identity cloud accounts. For more information, refer to Use Azure Managed Identities.

Improvements​

  • When you add a GCP backup location, you can now upload the JSON key file of the service account instead of pasting its contents. This matches how you add a GCP cloud account. For more information, refer to Add a Backup Location using Static Credentials.
  • Palette now records an audit log entry each time someone downloads a kubeconfig file for a cluster or virtual cluster from the Palette UI or the API, including failed and denied attempts. Each entry identifies the user, the cluster, and the kubeconfig file. For more information, refer to Kubeconfig Downloads.

Bug Fixes​

  • Fixed an issue that caused cluster cost, resource usage, and workload summary data to stop updating for clusters whose data grew past the 16 MB MongoDB document size limit.
  • Fixed an issue that caused Palette to add an empty auth value, Og==, to the containerd configuration of self-hosted Palette and Palette VerteX nodes when the custom image registry was configured with an empty username and password. Palette now leaves out the authentication configuration when both values are empty.
  • Fixed an issue that stopped cluster creation before any worker nodes were provisioned. This occurred when CNI pack pods that cannot run on control plane nodes, such as Cilium Hubble Relay and Hubble UI, were still pending 30 minutes after the CNI pack was installed. Pending CNI pack pods no longer block worker node creation.
  • Fixed an issue that left the deprecated capa-eks-control-plane-controller-manager Deployment in the cluster namespace of EKS clusters without add-on packs. The Deployment's pod restarted repeatedly. Palette now removes the Deployment and its related namespace-scoped resources.
  • Fixed an issue that caused a pack to stay in progress indefinitely when its Deployment rollout failed with ProgressDeadlineExceeded, for example, after a cluster profile update with incorrect pack values. Palette now reports the pack as failed and shows the error message from the Deployment.
  • Fixed an issue that caused Palette and Cluster API controller pods to keep running on workload node pools after you dedicated a node pool for system pods on an existing EKS cluster. Palette now labels the nodes in the dedicated node pool and schedules these controllers on them.
  • Fixed an issue that caused the Palette agent on a workload cluster to delete and recreate the namespaces that Palette created through cluster Role-Based Access Control (RBAC) settings or a workspace. This occurred when the Palette management plane briefly lost access to its database, and the recreated namespaces were empty.
  • Fixed an issue that prevented the cluster autoscaler pod from being scheduled on control plane nodes. When all worker node pools carried a NoSchedule taint, the pod stayed Pending after a node drain, repave, or upgrade, and node pool autoscaling stopped.
  • Fixed an issue that caused the AKS-managed CSI drivers on AKS clusters deployed with a service principal cloud account to use the cloud account's service principal instead of the AKS control plane identity. Dynamic volume provisioning failed with HTTP 403 errors when only the AKS control plane identity had storage permissions. Palette no longer creates the azure-cloud-provider secret in the kube-system namespace of AKS clusters. After the upgrade, Palette removes the secret that it created and restarts the drivers, and persistent volume claims stuck in Pending then bind automatically. Refer to the Upgrade Notes for more information.
  • Fixed an issue that caused repeated Operation 'cloudconfig.get' is forbidden errors after you switched projects while a cluster's details page was still loading. The errors appeared even though your role had the required permissions.
  • Fixed issues that caused clusters attached to a cluster template to revert to the default values of their profile variables after the template's cluster profile version changed. The issues occurred when you selected Review & Save on a tenant-scoped cluster template from a project, or when a Palette API or Terraform request did not list every profile variable.
  • Fixed an issue that caused the Palette UI to reject valid node pool Additional Labels whose value is a single character, such as workload:a, and to highlight the wrong label.
  • Fixed an issue that caused Headlamp to stay on its loading screen when you opened it from a cluster's Connect link through Spectro Proxy. This occurred when Palette took longer than 15 seconds to respond, for example, over a slow network connection.
  • Fixed an issue that caused cluster profile creation through the Palette API or Terraform to fail with ErrPackUidInvalid: Pack uid '<uid>' not found, even though the requested pack version existed. This occurred when the request specified a pack by registry, name, and version without a pack UID.
  • Fixed an issue that caused clusters to remain marked as cluster group hosts after their cluster group was deleted. Palette now clears this state when a cluster group is deleted. Upgrading Palette also clears it from clusters that are already affected.
  • Fixed an issue that caused clusters created from the Palette UI to have the Update worker pools in parallel option (updateWorkerPoolsInParallel) turned on, while clusters created through the Palette API or Terraform had it turned off. Clusters created from the Palette UI now have this option turned off by default.

Deprecations and Removals​

  • Palette removes the deprecated GET /v1/clusterprofiles endpoint. Use POST /v1/dashboard/clusterprofiles to list cluster profiles instead. Other /v1/clusterprofiles endpoints, such as those that create, update, and publish cluster profiles, continue to work. Update any automation, script, or integration that calls GET /v1/clusterprofiles before you upgrade, because those calls fail after the upgrade.
  • The POST /v1/users/kubectl/session, GET /v1/users/kubectl/session/{sessionUid}, and DELETE /v1/users/kubectl/session/{sessionUid} Palette API endpoints are now deprecated. These endpoints supported web-based kubectl, which is no longer available in Palette, so they have no replacement. Remove any calls to these endpoints from your automation, scripts, and integrations. Refer to the Deprecations table for the removal timeline.

Edge​

info

The CanvOS version corresponding to the 4.10.44 Palette release is 4.10.14.

Breaking Changes​

  • Local UI now rejects cluster updates that move Kubernetes more than one minor version ahead of the version last applied to the cluster or that downgrade Kubernetes after a successful upgrade. Local UI displays an error, and the cluster stays on its current configuration. To upgrade by more than one minor version, upload a cluster profile version for each minor version in turn. Requests to the Edge Management API fail with an HTTP 400 response and the K8sMultiMinorUpgradeNotSupported or K8sDowngradeAfterUpgradeNotSupported error code. Refer to Update Local Cluster for more information.

Upgrade Notes​

  • After you upgrade the Palette agent on an existing Edge cluster, Palette manages its default Cert Manager installation with Helm so that you can replace it with the Cert Manager pack. The Cert Manager pods might restart once during the change. Certificates that Cert Manager has already issued remain valid.

Features​

  • You can now add the Cert Manager pack to the cluster profiles of Edge clusters to customize Cert Manager, either when you deploy the cluster or later. The pack takes over the default Cert Manager installation from Palette, and Palette restores its default configuration if you remove the pack. Refer to Cert Manager Add-On for more information.
  • Edge clusters can now authenticate to a private Amazon Elastic Container Registry (ECR) repository without storing static AWS credentials on the Edge hosts. Palette issues short-lived ECR tokens that the Edge hosts use to pull provider images during cluster creation and upgrades. This feature is available only for centrally managed clusters in appliance mode. Refer to Deploy Cluster with an Amazon ECR Provider Registry for more information.
  • Centrally managed Edge Native clusters now support an update lock that holds cluster profile updates on the Edge hosts until an operator releases them in Local UI or through the Edge Management API. Each release lets one update through, and then the lock engages again. Turn on Lock Profile Updates when you create the cluster or later in the cluster settings. Refer to Lock Profile Updates for more information.
  • Local UI now lists the Local UI server certificate, the per-node mutual TLS (mTLS) certificates, and the registry certificates alongside the Kubernetes certificates of a cluster. For each certificate, Local UI shows the expiration date and whether Palette Edge renews it automatically. Local UI also warns you when certificates are close to their expiration date or have expired. Refer to View Cluster Certificates for more information.

Improvements​

  • You can now change how long each OS upgrade attempt can run on an Edge host with the system.upgradeTimeoutMinutes parameter in the OS layer of the cluster profile. The default limit is now 45 minutes, up from 15 minutes. Increase the limit for Edge hosts with slow network connections. Refer to Configure the OS Upgrade Timeout for more information.
  • When you sort the Edge host grid view by a column, Palette now orders all Edge hosts, so the order continues across pages instead of restarting on each page. This applies to the Edge Hosts tab and to the grid view that appears when you create an Edge cluster or add Edge hosts to an existing cluster. Refer to Edge Host Grid View for more information.
  • Palette and Local UI now display a warning when you create an Edge cluster without an overlay network, because IP address changes in a Dynamic Host Configuration Protocol (DHCP) network can cause cluster instability. Refer to Enable Overlay Network for guidance on when to enable an overlay network.
  • When you create a cluster in Local UI, you can now enter a Fully Qualified Domain Name (FQDN) in the Virtual IP Address (VIP) field, in addition to an IPv4 address. Refer to Create Local Cluster for more information.
  • The mTLS certificates that Edge hosts use to secure communication with each other are now valid for 10 years instead of 5 years. Refer to Renew Certificates for Local Clusters for more information.

Bug Fixes​

  • Fixed an issue that caused provider image downloads on Edge hosts with limited network bandwidth to fail repeatedly. The Palette agent now stops a download attempt only when it receives no data for 5 minutes, and the next attempt resumes the download. During a cluster update that changes the provider image, each download attempt still ends after 15 minutes. To resume a download, the Edge host needs free disk space of about three times the compressed size of the provider image. Palette also displays Edge host events during cluster provisioning when a provider image download starts, resumes, retries, or completes. For more information, refer to Troubleshooting - Edge.
  • Fixed an issue that caused the Palette agent to reset an Edge host removed from a node pool without draining it first, so Kubernetes did not enforce PodDisruptionBudgets. The Palette agent now drains each removed host before it resets the host. If a PodDisruptionBudget blocks the drain, the host stays cordoned in the cluster until the drain can complete, for example, after you move the blocking workload to another node.
  • Fixed an issue that caused the Palette Edge Interactive Installer to offer to continue after the installation media, such as a virtual CD or USB drive, was disconnected. The disk wipe failed with fork/exec /usr/sbin/wipefs: input/output error, and the installation later failed with a mount failed error. The installer now stops the installation and prompts you to reconnect the installation media and reboot the host.
  • Fixed an issue that prevented a control plane node from joining an Edge cluster that uses Palette eXtended Kubernetes - Edge (PXK-E) after a join attempt failed. The node kept the cluster's virtual IP (VIP) address from the failed attempt, so each retry failed with connect: connection refused errors for the VIP. This was most likely to occur when two control plane nodes joined the cluster at the same time. A control plane node no longer keeps the VIP after a failed join attempt.
  • Fixed an issue that caused Local UI to display only the message Client sent an HTTP request to an HTTPS server. when you opened it with an http:// address. Local UI now redirects HTTP requests on its port to HTTPS. Refer to Access Local UI for more information.
  • Fixed issues that allowed a Local UI session to continue after it expired because of inactivity. Local UI sessions now end after the idle time set in stylus.localUI.session.inactivityTimeoutInMinutes plus a 2-minute grace period. The stylus.localUI.session.refreshTimeoutInMinutes parameter now sets the maximum session length from login. Each token renewal now returns a new refresh token, and each refresh token expires after the same idle time and grace period. Edge Management API clients must use the refresh token from the latest renewal response. Refer to Edge Installer Configuration Reference for more information.
  • Fixed an issue that caused containerized workloads that use sudo or PAM to fail on Edge hosts built from Ubuntu 24.04 CanvOS images with a hardened /etc/shadow file, for example, one with file mode 0000. The workloads failed with sudo: PAM account management error: Authentication service cannot retrieve authentication info. Edge artifacts built with CanvOS version 4.10.14 or later include the fix.
  • Fixed an issue that caused audit log entries for cloud configuration updates on Edge clusters, such as NTP, VIP, or SSH key changes, to show the cluster UID instead of the cluster name in the Resource Name field.

VerteX​

Features​

  • Includes all Palette features, improvements, breaking changes, and deprecations in this release. Refer to the Palette section for more details.

Automation​

info

The Palette CLI version corresponding to the 4.10.44 Palette release is 4.10.16. Refer to CLI Tools for the download URL and checksum.

Features​

Bug Fixes​

  • Fixed an issue in the Spectro Cloud Terraform provider that left an empty cluster_profile entry in the Terraform state of cluster resources, such as spectrocloud_cluster_eks, after an apply that changed a cluster_profile block with both pack overrides and variables. As a result, later terraform plan runs reported a cluster profile change when nothing had changed.

November 15, 2026 - Component Updates​

The following components have been updated for Palette version 4.10.41.

ComponentVersion
Spectro Cloud Terraform provider0.30.6
Spectro Cloud Crossplane provider0.30.6

Improvements​

  • Added Kubernetes 1.34.13 and 1.35.10 to the FIPS and non-FIPS variants of the edge-k8s pack for Edge clusters.
  • Added Kubernetes 1.36 to the non-FIPS variant of the kubernetes-gke pack for GKE clusters.

Packs​

Pack NameLayerNon-FIPSFIPSNew Version
argo-cdaddon✅❌10.17.2
aws-cluster-autoscaleraddon✅❌1.37.0
cert-manageraddon✅✅1.21.3
cni-cilium-osscni✅❌1.20.4
edge-k8sk8s✅✅1.34.13
edge-k8sk8s✅✅1.35.10
falcoaddon✅❌9.2.2
flux-cdaddon❌✅2.20.0
headlampaddon✅❌0.45.0-rev1
karpenteraddon✅✅1.15.1
kubernetes-gkek8s✅❌1.36
lb-metallb-helmaddon✅❌0.16.2
opentelemetryaddon✅❌0.162.0
piraeus-netiface-builderaddon❌✅0.1.2
piraeus-netiface-builderaddon❌✅0.1.3
piraeus-operatorcsi✅✅2.12.1
piraeus-operator-addonaddon✅✅2.12.1
prometheus-agentaddon✅❌29.47.1
prometheus-operatoraddon✅❌93.2.0
reloaderaddon✅❌1.4.23
traefikaddon✅❌42.3.0

November 6, 2026 - Component Updates​

The following components have been updated for Palette version 4.10.13 - 4.10.34.

ComponentVersion
Artifact Studio4.10.8
Spectro Cloud Terraform provider0.30.5
Spectro Cloud Crossplane provider0.30.5
Palette Management Appliance4.10.33
VerteX Management Appliance4.10.33

Improvements​

  • Added K3s 1.34.12, 1.35.9, 1.36.5, and 1.37.1 to the non-FIPS variant of the edge-k3s pack for Edge clusters.

Bug Fixes​

  • Fixed an issue in the Spectro Cloud Terraform and Crossplane providers that caused Day-2 cloud_config updates to the spectrocloud_cluster_edge_native resource to fail when ssh_keys was not set.

Packs​

Pack NameLayerNon-FIPSFIPSNew Version
argo-cdaddon✅❌10.11.7
aws-efsaddon✅❌3.5.4
csi-aws-ebscsi✅✅1.67.2
csi-aws-efscsi✅❌3.5.4
csi-gcp-drivercsi✅❌1.28.1
edge-k3sk8s✅❌1.34.12
edge-k3sk8s✅❌1.35.9
edge-k3sk8s✅❌1.36.5
edge-k3sk8s✅❌1.37.1
flux-cdaddon❌✅2.19.1
flux-cdaddon✅❌2.19.1-rev1
flux-cdaddon✅❌2.20.0
kgatewayaddon✅✅2.5.0
prometheus-agentaddon✅❌29.40.3
prometheus-operatoraddon✅❌92.17.1
tetragonaddon✅❌1.8.0
traefikaddon✅❌42.2.0

October 30, 2026 - Component Updates​

The following components have been updated for Palette version 4.10.13 - 4.10.34.

ComponentVersion
Artifact Studio4.10.6
Palette Management Appliance4.10.33
VerteX Management Appliance4.10.33

Improvements​

  • Technical preview feature badgeTechnical preview feature badge Artifact Studio v2 is available as a Tech Preview. It adds self-service accounts, organization entitlements, and the Artifact Studio API. Your organization's entitlements determine which appliances, packs, and profiles you can download, and the API's /v2/ endpoints support pagination. Artifact Studio v1 is the default. Refer to Artifact Studio v2 and the Artifact Studio API for more information.
  • Added Kubernetes 1.37 to the FIPS and non-FIPS variants of the kubernetes-eks and kubernetes-aks packs for Amazon EKS and Azure AKS clusters.

Packs​

Pack NameLayerNon-FIPSFIPSNew Version
argo-cdaddon✅❌10.11.6
aws-albaddon✅❌3.6.1
aws-albaddon✅❌3.6.2
csi-aws-ebscsi✅✅1.67.1
external-secrets-operatoraddon✅❌2.14.0
falcoaddon✅❌9.2.1
harboraddon✅❌1.19.4
kubernetes-aksk8s✅✅1.37
kubernetes-eksk8s✅✅1.37
kyvernoaddon✅❌1.19.2
openobserveaddon✅❌1.0.3
prometheus-agentaddon✅❌29.39.0
prometheus-operatoraddon✅❌92.14.1
registry-connectaddon✅✅0.2.3
traefikaddon✅❌42.1.3

October 29, 2026 - PaletteAI VM Launchpad 4.10.25​

This release delivers version 4.10.25 of the PaletteAI VM Launchpad appliance, which runs on Palette 4.10.34.

Breaking Changes​

  • If you turned on Restrict Allowed VLANs (vlanFiltering.enabled) in an earlier release, review its settings before you upgrade. The VM VLAN Range (vlanFiltering.env.allowedVlans) replaces the vlanFiltering.env.vlanRange and vlanFiltering.env.enforceVlanRange settings, and VMO ignores the old values without a warning. After the upgrade, the VLAN filtering agent removes VLANs that no Network Attachment Definition (NAD) references. It also removes NAD VLANs that are outside the VM VLAN Range, including VLANs that NAD discovery added in an earlier release. Add every VLAN that your virtual machine networks use to the VM VLAN Range before you upgrade. The vlanFiltering.env.nadDiscovery and vlanFiltering.env.vlanDesiredStateSync parameters are deprecated. Refer to Upgrade Considerations.
  • Nodes on non-FIPS appliances now reserve memory for host services, and etcd and the Kubernetes API server now request memory on each control plane node of every appliance. This prevents a large virtual machine from taking the memory that the control plane needs. On non-FIPS appliances, the upgrade lowers the allocatable memory of each node by about 4 GiB and its allocatable ephemeral storage by 10 GiB. A node that is already near its limit can reject or evict pods when the Kubelet restarts during the upgrade. Before you upgrade, confirm that each node has room for the reservation and the new memory requests. Refer to Check Node Memory Before You Upgrade.

Features​

  • PaletteAI VM Launchpad now supports VLAN filtering on Linux Bridge networks. Virtual machines receive only the VLANs that their NADs assign, and only when the VM VLAN Range permits them. The appliance protects the VLANs that the nodes use for their own traffic, and the Networks page shows a VLAN filtering panel when a node has a problem. Refer to Bridge VLAN Filtering.

Improvements​

  • VMO now uses KubeVirt 1.9.0, which turns on its Beta features by default. VMO disables the Beta features that it has not validated, on both new installations and upgrades, and lists them in the new Disabled feature gates field on the KubeVirt Configuration page. KubeVirt 1.9.0 also deprecates the HotplugVolumes feature gate in favor of DeclarativeHotplugVolumes, which is on by default. VMO still includes HotplugVolumes in Feature gates by default on both new installations and upgrades, so remove it from Feature gates. Disk hot-plug keeps working as long as DeclarativeHotplugVolumes stays on. Refer to Feature Gate Considerations.
  • VMO now uses Containerized Data Importer (CDI) 1.66.1, which fixes importing a disk image to a DataVolume on a block device with a 4K sector size.
  • The Forwarding Token in the Logs section of the Metrics and Logs page now accepts a single Key=Value header, so you can forward logs to Datadog (DD-API-KEY=<datadog-api-key>), New Relic (api-key=<license-key>), or Grafana Cloud (Authorization=Basic <credentials>). Only the Metrics section accepts more than one comma-separated Key=Value pair. Refer to Metrics and Logs.
  • Applied critical security fixes to improve the appliance's security posture.

Bug Fixes​

  • Fixed an issue that caused a blank disk hot-plugged to a running virtual machine to always use the ReadWriteMany access mode and the Block volume mode. On storage that allows only the Filesystem volume mode, the DataVolume never became ready. The Create New Blank Disk form now has Access Mode and Volume Mode settings. Access Mode defaults to ReadWriteMany, and Volume Mode defaults to Auto (from StorageProfile), which uses the volume mode from the StorageProfile of the StorageClass.
  • Fixed an issue that caused a DataVolume to show the Unknown phase, with no reason, when its PersistentVolumeClaim (PVC) could not be provisioned, such as when the storage backend was full or offline. The DataVolume now shows a warning with the reason, and its details list the DataVolume conditions and the PVC events. Uploads to a DataVolume in this state now fail with the reason instead of timing out after 2 minutes.
  • Fixed an issue that caused the confirmation dialog to grow taller than the browser window when you deleted a resource that is in use, such as a Portworx StorageClass with attached DataVolumes. This cut off the warning, the list of attached resources, and the OK button. The dialog now fits the window, and long lists scroll.
  • Fixed an issue that caused long error messages to overflow their notification and hide its close button. Error notifications now also stay visible for 5 seconds.

October 28, 2026 - Release 4.10.34​

Improvements​

  • Applied further fixes for CVE-2026-111943, the eMudhra certificate authority (CA) compromise. Provider images and Edge installer ISOs that you build with CanvOS 4.10.11 or later no longer trust the affected root certificates. Refer to Security Advisory 024 for the affected deployments and recommended actions. Appliance mode Edge hosts that use images built with an earlier CanvOS version can use the workaround in Troubleshooting - Edge. The workaround applies to hosts that run Ubuntu, SUSE Linux Enterprise Micro, Red Hat Enterprise Linux, or Rocky Linux.

Edge​

info

The CanvOS version corresponding to the 4.10.34 Palette release is 4.10.11.

Automation​

info

The Palette CLI version corresponding to the 4.10.34 Palette release is 4.10.12. Refer to CLI Tools for the download URL and checksum.

October 24, 2026 - Release 4.10.33​

Breaking Changes​

  • To improve security, the GET /v1/spectroclusters/{uid}/assets/frpKubeconfig API endpoint now requires the cluster.adminKubeconfigDownload permission, the same permission that the admin kubeconfig endpoint requires. Users who have only the cluster.get permission, such as users assigned the Cluster Viewer role, can no longer download a kubeconfig file through this endpoint. Their requests fail with the OperationForbidden error. If your automation calls this endpoint, use an API key that belongs to a user with the Cluster Admin role or a custom role that includes the cluster.adminKubeconfigDownload permission. Refer to Kubeconfig for more information.

Improvements​

  • Applied fixes for CVE-2026-111943, the eMudhra certificate authority (CA) compromise. Refer to Security Advisory 024 for the affected deployments and recommended actions. Appliance mode Edge hosts that run Ubuntu, SUSE Linux Enterprise Micro, Red Hat Enterprise Linux, or Rocky Linux can use the workaround in Troubleshooting - Edge.

Bug Fixes​

  • Fixed an issue where provider image downloads on Edge hosts with limited network bandwidth repeatedly failed after 45 minutes with the error context deadline exceeded. The cluster remained in the Provisioning state, or the agent upgrade did not complete. The time limit for each download and extraction attempt is now 6 hours. For more information, refer to Troubleshooting - Edge.

Edge​

info

The CanvOS version corresponding to the 4.10.33 Palette release is 4.10.10.

Automation​

info

The Palette CLI version corresponding to the 4.10.33 Palette release is 4.10.11. Refer to CLI Tools for the download URL and checksum.

October 23, 2026 - Component Updates​

The following components have been updated for Palette version 4.10.13 - 4.10.32.

ComponentVersion
Artifact Studio4.10.5
Spectro Cloud Terraform provider0.30.4
Spectro Cloud Crossplane provider0.30.4
Palette Management Appliance4.10.27
VerteX Management Appliance4.10.27

Improvements​

  • Deprecated the override_kubeadm_configuration machine pool parameter for EKS, AKS, and GKE clusters in the Spectro Cloud Terraform and Crossplane providers. Palette does not support this parameter for managed Kubernetes clusters, so it has no effect. The Terraform provider now displays a warning when you set it.
  • Built and published Launchpad AI model recipes as standalone content archives in Artifact Studio, allowing new model support and vLLM inference engine updates to ship independently of core product release cycles.

Packs​

Pack NameLayerNon-FIPSFIPSNew Version
argo-cdaddon✅❌10.10.5
aws-efsaddon✅❌3.5.2
cni-aws-vpc-eks-helmcni❌✅1.23.2
cni-cilium-osscni✅❌1.20.3
cni-flannelcni✅✅0.28.10
csi-aws-ebscsi✅✅1.67.0
csi-aws-efscsi✅❌3.5.2
csi-longhorncsi❌✅1.13.0
csi-longhorn-addonaddon❌✅1.13.0
csi-vsphere-csicsi✅✅3.8.0
external-secrets-operatoraddon✅❌2.13.0
prometheus-agentaddon✅❌29.37.1
prometheus-operatoraddon✅❌92.11.0
volume-snapshot-controlleraddon✅❌8.7.0

October 22, 2026 - Release 4.10.32​

Improvements​

  • Improved the load times of the Users and Teams tabs on the Users & Teams page for tenants with a large number of users and teams. Palette now retrieves users and teams through the new paginated POST /v2/users/summary and POST /v2/teams/summary API endpoints. The POST /v1/users/summary and POST /v1/teams/summary endpoints continue to work as before.

October 16, 2026 - Component Updates​

The following components have been updated for Palette version 4.10.13 - 4.10.27.

ComponentVersion
Palette Management Appliance4.10.25
VerteX Management Appliance4.10.25

Improvements​

  • Added Kubernetes 1.37.0 to the Palette eXtended Kubernetes - Edge pack and updated the RKE2 pack to Kubernetes 1.33.13, 1.34.10, 1.35.7, and 1.36.3.

Packs​

Pack NameLayerNon-FIPSFIPSNew Version
argo-cdaddon✅❌10.10.3
aws-albaddon✅❌3.6.0
cni-calicocni❌✅3.33.0
cni-calico-azurecni❌✅3.33.0
csi-aws-ebscsi✅✅1.66.1
csi-gcp-drivercsi✅❌1.27.1
edge-k8sk8s✅✅1.37.0
karpenteraddon✅✅1.15.0
kubernetes-rke2k8s✅❌1.33.13-rke2r1-build20260612
kubernetes-rke2k8s✅❌1.34.10-rke2r1-build20260723
kubernetes-rke2k8s✅❌1.35.7-rke2r1-build20260723
kubernetes-rke2k8s✅❌1.36.3-rke2r1-build20260723
opentelemetryaddon✅❌0.159.0
prometheus-agentaddon✅❌29.36.1
prometheus-operatoraddon✅❌92.3.0
tigera-operatorcni✅❌3.33.0
traefikaddon✅❌41.7.1
zot-registryaddon❌✅0.1.125

October 15, 2026 - Release 4.10.27​

Improvements​

  • Strengthened the authorization checks on the Edge host APIs. Palette now verifies permissions against the specific Edge host in each request. If you call these APIs from scripts or automation, review the following changes in behavior.

    • Registration requests to POST /v1/edgehosts/register must include a registration token in the hostAuthToken field for every Edge host type, including when you re-register an existing Edge host. Requests without a token fail with the error Parameter 'hostAuthToken' should not be empty. Include a registration token associated with the Edge host's project in the request.

    • System-scope roles no longer grant Edge host permissions. Use a project-scope role that includes the required edgeHost permissions instead.

    • Updating the pairing key with PATCH /v1/edgehosts/{uid}/hostPairingKey or the checksum with PATCH /v1/edgehosts/{uid}/hostCheckSum requires the edgeHost.update permission on the Edge host in its project. The Tenant Admin role alone no longer grants this access outside the Edge host's project context.

    • Resetting an Edge host with PUT /v1/edgehosts/{uid}/reset requires the edgeHost.get permission on the Edge host, as well as the existing cluster permission.

    • GET /v1/edgehosts/{uid} now returns a masked value for the temporary SSH user's key in the sshKey field. Remote Shell sessions that you open in Palette are not affected.

    Edge hosts that register with a valid registration token are not affected. Apart from the masked sshKey value, users with the Project Admin, Project Editor, or Tenant Admin role who work in the Edge host's project are also not affected. For more information, refer to Edge Host Registration.

  • Applied security fixes to improve the platform's security posture.

Bug Fixes​

  • Fixed an issue where changing the number of nodes in a node pool from the cluster's Nodes tab turned off the Skip worker node update (Optional) option for that node pool. As a result, the next Kubernetes upgrade also upgraded the node pool's worker nodes. Changing the node count by editing the node pool configuration did not turn off the option.
  • Fixed an issue where the Teams tab on the Users & Teams page did not load for tenants with 100 or more teams. The tab stayed on a loading indicator, and the user service logs showed the error BSONObj exceeds maximum nested object depth.
  • Fixed an issue where control plane updates failed on clusters whose Kubernetes pack extraArgs include an unquoted integer of 1,000,000 or greater, such as quota-backend-bytes: 4294967296. After an upgrade to Palette agent version 4.10, the next cluster profile or pack change converted the value to scientific notation, such as 4.294967296e+09. The new control plane nodes then failed to join the cluster with the error etcd-join: the etcd member is not started. Palette now preserves integer values. Affected clusters repave their control plane nodes once after the upgrade to apply the corrected values.
  • Fixed an issue where the Palette Management Appliance and VerteX Management Appliance could not deploy new clusters when an administrator turned on Enable Certificate Rotation for the internal Zot registry after upgrading from version 4.9. Cluster deployments failed with the error failed to get in-cluster registry: in-cluster registry credential not found. Palette and Palette VerteX now update the registry certificate authority (CA) automatically, so the registry CA update script is required only on version 4.10.25. Refer to the Upgrade Notes for Palette and Palette VerteX for more information.
  • Fixed an issue where Palette kept duplicate machine records for an Edge host after its node re-registered with a new Kubernetes node UID, for example, after running kubectl delete node and restarting the Kubelet. The duplicate records appeared as extra nodes in Palette, and removing an outdated record could release the Edge host from its cluster and reset it. Palette now rejects a second machine record for an Edge host that already has a running machine in the same node pool, and the Palette agent no longer deletes the machine record of a running node. Palette also removes existing duplicate machine records automatically after the upgrade.
  • Fixed an issue on clusters that use a proxy server, including Edge clusters, where Palette task pods remained in the Terminating state and their tasks did not complete. Affected tasks include collecting node logs, applying OS patches, and updating NTP and SSH key settings, and the task jobs reported the error pod updates may not change fields other than. Palette now injects proxy settings, such as environment variables and CA certificates, only when a pod is created. Palette also keeps a pod's existing pod affinity and pod anti-affinity rules when it adds node affinity rules.
  • Fixed an issue on two-node Edge clusters where, after a failover, a demoted node could stop replicating data from the leader while the cluster continued to report a healthy status. A later failover could then promote the node with out-of-date data. Palette now stops the node's local Kubernetes and datastore processes before demoting it. Palette also generates an Error audit event when replication stalls and an Info audit event when replication recovers.
  • Fixed an issue on locally managed Edge clusters and Edge clusters that use Canonical Kubernetes where the node's <node>-public-key secret in the spectro-system namespace was not created if bootstrap hit a temporary failure, such as DNS or the Kubernetes API server not being ready. The task still reported success. Services that validate Local UI tokens then rejected users on that node, and /var/log/public-key-manager.log showed the error Failed to create secret <node>-public-key. The task now retries until the secret exists.

Edge​

info

The CanvOS version corresponding to the 4.10.27 Palette release is 4.10.9.

Automation​

info

The Palette CLI version corresponding to the 4.10.27 Palette release is 4.10.10. Refer to CLI Tools for the download URL and checksum.

October 9, 2026 - PaletteAI VM Launchpad 4.10.20​

This release delivers version 4.10.20 of the PaletteAI VM Launchpad appliance, which runs on Palette 4.10.25.

Breaking Changes​

  • Only the Platform Admin VMO role can now create, edit, and delete cluster-scoped instance types and preferences. Users with the Editor role can still view and select them, but can no longer change them. Assign the Platform Admin role to users who manage cluster-scoped instance types and preferences. Refer to VMO Roles.

Features​

  • PaletteAI VM Launchpad can now forward appliance metrics and logs to backends that accept the OpenTelemetry Protocol (OTLP) over HTTP, as well as to Splunk HTTP Event Collector (HEC). The Forwarding URL determines the protocol. The Logs section of the Metrics and Logs page now controls log delivery and has its own Forwarding Token and CA Certificate settings. Both sections have a new Skip TLS Verification setting, which is off by default. When Metrics Forwarding is enabled, the appliance also forwards its logs to the metrics receiver unless you configure the Logs section. Refer to Metrics and Logs.
  • Portworx appliances now support Everpure FlashArray as a storage backend for the Portworx storage cluster. Refer to Storage.

Improvements​

  • Non-FIPS appliances now use hardened images for the MetalLB, Local Path Provisioner, Traefik, Piraeus, and Registry Connect packs, and for the VMO pack components where a hardened image is available.

Bug Fixes​

  • Fixed an issue that caused namespaces enabled for virtual machines to disappear from the VMO UI after an update to the VMO pack, leaving only the default namespace.
  • Fixed an issue that caused the namespace list in the Upload ISO form to include namespaces that the user could not access, so the upload failed with a permission error after it started. The namespace lists in the Upload ISO and Build Golden Image forms now show only namespaces that the user can access.
  • Fixed an issue that prevented single-node non-FIPS Piraeus appliances from storing images in the appliance registry, so cluster bootstrap failed and pods remained in ImagePullBackOff.
  • Fixed an issue that hid the sub-steps and tips of a Getting Started wizard step after you completed the step. You can now reopen them.
  • Fixed an issue that caused VMO to delete uploaded images about 30 minutes after the upload started. VMO no longer deletes uploads that remain in the UploadReady phase, so remove abandoned uploads manually.
  • Fixed an issue that caused the Create VM wizard to create a Filesystem PersistentVolumeClaim (PVC) instead of a Block PVC when the root disk of a virtual machine template imports from a container registry. The disk import failed with qemu-img: Could not create '/data/disk.img': Permission denied, or the PVC remained Pending.
  • Fixed an issue that caused virtual machines created from an edited template to fail with disks[0].Name 'datavolume-os' not found.
  • Fixed an issue that prevented virtual machines on non-FIPS Portworx appliances from attaching volumes from a ReadWriteMany (RWX) storage class, which also caused live migration to fail. The appliance image now includes the NFS packages that Portworx requires, for both new installations and upgraded appliances.
  • Fixed an issue that marked the Set Up a Storage Class step of the Getting Started wizard as complete when you opened the Create Storage Class wizard and canceled it without creating a StorageClass.
  • Fixed an issue that caused VMO Manager to run out of memory and restart when multiple users logged in at the same time. Each restart logged out every user.
  • Fixed an issue that caused a stopped virtual machine to show Stopping indefinitely, with no error, when its disk could not be provisioned. The virtual machine now shows Stopped, and a banner explains why the disk cannot be provisioned.

October 9, 2026 - Release 4.10.25​

Security Notices​

  • To improve the security posture of using Palette as an identity provider, Palette now validates the callback URL on every authentication request. Refer to Register OIDC Callback URLs for more information.

  • Review the Security Bulletins page for the latest security advisories.

Palette Enterprise​

Breaking Changes​

  • Palette now enforces authorization checks on nine APIs that previously required no permission. Automation that calls these endpoints must use a principal that holds the required permission. Otherwise, the request fails with an authorization error.

    APIRequired permission
    PATCH /v1/cloudaccounts/{uid}/geoLocationUpdate permission on the cloud account
    POST /v1/spectroclusters/{uid}/workloads/syncUpdate permission on the cluster
    GET /v1/spectroclusters/{uid}/edge-native/edgeHostsGet permission on the cluster
    GET /v1/users/assets/vsphere/dnsMappingGet permission on the DNS mapping object
    GET /v1/users/assets/vsphere/dnsMappingsList permission on DNS mapping objects
    GET /v1/registries/oci/summaryList permission on registries
    GET /v1/packs/{packName}/registries/{registryUid}Get permission on the registry
    GET /v1/tenants/{tenantUid}/subscriptions/metadataTenant Admin role on the tenant
    GET /v1/tenants/{tenantUid}/subscriptions/aws/{uid}Tenant Admin role on the tenant

    The related endpoint POST /v1/spectroclusters/{uid}/workloads/{kind}/sync, which syncs a single workload kind, requires the same Update permission on the cluster as POST /v1/spectroclusters/{uid}/workloads/sync in the table above.

  • To improve security, virtual machine operations now also require the cluster.get permission on the cluster that hosts the virtual machine (virtualCluster.get for virtual clusters). Users who have a virtual machine role, such as Virtual Machine Power User or Virtual Machine User, but no role that includes cluster.get in the cluster's project can no longer view or manage the virtual machines on that cluster. Assign these users the Cluster Viewer role, or a custom role that includes cluster.get, in the cluster's project. If you upgrade from Palette 4.10.17-patch.5, operations other than viewing no longer require the cluster.update permission. Users who have the Cluster Editor role only for virtual machine operations now need only the Cluster Viewer role. Refer to VM User Roles and Permissions for more information.
  • Palette now validates the callback URL on every authentication request when it acts as an identity provider (IdP). Earlier releases accepted any callback URL. A tenant admin must register callback URLs that Palette does not allow by default. Authentication that succeeded before the upgrade fails until a tenant admin registers the callback URL, with some exceptions.

    You do not need to register the callback URLs after upgrade in the following situations.

    • The callback URL is on your Palette domain.

    • The callback URL is on the local machine, such as localhost or 127.0.0.1.

    • The callback URL belongs to a Virtual Machine Orchestrator or VM Migration Assistant cluster that exists at the time of the upgrade with its address set to a literal value or a variable.

      Palette registers the Virtual Machine Orchestrator and VM Migration Assistant callback URLs automatically at upgrade, including an address set with a variable. However, Palette cannot resolve an address set with a macro, so sign-in to those clusters fails until a tenant admin registers the address. If you change a cluster's callback URL after the upgrade, you must register the new value.

    For more information, refer to Register OIDC Callback URLs.

  • To improve security, Palette SaaS now rejects Splunk audit trail endpoints that point at private or reserved IP addresses. Palette checks the HTTP Event Collector (HEC) URL when you create, update, or validate a Splunk audit trail, and each time it sends audit logs to Splunk. Palette rejects a URL that is, or resolves to, a private (RFC 1918), localhost, link-local, multicast, carrier-grade NAT (100.64.0.0/10), or unspecified address, including the cloud instance metadata address 169.254.169.254, as well as a hostname that it cannot resolve. A rejected request returns the InvalidParam error code, and an existing audit trail that points at a blocked address stops delivering logs after the upgrade. Self-hosted Palette and Palette VerteX are not affected. If you use Palette SaaS with a Splunk instance on an internal IP address, expose a publicly routable HEC endpoint and update the audit trail to use it. Refer to Push Audit Trails to Amazon CloudWatch or Splunk for more information.
  • Palette now requires a role name when you create, update, or clone a role. A request with an empty or whitespace-only metadata.name fails with an HTTP 400 response and the InvalidName error code. Palette does not modify existing roles, but you must add a name to any unnamed role the next time you update it. Update any automation that creates or modifies roles through the API and might pass an unset name. The Spectro Cloud Terraform provider already requires a role name, so this change does not affect Terraform configurations.
  • Registry names must now be unique when you rename a pack or Helm registry, consistent with the existing check when you create a registry. Renaming a registry to a name that another registry in the same tenant already uses through PUT /v1/registries/pack/{uid} or PUT /v1/registries/helm/{uid} fails with an HTTP 409 response and the ResourceNameAlreadyExists error code. Name matching is case-insensitive and applies across all registry types, so a pack registry cannot take the name of a Helm or OCI registry. If two registries already share a name, rename one of them to a unique name. Update any automation or Terraform configuration that renames registries so that it uses unique names.
  • The private and synchronization settings of a Helm registry are now fixed when you create the registry, because Palette determines synchronization support from the private setting at creation. The Palette UI no longer offers the Synchronization toggle when you edit an existing Helm registry. A PUT /v1/registries/helm/{uid} request that changes spec.isPrivate fails with an HTTP 400 response and the HelmRegistryImmutableField error code. A request that omits isPrivate is treated as public, so an update to a private registry must include "isPrivate": true. To change either setting, delete the registry, add it again with the setting you need, and update any cluster profiles that reference it. Refer to Synchronization Behavior for more information.
  • K3s is no longer a supported distribution for virtual clusters and cluster groups. You cannot create new K3s cluster groups, and you cannot upgrade existing K3s cluster groups to vCluster 0.36.x. Existing K3s virtual clusters continue to run on their current vCluster version. To move to vCluster 0.36.x, create a new Kubernetes-based cluster group and migrate your workloads. Refer to Upgrade Cluster Groups for more information.

Upgrade Notes​

  • User action required: The Palette Management Appliance and VerteX Management Appliance can now rotate the TLS certificate of their internal Zot registry automatically. Rotation is turned off by default when you upgrade. After you upgrade, resume any paused agent upgrades, wait for the Palette agent upgrade to complete, turn on Enable Certificate Rotation in the Local UI, and then run the registry certificate authority (CA) update script. If you leave rotation turned off, the registry certificate eventually expires, unless you rotate it manually. Refer to the Upgrade Notes for Palette and Palette VerteX for more information.
  • Self-hosted Palette and VerteX now warn you to configure the image pull secret before you upgrade or install. If your installation pulls images from the Spectro Cloud registry instead of a local registry and the secret is not configured, Local UI (for appliances) and the system console (for Enterprise Clusters) display a warning. The warning is not enforced and does not block the upgrade or installation: the installation stays available and existing workload clusters are unaffected. However, new cluster deployments and day-2 operations that pull Spectro Cloud images fail until you configure the pull secret. Airgapped installations and installations that use a mirrored registry are not affected. For more information, refer to Configure Image Pull Secret.
  • If you have cluster groups on the K3s distribution, you cannot upgrade them to vCluster 0.36.x or later, because K3s is no longer a supported virtual cluster distribution. Existing K3s virtual clusters continue to run on their current vCluster version. Refer to Upgrade Cluster Groups for more information.
  • If your EKS clusters use the irsaRoles field of the EKS pack, add the iam:ListRoles action to the PaletteControllersEKSPolicy AWS IAM policy. Palette requires this action to delete the IAM Roles for Service Accounts (IRSA) roles it manages when you delete an EKS cluster. IRSA roles of clusters deleted without this permission remain in your AWS account. You can identify them by the palette.spectrocloud.com/cluster-uid tag and delete them manually. Refer to Controllers EKS Policy for the updated policy.

Features​

  • EKS and GKE clusters now support dedicating a worker node pool for system pods, both Palette and non-Palette. Select Dedicate node pool for system pods in the pool configuration to keep non-system workloads off the pool, reserving the other node pools in the cluster for your own workloads. Refer to Dedicated node pool for system pods for more information.
  • You can now attach a running, Palette-provisioned cluster to an existing cluster template through the Palette UI or the Spectro Cloud Terraform provider, bringing the cluster under template governance for future upgrades and policy enforcement. Refer to Attach an Existing Cluster to a Cluster Template.
  • Self-hosted Palette and VerteX can now export platform metrics to an external observability stack, such as Splunk, using an OpenTelemetry collector. You configure and manage the export from the Metrics tab in the system console. For more information, refer to Export Platform Metrics.

Improvements​

  • Beneath the pull secret configuration field, the system console now shows who configured the image pull secret and when.
  • When adding a worker node pool to a running cluster, you can now click Copy from Control Plane Pool to reuse the control plane pool's configuration in the new pool. The copied fields remain editable.
  • Palette upgraded the Cluster API providers that it uses internally, including the AWS, vSphere, Azure, and GCP providers. For the current provider versions, refer to Cluster API Versions.

Bug Fixes​

  • Fixed an issue that caused cluster backups that Velero reported as PartiallyFailed to display as Completed in Palette. These backups now display as Partially Failed in the Backups tab, where you can still delete or restore them.
  • Fixed an issue that caused Additional Labels configured on EKS worker node pools to reach new nodes only after the nodes joined the cluster. Workloads that selected nodes by these labels could fail to schedule, and node pool updates could time out while draining nodes. Palette now applies the labels when it creates the nodes.
  • Fixed an issue that caused the AWSManagedMachinePool resource of EKS worker node pools to list each Availability Zone twice when the node pool specified both Availability Zones and subnets.
  • Fixed an issue that could prevent Palette from applying add-on pack changes to EKS clusters and other clusters with managed worker node pools after a worker node pool recovered from a cloud provider failure, such as a deleted EC2 launch template version.
  • Fixed an issue that caused Palette to retry a pack uninstall indefinitely when the uninstall could not complete, for example, when removing the VMO pack from a cluster that still had virtual machines. Palette now stops after 12 failed attempts and reports a pack error with the reason PackUninstallStalled.
  • Fixed an issue that caused Palette to unexpectedly repave all control plane nodes after a cluster change that should not trigger a repave. This affected clusters first deployed with Palette 4.7 or earlier whose Palette agent was upgraded to version 4.8 or later. The agent upgrade itself did not trigger the repave. Instead, the repave started with the next cluster change, such as updating clientConfig.oidc-client-secret in the Kubernetes pack values. With this fix, these changes no longer repave control plane nodes.
  • Fixed an issue that caused sustained high CPU usage by the palette-controller-manager pod in clusters with add-on packs from the legacy Public Repo registry, registry.spectrocloud.com. Palette reconciled these packs every few seconds even when their configuration had not changed. Add-on packs from OCI registries were not affected.
  • Fixed an issue that caused audit trail events pushed to Splunk to omit the event time from the event body, so records exported or forwarded from Splunk did not show when the action occurred. The event body now includes a timestamp field in UTC.

Deprecations and Removals​

  • EKS Hybrid Nodes, deprecated in May 2026, are now disabled. Hybrid node configuration controls are greyed out in the Palette UI on imported EKS clusters, and API requests that enable or configure hybrid nodes are rejected. Existing clusters with hybrid nodes continue to operate, but you cannot add or modify hybrid configuration. Deploy your workloads to EKS clusters instead.
  • K3s is no longer a supported distribution for Palette and Palette VerteX virtual clusters or cluster groups. Palette and Palette VerteX virtual clusters now run on vCluster 0.36.x, and upstream vCluster removed K3s in 0.33. Existing K3s virtual clusters continue to run on their current vCluster version, but their cluster groups cannot be upgraded to newer vCluster versions. To move to a newer vCluster version, create a new Kubernetes-based cluster group and migrate your workloads.

Edge​

info

The CanvOS version corresponding to the 4.10.25 Palette release is 4.10.8.

Breaking Changes​

  • Palette now validates the control plane virtual IP (VIP) address of Edge Native clusters on cluster creation and day-2 updates. When the control plane endpoint type is VIP, the host must be a valid IPv4 address or a Fully Qualified Domain Name (FQDN). Otherwise, the request fails with an HTTP 400 response and the InvalidAddress error code. After a cluster is provisioned, you cannot change its control plane endpoint type or host. A day-2 update through PUT /v1/cloudconfigs/edge-native/{configUid}/clusterConfig that changes either value fails with an HTTP 403 response and the ControlPlaneEndpointUpdateForbidden error code. The Palette UI already enforces these rules. Update any Terraform configuration or automation that passes an invalid VIP or changes the VIP of a provisioned Edge Native cluster.
  • When you update an Edge Native node pool through the PUT /v1/cloudconfigs/edge-native/{configUid}/machinePools/{machinePoolName} endpoint, the number of entries in cloudConfig.edgeHosts must now match poolConfig.size. This safeguard prevents a partial host list from removing Edge hosts from the pool unintentionally, and applies to control plane and worker node pools. A request whose host count does not match the pool size fails with an HTTP 400 response and the ShouldBeEqualToValue error code. If your automation updates node-level taints or labels on some of the hosts in a pool, include every host in the pool in the request. To remove a host, reduce poolConfig.size and omit the host from cloudConfig.edgeHosts. Refer to Node-level Taints and Labels for Edge Native for more information.

Features​

  • On connected Edge Native clusters, you can now assign taints and labels to individual Edge hosts, so hosts in the same node pool can play different roles without needing separate pools. Set the values on each host row inside a pool's Edge Hosts section, either during cluster creation or when you edit a pool on an existing cluster. This supports patterns such as a three-node control plane where two hosts run application workloads and a third, lightweight host participates only in etcd quorum. Refer to Node-level Taints and Labels for Edge Native for more information.
  • The Palette UI now supports adding a temporary fourth control plane host to an Edge cluster node pool, so you can replace a control plane Edge host by adding the replacement before you remove the original and maintain etcd quorum throughout the operation. While the node pool contains four control plane hosts, Palette displays a warning that the cluster is in a temporary intermediate state. Refer to Replace a Control Plane Edge Host for the procedure, including how to verify etcd membership health before you remove the original host.
  • Technical preview feature badgeTechnical preview feature badge Connected Edge clusters now support Hadron, Spectro Cloud's minimal, immutable Edge operating system built on the CNCF Kairos project. Use the EdgeForge workflow to build Hadron-based installer ISOs and provider images. Refer to Build Hadron Images for more information.

Improvements​

  • The HTTPS certificate that Edge hosts generate for Local UI is now valid for 10 years instead of 5 years. Edge hosts that already have a Local UI certificate keep their existing certificate.
  • The Edge Hosts page now shows the Palette agent version installed on each Edge host and indicates when a newer version is available, so you can determine which hosts are behind and the version they would upgrade to. Refer to Identify the Latest Palette Agent Version for more information.

Bug Fixes​

  • Fixed an issue where the Edge Local UI reported an expired OS password as incorrect credentials at login. The login page now detects an expired password, whether an administrator or a PAM policy expired it, and lets you change the password there instead of requiring SSH or Palette TUI console access.
  • Fixed an issue that prevented add-on packs whose name contains a forward slash, such as packs sourced from a namespace in an OCI registry, from deploying to Edge clusters. The cluster reported FailedToDownloadPack events with failed to locate pack errors.
  • Fixed an issue that caused Kubernetes to fail to start, with errors such as bind: cannot assign requested address, after a single-node Edge cluster with an overlay network rebooted while the Edge host could not connect to Palette.
  • Fixed an issue that prevented OS and Kubernetes upgrades from starting on Edge clusters when the OS pack values in the cluster profile could not be parsed, for example, because of incorrect YAML indentation. The cluster now reports an OSPackValuesValid condition that includes the parsing error and line number.
  • Fixed an issue that prevented Kubernetes upgrades on Edge clusters that do not use Trusted Boot and were first deployed with Palette agent version 4.8.8 or earlier. The upgrade pods were rejected with the error maximum memory usage per Container is 1Gi, but limit is 1536Mi.
  • Fixed an issue that could prevent Edge hosts from resetting and returning to registration mode after their cluster was deleted, if a cleanup step failed during the reset. Affected hosts could not reconnect to Palette.
  • Fixed an issue that caused the certificate check that runs during Edge cluster creation to restart control plane components, including kube-apiserver, on clusters that use Palette eXtended Kubernetes - Edge (PXK-E). The restart could interrupt add-on packs that were still installing.
  • Fixed an issue that prevented automatic and manual certificate renewal on Edge clusters when any aggregated APIService in the cluster was unavailable, for example, when metrics-server was not ready or an APIService remained after uninstalling an add-on such as prometheus-adapter. Renewal failed with the error unable to retrieve the complete list of server APIs, and the certificates were not renewed.
  • Fixed an issue that prevented kube-vip from starting on agent mode Edge hosts that were added to a new cluster after a reset, for example, after their previous cluster was deleted. The control plane VIP of the new cluster was not claimed, and cluster provisioning did not complete.

VerteX​

Features​

  • Includes all Palette features, improvements, breaking changes, and deprecations in this release. Refer to the Palette section for more details.

Virtual Machine Orchestrator (VMO)​

VMO Pack​

Improvements​
  • VM migrations now support VDDK 9 for source VMs on VMware vSphere 7.0 and 8.0. Because Broadcom has removed public VDDK downloads, downloading VDDK requires a Broadcom account with the appropriate product entitlements and an active vSphere subscription.
Bug Fixes​
  • Fixed an issue that caused the removal of the VMO pack to delete the Containerized Data Importer (CDI) even when KubeVirt blocked the removal because virtual machines still existed on the cluster and the uninstall strategy was BlockUninstallIfWorkloadsExist. The removal now stops before it deletes any components, so disk imports, cloning, and VM provisioning continue to work.

Automation​

info

The Palette CLI version corresponding to the 4.10.25 Palette release is 4.10.9. Refer to CLI Tools for the download URL and checksum.

Features​

Docs and Education​

  • The documented Controllers EKS Policy now includes the iam:ListRoles action, which Palette uses to delete the IRSA roles it manages when you delete an EKS cluster. Refer to the Upgrade Notes for the action required on existing EKS clusters.

October 9, 2026 - Component Updates​

The following components have been updated for Palette version 4.10.25.

ComponentVersion
Spectro Cloud Terraform provider0.30.3
Spectro Cloud Crossplane provider0.30.3

Improvements​

  • Updated Kubernetes core and edge pack distributions across supported minor versions to include pause image enhancements and upgrade constraints.
  • Updated Kubernetes pack distributions with the latest maintenance releases and configuration improvements.
  • Added support for deploying the spectro-mgmt-plane management chart on Amazon EKS 1.36 clusters.

Bug Fixes​

  • Fixed an issue in MetalLB 0.16.1 Helm packs where liveness and readiness probe ports could not be customized.
  • Fixed an issue in Piraeus where RWX (NFS) volumes remained unmountable on single-node clusters when creating multiple RWX persistent volume claims simultaneously.

Packs​

Pack NameLayerNon-FIPSFIPSNew Version
argo-cdaddon✅❌10.9.2
argo-cdaddon✅❌10.9.6
aws-efsaddon✅❌3.5.1
calico-network-policyaddon✅❌3.33.0
cni-aws-vpc-eks-helmcni✅❌1.23.0
cni-aws-vpc-eks-helmcni✅❌1.23.2
cni-calicocni✅❌3.33.0
cni-calico-azurecni✅❌3.33.0
csi-aws-ebscsi✅✅1.66.0-rev1
csi-aws-efscsi✅❌3.5.1
csi-azurecsi✅✅1.36.0
csi-longhorncsi✅❌1.13.0
csi-longhorn-addonaddon✅❌1.13.0
edge-k8sk8s✅✅1.34.11
edge-k8sk8s✅✅1.36.4
external-dnsaddon✅❌0.23.0
external-secrets-operatoraddon✅❌2.11.0
falcoaddon✅❌9.2.0
harboraddon✅❌1.19.2-rev2
kubernetesk8s✅✅1.34.11
kubernetesk8s✅✅1.35.8
kubernetesk8s✅✅1.36.4
kubernetesk8s✅✅1.37.0
lb-metallb-helmaddon✅❌0.16.1-rev2
nvidia-gpu-operator-aiaddon✅❌26.7.1
openobserveaddon✅❌1.0.2
piraeus-operatorcsi✅❌2.12.0
piraeus-operator-addonaddon✅❌2.12.0
prometheus-agentaddon✅❌29.35.0
prometheus-operatoraddon✅❌91.8.1
prometheus-operatoraddon✅❌91.9.0
traefikaddon✅❌41.6.1
zot-registryaddon✅❌0.1.125

October 6, 2026 - Release 4.10.17-patch.5​

Palette 4.10.17-patch.5 is a cumulative security update for Palette 4.10.17. It includes every fix from the earlier 4.10.17 security patches, so upgrading to 4.10.17-patch.5 also applies those fixes. The patch suffix is a build sequence number. It is not a severity rating or a count of fixes.

The following table lists the release date of each patch in this update.

VersionRelease Date
4.10.17-patch.5October 6, 2026
4.10.17-patch.4October 3, 2026
4.10.17-patch.2October 2, 2026
4.10.17-patch.1October 1, 2026

Breaking Changes​

  • To improve security, virtual machine operations now also require permissions on the cluster that hosts the virtual machine. Viewing virtual machines and snapshots requires the cluster.get permission, and all other operations, such as starting, stopping, cloning, migrating, and creating snapshots, require the cluster.update permission (virtualCluster.get and virtualCluster.update for virtual clusters). Users assigned the Virtual Machine Power User or Virtual Machine User role together with the Cluster Viewer role can still view virtual machines but can no longer perform other operations on them. Assign these users the Cluster Editor role, or a custom role that includes cluster.update, in the cluster's project. Refer to VM User Roles and Permissions for more information.

Improvements​

  • Applied security fixes to improve the platform's security posture.

September 25, 2026 - PaletteAI VM Launchpad 4.10.13​

This release delivers version 4.10.13 of the PaletteAI VM Launchpad appliance, which runs on Palette 4.10.17.

Features​

  • PaletteAI VM Launchpad now ships as high-availability appliance variants backed by either Piraeus/LINSTOR or Portworx storage. The Piraeus/LINSTOR variant is available in both FIPS and non-FIPS builds; the Portworx variant is available as non-FIPS only. Refer to Install PaletteAI VM Launchpad and Storage for variant selection and storage configuration.
  • The appliance UI now includes a Force Stop action for virtual machines. Use Force Stop to immediately power off a virtual machine whose guest operating system does not respond to a graceful Stop. Refer to Manage Virtual Machines.
  • PaletteAI VM Launchpad can now forward appliance metrics to Splunk, so you can retain metrics longer and view them in your own tooling. Refer to Metrics and Logs.
  • The CDI upload proxy and KubeVirt export proxy are now exposed, so virtctl image-upload and virtual machine disk exports work through the appliance.
  • The appliance now generates the OIDC client secrets for VMO, Headlamp, and the Keycloak admin account per appliance, instead of setting them through hidden profile variables.

Improvements​

  • The virtual machine resource configuration form now includes a Memory Limits field, corrects a misleading CPU request placeholder, and allows editing CPU and memory requests and limits on day 2.
  • The Local UI installation wizard now includes a verify-password field for the VMO password.
  • The ephemeral-storage warning banner can now be dismissed.
  • The appliance now bundles Palette VMM version 4.10.x.
  • All Helm hook workloads now run as non-root, with the exception of the privileged Multus cleanup job.
  • The spectro-kubectl images have been rebuilt on Go 1.26.7 to address security vulnerabilities.

Bug Fixes​

  • Fixed an issue that caused the CDI upload proxy and KubeVirt export proxy ingress to unconditionally render a Traefik configuration, which broke the Nginx reference architecture. This affected VMO pack versions 4.10.2 through 4.10.6.
  • Fixed an issue that caused failures when editing a virtual machine's CPU configuration.
  • Fixed an issue that caused cloud-init networkData to be silently dropped. The appliance now sets networkDataSecretRef so the network configuration is applied.
  • Fixed an issue that caused the virtual machine creation wizard to skip a network interface with bootOrder=2 when UEFI, Secure Boot, and TPM were enabled.
  • Fixed an issue that prevented Windows virtual machines using windows.* cluster preferences from scheduling on nodes that did not have Hyper-V enlightenment labels.
  • Fixed an issue that prevented creating Windows virtual machines from a template.
  • Fixed an issue that prevented Sysprep from running when the QEMU guest agent installation was enabled on Windows.
  • Fixed an issue that caused golden images to build non-bootable disks, produce an invalid template, or ignore the EFI boot checkbox.
  • Fixed an issue that caused the Golden Image Builder to mis-detect the guest operating system from the ISO filename. Refer to Golden Images.
  • Fixed an issue that caused DataVolumes to display ImportScheduled instead of ImportInProgress while an import was running.
  • Fixed an issue that caused image upload in Direct mode to fail on MAAS and Edge clusters.
  • Fixed an issue that caused registry (zot) images to disappear from the UI after a pod restart.
  • Fixed an issue that caused the Storage HA Not Available banner to appear on clusters that do not use LINSTOR.
  • Fixed an issue that caused some Traefik routes to be inaccessible.
  • Fixed an issue that caused Headlamp SSO to land on a 404 tenantApps URL after Keycloak login through direct load balancer access.
  • Fixed an issue that caused the Keycloak configure hook to extract the wrong client ID, which caused a CrashLoopBackOff.
  • Fixed an issue that caused the namespace list to include namespaces a user could not access. VMO Manager now scopes the namespace list to the user's Kubernetes token.
  • Fixed an issue that prevented cAdvisor metrics from reaching Victoria Metrics, which caused missing dashboard telemetry. This was a regression introduced in 4.9.16.
  • Fixed an issue that caused the audit cleanup pod to error out.
  • Fixed an issue that caused the Resource Summary to show usage against allocated resources instead of against cluster capacity.
  • Fixed an issue that caused the Total vCPUs count to be miscalculated. The count now multiplies cores, sockets, and threads.
  • Fixed an issue that caused the node CPU history chart to drop to false 0% values.
  • Fixed an issue that caused the dashboard Issues count to apply an incorrect virtual machine status filter when selected.
  • Fixed an issue that prevented dashboard widget changes from persisting in Palette mode.
  • Fixed an issue that caused transitional states, such as Stopping, to be counted under Provisioning.
  • Fixed an issue that caused edit modals to show a false unsaved-changes prompt when pressing Escape.
  • Fixed an issue that caused incorrect breadcrumb labels and a stale System > Storage reference.
  • Fixed an issue that prevented special characters and uppercase letters from pasting correctly in the VNC console.

September 24, 2026 - Component Updates​

The following components have been updated for Palette version 4.10.16 - 4.10.17.

ComponentVersion
Spectro Cloud Terraform provider0.30.2
Spectro Cloud Crossplane provider0.30.2

Bug Fixes​

  • Fixed an issue in the Spectro Cloud Terraform provider where concurrent resource reconciliation could leak the ProjectUid header across tenant and project scopes, leading to permission errors when managing mixed-scope resources.

Packs​

Pack NameLayerNon-FIPSFIPSNew Version
aws-efsaddon✅❌3.5.0
cert-manageraddon✅✅1.21.2
cni-cilium-osscni✅❌1.20.2
csi-aws-efscsi✅❌3.5.0
csi-azurecsi✅✅1.35.0
edge-canonicalk8s✅❌1.35.8
edge-k8sk8s✅✅1.35.8
flux-cdaddon✅❌2.19.1
kyvernoaddon✅❌1.19.1
prometheus-agentaddon✅❌29.31.1
prometheus-operatoraddon✅❌91.4.1
traefikaddon✅❌41.6.0

September 18, 2026 - Component Updates​

The following components have been updated for Palette versions 4.10.16 - 4.10.17 and 4.9.51 - 4.10.17.

ComponentVersion
Artifact Studio4.10.4
Spectro Cloud Terraform provider0.30.1
Spectro Cloud Crossplane provider0.30.1
Palette Management Appliance4.10.17
VerteX Management Appliance4.10.17

Improvements​

  • Palette now blocks access to production SaaS, Artifact Studio, and product downloads from comprehensively sanctioned jurisdictions at the network edge, to meet export control compliance requirements.

Packs​

Pack NameLayerNon-FIPSFIPSNew Version
argo-cdaddon✅❌10.9.1
csi-aws-ebscsi✅✅1.66.0
csi-longhorncsi✅❌1.12.1
csi-longhorn-addonaddon✅❌1.12.1
csi-portworx-genericcsi✅❌3.7.0
external-dnsaddon✅❌0.22.0
kgatewayaddon❌✅2.2.2
portworx-add-oncsi✅❌3.7.0
prometheus-agentaddon✅❌29.30.0
prometheus-operatoraddon✅❌91.4.0
reloaderaddon✅❌1.4.22

September 17, 2026 - Release 4.10.17​

Improvements​

  • Added FIPS provider images to the content bundle when building content with the --fips flag in Palette CLI.

Bug Fixes​

  • Fixed an issue where mixed compression formats in spectro-ui container image manifests caused unpack failures in containerd.
  • Fixed an issue where airgapped Enterprise Cluster installations failed to pull cert-manager images because the generated manifest referenced the upstream registry and an unpublished image tag instead of the local registry.
  • Fixed an issue that caused the Palette CLI shell completion command to fail.

Automation​

info

The Palette CLI version corresponding to the 4.10.17 Palette release is 4.10.4. Refer to CLI Tools for the download URL and checksum.

September 11, 2026 - Component Updates​

The following components have been updated for Palette version 4.10.13 - 4.10.16.

ComponentVersion
Spectro Cloud Terraform provider0.30.1
Spectro Cloud Crossplane provider0.30.1
Palette Management Appliance4.10.14
VerteX Management Appliance4.10.14

Improvements​

  • Added support for deploying the NVIDIA GPU operator in airgapped Edge clusters using provider images with pre-installed NVIDIA drivers.
  • Deprecated Portworx pack versions 3.3 and older to align with upstream vendor support lifecycle policies.
  • Made admin_kube_config reads during cluster Read() operations conditional, gracefully handling 403 forbidden responses when users lack download permissions.

Bug Fixes​

  • Fixed an issue in the Harbor pack where Palette-managed projects were created as public without a configuration setting to set them to private.
  • Fixed an issue where running the Palette CLI shell completion command failed if the user had not first executed palette login.

Packs​

Pack NameLayerNon-FIPSFIPSNew Version
argo-cdaddon✅❌10.8.2
cni-antreacni✅❌2.7.0
cni-calicocni✅✅3.32.2
cni-calico-azurecni✅✅3.32.2
cni-cilium-osscni✅❌1.20.1
csi-aws-ebscsi✅✅1.65.0
external-secrets-operatoraddon✅❌2.10.0
prometheus-agentaddon✅❌29.27.2
prometheus-operatoraddon✅❌90.0.0
tigera-operatorcni✅❌3.32.2
traefikaddon✅❌41.5.0

September 10, 2026 - Release 4.10.16​

Bug Fixes​

  • Fixed an issue where stylus-operator did not retry the registry connection when the zot-secret appeared late, which left pack images missing from Zot.
  • Fixed an issue where Edge Kubernetes upgrades to version 1.33.5 stalled following a hostname-validation reboot due to x509 certificate verification failures during Stylus pack download and edge login.
  • Fixed an issue on two-node Edge clusters where PostgreSQL failed to start after a provider image swap, because the TLS key pair paths were split across persistent and ephemeral storage.
  • Fixed an issue where Private Cloud Gateway (PCG) migrations did not transfer IP pool ownership.
  • Fixed an issue where Ally upgrades retriggered on-boot operating system patching drains across nodes, causing nodes with failed drains to remain cordoned indefinitely.
  • Fixed an issue during vSphere Private Cloud Gateway (PCG) migrations where the cloud account's overlord UID was not transferred, preventing migrated clusters from being deleted.
  • Fixed an issue where two-node cluster creation deadlocked permanently when the API VIP was unavailable due to nodes mutually fencing each other and evicting kube-vip.

Edge​

info

The CanvOS version corresponding to the 4.10.16 Palette release is 4.10.4.

September 8, 2026 - Release 4.10.14​

Bug Fixes​

  • Fixed an issue where some OIDC sign-ins failed with an OidcRedirectUriNotAllowed error. This affected kubectl oidc-login authentication from headless hosts and access to the Virtual Machine Orchestrator (VMO) console.

September 6, 2026 - Release 4.10.13​

Security Notices​

Palette Enterprise​

Breaking Changes​

  • Palette now validates user-supplied cluster profile and app profile versions against the Semantic Versioning specification when a profile is created or updated through the Palette UI, API, Terraform provider, or Crossplane provider. Values such as 1.2.3 and 1.2.3-rc.1 are accepted; values such as 2.2.2.develop or V0.0.1 that earlier releases accepted are now rejected whenever a version is set (on create, on clone, when creating a new profile version, or when changing the version of an existing profile). Existing profiles carrying a malformed version continue to function, and no pre-upgrade or post-upgrade action is required. If you update a profile's version to a valid value, later attempts to set a malformed value on that profile fail, including reverting to the original value. These new requirements do not apply to external registry and chart tags, including Zarf UDS tags. For the accepted format, refer to Version a Cluster Profile and Version an App Profile.
  • The OCI Helm registry validation endpoint now returns an error code that reflects the actual failure. Previously, any validation failure (including request timeouts, DNS failures, and TLS errors) returned InvalidRegistryCredentials. Timeout and cancellation failures now return a timeout error, connectivity failures return an unreachable error, and InvalidRegistryCredentials is reserved for authentication failures. Existing registries and credentials are unaffected, and no pre-upgrade or post-upgrade action is required. If you have automation that treats InvalidRegistryCredentials as the catch-all validation failure, narrow it to authentication failures and handle timeout and connectivity errors with a retry instead. Refer to Registries for more information.
  • Palette removes the deprecated GET /v1/projects endpoint. Use POST /v1/dashboard/projects instead, which supports the filtering and sorting that the removed endpoint lacked. Update any automation, script, or integration that calls GET /v1/projects before you upgrade, because those calls fail after the upgrade.
  • Creating or updating a registry now requires a non-empty metadata.name. Palette validates the field at the API layer and rejects a missing or empty name with an HTTP 400 response that identifies metadata.name. Earlier releases performed no field-level check, so you could create a registry with an empty name, and only the uniqueness constraint on the tenant and name pair rejected it later, which surfaced as a misleading duplicate-name error. The validation applies to pack, Helm, basic OCI, and ECR OCI registries. Existing registries are unaffected, and no pre-upgrade or post-upgrade action is required. Update any automation or Terraform configuration that templates a registry payload and might pass an unset name. Refer to Registries for more information.
  • Updating an OCI registry that uses the Helm or Zarf provider no longer accepts a change to the synchronization setting. Palette treats synchronization as a create-only setting for these registries, and the Palette UI does not offer the toggle after creation. The update API accepted the change and returned success without applying it. The API now rejects the request with an HTTP 400 response and the message Registry synchronization setting cannot be changed after the registry is created. This applies to both basic and ECR OCI registries. If you manage registries with the Terraform provider, remove any change to is_synchronization on an existing spectrocloud_registry_oci resource. Such a change previously produced persistent terraform plan drift, because each apply reported success without taking effect. Refer to Registries for more information.
  • The Edge host search API, POST /v1/dashboard/edgehosts/search, now has its own sort field contract. The accepted sort fields are name, state, healthState, creationTimestamp, and lastModifiedTimestamp. Palette removes the environment, clusterName, and clusterState fields, which do not apply to Edge hosts. A request that sorts on one of the removed fields now returns an HTTP 422 response instead of an HTTP 500 response. Sorting on healthState was advertised in earlier releases but did not return results, and now works. Remove the unsupported fields from any automation or integration that sorts Edge host search results.

Features​

  • Apache CloudStack support is now generally available. Palette removes the ApacheCloudStack feature flag and enables Apache CloudStack for all tenants. You can deploy and manage Kubernetes clusters in your Apache CloudStack environment through a Private Cloud Gateway (PCG). Refer to Apache CloudStack for more information.
  • Canonical Kubernetes (CK8s) clusters on MAAS now support SSH key management. You can configure SSH Keys on the cluster's cloud configuration during cluster creation and on Day-2 through the Palette UI, API, Terraform provider, and Crossplane provider, on both Palette and Palette VerteX. Palette injects the keys into the spectro user's ~/.ssh/authorized_keys on every control plane and worker node, and preserves any users that MAAS or the machine image already configured. Changing the keys on a deployed CK8s cluster repaves the cluster nodes. Refer to SSH Keys on MAAS Cluster Nodes for more information.
  • You can now use dynamic credentials when backing up edge-native, AKS, vSphere, and other non-AWS clusters to AWS S3, which allows you to authenticate backups without storing long-lived static keys. Previously, only AWS IaaS and EKS clusters supported dynamic credentials. Refer to Add Backup Location using Dynamic Credentials for more information.
  • Canonical Kubernetes (CK8s) clusters on MAAS now support Network Time Protocol (NTP) server configuration. You can configure NTP Servers on the cluster's cloud configuration during cluster creation and on Day-2 through the Palette UI, API, Terraform provider, and Crossplane provider, on both Palette and Palette VerteX. The servers you specify replace the NTP configuration that MAAS provides to each control plane and worker node. Changing the servers on a deployed CK8s cluster repaves the cluster nodes. Refer to NTP Servers on MAAS Cluster Nodes for more information.
  • Palette now supports the option to skip worker node upgrades on Azure IaaS, GCP IaaS, and Apache CloudStack clusters. For example, if you have worker pools running critical databases or real-time processing services, you can enable this option to maintain service continuity during control plane upgrades, then schedule worker node updates during planned maintenance windows.

    The version difference between the control plane and worker nodes must not exceed the N-3 minor version skew supported by Kubernetes. Palette enforces this during cluster profile updates and blocks you from updating if you attempt to exceed the N-3 threshold.

  • You can now view whether the image pull secret that is specified during install has propagated properly to all clusters. Clusters cannot pull security-hardened images from Spectro Cloud's OCI registries until they have the pull secret. Refer to Monitor Propagation of the Image Pull Secret for more information.

Improvements​

  • The Host LXD-based control planes and Use LXD VMs toggles in cluster and node configuration now display a tooltip clarifying that LXD-based VMs on MAAS are supported only with Palette eXtended Kubernetes clusters. Enabling LXD with other Kubernetes distributions results in deployment failures.
  • Headlamp, the modern replacement for the deprecated Kubernetes Dashboard, is now available on imported clusters as well as Palette-managed clusters.
  • Palette now updates IAM roles in AWS when the irsaRoles parameter of the EKS pack is updated. Previously, specified IAM roles were created in AWS, but not updated when the irsaRoles field changed. Edits to irsaRoles in the cluster profile take effect on the next reconcile. For example, policies are attached and detached, service account changes update the trust policy, and removing an entry deletes the underlying IAM role. When you delete an EKS cluster, Palette deletes every IRSA role that it manages for that cluster.

    If you have existing EKS clusters that use the irsaRoles field, you should add the iam:ListRolePolicies action to the PaletteControllersEKSPolicy AWS IAM policy so that Palette can fully manage the lifecycle of the roles.

  • The cert-manager chart deployed with the Palette management plane has been upgraded from version 1.14 to version 1.20.2.

Bug Fixes​

  • Fixed an issue where using the Palette UI to open and save a cluster profile created with the API or Terraform could reorder its packs, surfacing as unexpected terraform plan drift for profiles managed as code.
  • Fixed an issue that caused the import-presetup-logs ConfigMap to grow without limit each time the setup job for importing a cluster ran. The ConfigMap eventually exceeded the maximum request size, causing request is too large errors and leaving the management plane API server unresponsive.
  • Fixed an issue that caused the palette-webhook deployment to stall with a pod stuck in Pending and the deployment reporting ProgressDeadlineExceeded on Amazon EKS clusters with three or fewer nodes. The stall occurred both when a Palette agent upgrade updated the webhook and after a cluster pivot.
  • Fixed an issue that caused the CoreDNS deployment in the kube-system namespace to be replaced with a virtual cluster CoreDNS configuration on MAAS clusters that host virtual clusters. Cluster DNS resolution failed and the cluster reported an unknown status in Palette.
  • Fixed an issue that prevented the Palette UI from updating when all tags were removed from a cluster profile. The removed tags continued to display until the browser was refreshed.
  • Fixed an issue that caused intermittent 504 Gateway Timeout responses when attaching cluster profiles to a cluster through the PUT /v1/spectroclusters/{uid}/profiles API while that cluster was actively reporting its status. Attaching several profiles to the same cluster in quick succession was the most common trigger.
  • Fixed an issue that caused multi-line error messages to appear truncated in a cluster's Events tab, so an event showed only "Reconciler error" err=< without the failure description that followed it. The complete message is now recorded as a single event. Only error-level events were affected.
  • Fixed an issue that caused changing a cluster-level tag on a healthy Amazon EKS cluster to emit a ClusterUpgradeTriggered event and replace every node in the cluster's worker node pools. Pipelines that update tag values on each run, such as compliance tagging, could therefore repave nodes repeatedly. Changing tags directly on a worker node pool still replaces that pool's nodes.
  • Fixed an issue that caused sustained high CPU usage by the Palette management plane. The management plane repeatedly reconciled packs whose configuration had not changed, consuming approximately 1.3 vCPU per affected cluster continuously. Cluster provisioning and pack functionality were not affected.
  • Fixed an issue where OS patching on Ubuntu cluster nodes could leave a node in Ready,SchedulingDisabled and unable to accept workloads, and could prevent scheduled patches from running. Palette now requires OS patch schedules to run no more frequently than hourly.
  • Fixed an issue that caused the cert-manager add-on pack to fail with ChartInstallFailed when the pack enabled Gateway API support before the Gateway API custom resource definitions existed on the cluster. Palette now applies the setting once those definitions are available.

Deprecations and Removals​

  • K3s is deprecated as a Palette Virtual Cluster distribution and will be removed in a future Palette release. K3s was deprecated in Palette 4.8.0, when the cluster group distribution selector in the Palette UI began marking K3s as deprecated and moved the Kubernetes distribution to the top of the list. Existing virtual clusters that use K3s continue to run. Use the Kubernetes distribution when you create new cluster groups and virtual clusters. Refer to Palette Virtual Clusters for more information.

Edge​

info

The CanvOS version corresponding to the 4.10.13 Palette release is 4.10.3.

Breaking Changes​

  • The Palette agent now applies a fixed 45-minute time limit to each attempt to download and extract the provider image on an Edge host. The agent does not resume partial downloads, so each new attempt starts from the beginning. On connections with limited bandwidth, attempts can repeatedly reach the limit, which leaves the cluster in the Provisioning state or prevents the agent upgrade from completing. If your Edge hosts have limited bandwidth, make the provider image available on the host instead of downloading it. For the symptoms and steps, refer to the provider image download times out troubleshooting scenario.

Upgrade Notes​

  • Upgrade the Longhorn CSI pack to 1.12.0 if your Edge clusters run Linux kernel 6.19 or later, including Hadron 7.1.3 and Ubuntu 7.0.0-generic. Because of a known issue on these kernels, shared Read-Write-Many (RWX) volumes served by Longhorn CSI versions 1.8.x through 1.11.2, or by the Piraeus RWX driver, can fail to open with a remote I/O error. For the affected combinations, symptoms, and a node-level workaround, refer to the shared volumes fail on kernel 7.x troubleshooting scenario.

Features​

  • Locally managed Edge clusters in an airgapped environment that use Palette eXtended Kubernetes Edge (PXK-E) or Canonical Kubernetes now support decoupled control plane and worker node upgrades. Enable the Skip worker node update (Optional) toggle on a worker pool in Local UI to hold that pool at its current Kubernetes version while the control plane advances, up to the Kubernetes N-3 minor version skew. This reduces how many times worker nodes repave when crossing several Kubernetes minor versions. Scale-up on a pool with the toggle enabled is rejected, and disabling the toggle repaves the pool to the control plane version. Refer to Decoupled Control Plane and Worker Node Upgrades for more information.
  • Airgap content bundle uploads to Edge hosts are now chunked, resumable, and parallel by default. The Palette CLI splits the bundle into chunks and transfers them over multiple connections, which shortens upload times for large bundles on links with a high bandwidth-delay product, and verifies each chunk with SHA-256. An interrupted upload resumes from the chunks the Edge host already holds instead of restarting. You can also stream a bundle straight from a signed object store URL with the new --src-url flag rather than staging a local copy. The earlier single-stream upload remains available with --legacy, and the CLI falls back to it automatically against Edge hosts that predate chunked upload support. Refer to Upload Content Bundle for more information.
  • Edge clusters in appliance mode can now use systemd extensions to deliver Kubernetes and Palette Agent binaries at runtime, instead of embedding those binaries in the provider image. This feature is a Technical Preview. This applies in both connected and airgapped environments. On operating systems running systemd version 255 or later, provider images built with CanvOS 4.10.x exclude the binaries by default, and Stylus (Palette Edge node agent) 4.10.x delivers them through systemd extensions. Set system.uri: NA in the BYOOS pack for standard upgrades. Unified Kernel Image (UKI) deployments do not support systemd extensions and continue to receive these binaries embedded in the provider image. Refer to Deliver Kubernetes and Agent Binaries via systemd Extensions for build and upgrade guidance.

Improvements​

  • The Custom UI appliance install wizard now renders a specific inline error at any field whose value is invalid. Sections that contain an invalid field display an "invalid" badge on the section title, and the step counter reads as "X of Y complete."
  • Edge workflows have been updated to Kairos v4.1.2 with kairos-init v0.16.x. Day-1 and Day-2 upgrades from earlier Kairos builds are supported.
  • The Palette TUI landing page now signposts initial user setup. When no login user exists on the Edge host, the landing page displays the yellow warning Setup required: press F2 to create login user for ssh and LocalUI, and the footer reads <F2> Create login instead of <F2> Customize. Both revert automatically once a login user is created. Previously, the landing page did not indicate that a user account was missing, so the F2 shortcut for creating the initial login was not discoverable. Refer to Initial Edge Host Configuration with Palette TUI for more information.

Bug Fixes​

  • Fixed an issue that prevented digest-pinned application images from being redirected to the local registry on airgapped Edge clusters. Pods that referenced an image by digest rather than by tag attempted to pull from the upstream registry and remained in ImagePullBackOff.
  • Fixed an issue that caused packs whose name contains a forward slash, such as Helm OCI packs sourced from a private registry, to fail to download on Edge hosts with failed to rename pack: no such file or directory errors.
  • Fixed an issue that caused NTP servers entered in the Palette TUI as a comma-separated list without a space, such as 10.10.180.0,10.10.180.1, to generate an invalid time synchronization configuration. Both comma-separated and comma-and-space-separated lists are now accepted.
  • Fixed an issue that prevented remote shell access to an Edge host from being automatically disabled after 24 hours of inactivity. If the service handling remote shell sessions restarted unexpectedly, the Edge host remained enabled for remote shell indefinitely, leaving the tunnel and its temporary user credentials active.
  • Fixed an issue that prevented the NTP servers configured on an Edge host through the Palette TUI from being visible in Local UI. The Edge host overview page now lists them in an NTP Servers field.
  • Fixed an issue that left a two-node Edge cluster without a Kubernetes API server after the loss of the leader node. The surviving node did not complete its promotion, so the cluster virtual IP served no traffic. Loss of the follower node was unaffected.
  • Fixed an issue that could place the Kubernetes datastore of a two-node Edge cluster on ephemeral rather than persistent storage. Data written since the last persistent snapshot was lost on reboot, and the node could not be promoted during failover.
  • Fixed an issue that prevented leader promotion from completing on a connected-mode two-node Edge cluster while Palette was unreachable, leaving the local control plane unavailable until connectivity was restored. Promotion now completes without Palette connectivity.

VerteX​

Features​

  • FIPS-compiled vCluster is now available, letting strict-FIPS tenants provision virtual clusters using FIPS 140-3 approved cryptography. The FIPS-compiled pack is automatically selected when deploying virtual clusters on VerteX.

  • Includes all Palette features, improvements, breaking changes, and deprecations in this release. Refer to the Palette section for more details.

Improvements​

  • Palette VerteX now invalidates all active JWTs for a session when a user logs out or changes their password. Previously, tokens remained valid after logout and could be reused.

Virtual Machine Orchestrator (VMO)​

VMO Pack​

Bug Fixes​
  • Fixed an issue that caused the CA certificate help text in the Certificate validation step of the Create Provider wizard to refer to the OpenShift API endpoint for every provider type, including VMware vSphere.
  • Fixed an issue that caused uploading a golden image to fail with a Certificate Trust Required prompt that contained no URL, leaving no way to accept the certificate and complete the upload. This occurred when the upload was proxied through Palette because the cluster had no direct upload URL available. The upload pages now display a notice before an upload begins, and a proxied upload that fails states that direct upload access has to be enabled on the VMO pack.
  • Fixed an issue that caused hotplug disk attachments to stall on Portworx-backed clusters. The DataVolume for the hotplug disk previously remained in ImportScheduled when other virtual machines in the same cluster used only container disks. Hotplug attachment now proceeds correctly in this scenario.
  • Clarified access control for the KubeVirt Configuration page. The page is available only to users who hold cluster-admin privileges and who authenticate through OIDC. Refer to KubeVirt Configuration and VMO Roles for more information.
  • Fixed an issue that prevented a namespace created through the console from appearing in the UI until the page was reloaded. New namespaces now appear immediately after they are created.

Automation​

info

The Palette CLI version corresponding to the 4.10.13 Palette release is 4.10.3. Refer to CLI Tools for the download URL and checksum.

Features​

Improvements​

  • Palette CLI is now available for ARM Linux (arm64) and macOS Apple Silicon (arm64).
  • The Palette CLI now confirms content bundle uploads immediately. Previously, after the upload progress bar reached 100%, the CLI could stay silent for several minutes while the Edge host unpacked the bundle. The CLI now reports upload completion as soon as the transfer finishes.
  • Added the is_synchronization attribute to the spectrocloud_registry_helm Terraform resource to enable synchronization controls and align functionality with OCI registries.
  • Added support for configuring registry synchronization settings and credential options in the Terraform provider.
  • Enhanced Terraform provider registry resource handling and configuration options.

Bug Fixes​

  • Fixed an issue in spectrocloud_registry_helm where missing TLS configuration options prevented TLS authentication from being sent, causing Helm registries to be stored with TLS disabled.

Docs and Education​

  • Documentation has been added explaining that Local UI, the Palette TUI, and the Palette API all change Edge host passwords as root, and that PAM exempts root from password quality checks unless the image sets the enforce_for_root option. Without it, the check still logs a BAD PASSWORD message but returns success, so a password that does not meet your policy is accepted. Refer to Change User Password for the ways a password can be changed, and Build Edge Artifacts for SUSE, Ubuntu, and RHEL examples that set the option at image build time.

Packs​

Pack NameLayerNon-FIPSFIPSNew Version
argo-cdaddon✅❌10.4.2
aws-efsaddon✅❌3.4.2
calico-network-policyaddon✅❌3.32.2
csi-aws-ebscsi✅✅1.64.0
csi-aws-efscsi✅❌3.4.2
csi-longhorncsi❌✅1.12.0
csi-longhorn-addonaddon❌✅1.12.0
headlampaddon✅❌0.45.0
istioaddon✅❌1.30.4
karpenteraddon✅✅1.14.1
nvidia-gpu-operator-aiaddon✅❌26.7.0
open-policy-agentaddon✅❌3.23.1
prometheus-operatoraddon✅❌88.6.2
spectro-rbacADDON✅❌1.0.2
tetragonaddon✅❌1.7.1
traefikaddon✅❌41.4.0

Pack Notes​

  • The tigera-operator 3.32.1 pack has been republished as tigera-operator-3.32.1-rev1 to add the calico/csi and calico/node-driver-registrar images to its image manifest. The Tigera Operator deploys these two images whenever kubeletVolumePluginPath is set to a value other than None, but because they were missing from the manifest they were never mirrored into dedicated or airgapped registries, leaving the csi-node-driver DaemonSet in ImagePullBackOff. Calico networking was unaffected, because the Container Storage Interface (CSI) driver is a separate optional component.
  • Fixed an issue where the Spectro RBAC add-on pack preset included live example namespaces that were automatically deployed when no namespaces were specified in the profile layer.

Deprecations and Removals​

  • The Ingress Nginx pack is disabled in Palette 4.10.13. You can no longer create new cluster profiles that include the pack. Existing profiles that already contain it continue to launch new clusters, and running workloads are unaffected. The pack was deprecated in Palette 4.8.0, following the Kubernetes project's retirement of ingress-nginx. The pack remains disabled for approximately three months before it transitions to the Deleted state, after which existing profiles can no longer launch new clusters.