Version 1.70

Important

  • Prometheus has been replaced with Deckhouse Prom++. If you want to keep using Prometheus, disable the prompp module manually before upgrading DKP by running the command d8 platform module disable prompp.

  • Support for Kubernetes 1.33 has been added, while support for Kubernetes 1.28 has been discontinued. In future DKP releases, support for Kubernetes 1.29 will be removed. The default Kubernetes version (used when the kubernetesVersion parameter is set to Automatic) has been changed to 1.31.

  • Upgrading the cluster to Kubernetes 1.31 requires a sequential update of all nodes, with each node drained. You can control how node updates requiring workload disruptions are applied using the disruptions parameter section.

  • The built-in snapshot-controller and static-routing-manager modules will now be replaced with their external counterparts of the same name, sourced via ModuleSource deckhouse.

  • The new version of Cilium requires nodes to run Linux kernel version 5.8 or newer. If any node in the cluster has a kernel older than 5.8, the Deckhouse Kubernetes Platform upgrade will be blocked. Cilium Pods will be restarted.

  • All DKP components will be restarted during the update.

Major changes

  • You can now enforce two-factor authentication for static users. This is configured via the staticUsers2FA parameter section of the user-authn module.

  • When enabling a module (with d8 platform module enable) or editing a ModuleConfig resource, a warning is now displayed if multiple module sources are found. In such a case, explicitly specify the module source using the source parameter in the module’s configuration.

  • Improved error handling for module configuration. Module-related errors no longer block DKP operations. Instead, they are now displayed in the status fields of Module and ModuleRelease objects.

  • Improved virtualization support:
    • Added a provider for integration with Deckhouse Virtualization Platform (DVP), enabling deployment of DKP clusters on top of DVP.
    • Added support for nested virtualization on nodes in the cni-cilium module.
  • The node-manager module now includes several enhancements for improved node reliability and manageability:
    • You can now prevent a node from restarting if it still hosts critical Pods (labeled with pod.deckhouse.io/inhibit-node-shutdown). This can be necessary for workloads with stateful components, such as long-running data migrations.
    • Introduced API version v1alpha2 for the SSHCredential resource, where the sudoPasswordEncoded parameter allows specifying the sudo password in Base64 format.
    • The capiEmergencyBrake parameter allows you to disable Cluster API (CAPI) in emergency scenarios, preventing potentially destructive changes. Its behavior is similar to the existing mcmEmergencyBrake setting.

  • Added a pre-installation check to verify connectivity to the DKP container image registry.

  • Improved the log file rotation mechanism when using short-term log storage (via the loki module). Added the LokiInsufficientDiskForRetention alert to warn about insufficient disk space for log retention.

  • The documentation now includes a reference for the Deckhouse CLI (d8 utility) commands and parameters.

  • When using CEF encoding for collecting logs from Apache Kafka or socket sources, you can now configure auxiliary CEF fields such as Device Product, Device Vendor, and Device ID.

  • The passwordHash field in the NodeUser resource is no longer required. This allows you to create users without passwords — for example, in clusters that use external authentication systems (such as PAM or LDAP).

  • Added support for CRI Containerd v2 with CgroupsV2. The new version introduces a different configuration format and includes a mechanism to migrate between Containerd v1 and v2. You can change the CRI type used on nodes via the cri.type parameter and configure it using cri.containerdV2.

Security

  • Container image signature verification is now available in DKP SE+. This feature is now supported in DKP SE+ and EE.

  • The log-shipper, deckhouse-controller, and Istio (version 1.21) modules have been migrated to distroless builds. This improves security and ensures a more transparent and controlled build process.

  • New audit rules have been added to track interactions with containerd. The following are now monitored: access to the /run/containerd/containerd.sock socket, modifications to the /etc/containerd and /var/lib/containerd directories and the /opt/deckhouse/bin/containerd file.

  • Known vulnerabilities have been fixed in the following modules: loki, extended-monitoring, operator-prometheus, prometheus, prometheus-metrics-adapter, user-authn, and cloud-provider-zvirt.

Network

  • Added support for Istio version 1.25.2, which uses the Sail operator instead of the deprecated Istio Operator. Also added support for Kiali version 2.7, without Ambient Mesh support. Istio version 1.19 is now considered deprecated.

  • Added support for encrypting traffic between nodes and Pods using the WireGuard protocol (via the encryption.mode parameter).

  • Fixed the logic for determining service readiness in the ServiceWithHealthcheck resource. Previously, Pods without an IP address (for example, in Pending state) could be mistakenly included in the load balancing list.

  • Added support for the least-conn load balancing algorithm. This algorithm directs traffic to the service backend with the fewest active connections, improving performance for connection-heavy applications (such as WebSocket services). To use this algorithm, enable the extraLoadBalancerAlgorithmsEnabled parameter in the cni-cilium module settings and use the cilium.io/bpf-lb-algorithm annotation on the service and set it to a supported value: random, maglev, or least-conn.

  • Fixed an issue in Cilium 1.17 cilium-operator where IP addresses were not reused after a CiliumEndpoint was deleted. The issue was caused by improper cleanup of priority filters, which could lead to IP pool exhaustion in large clusters.

  • Refined the list of ports used for networking:

    • Added and updated:
      • 4287/UDP: WireGuard port used for CNI Cilium traffic encryption.
      • Instead of 4298/UDP and 4299/UDP, the range 4295–4299/UDP is now used for VXLAN traffic encapsulation between Pods.
    • Removed:
      • 49152, 49153/TCP: Previously used for live migration of virtual machines (in the virtualization module). Migration now occurs over the Pod network.

Component version updates

The following DKP components have been updated:

  • cilium: 1.17.4
  • golang.org/x/net: v0.40.0
  • etcd: v3.6.1
  • terraform-provider-azure: 3.117.1
  • Deckhouse CLI: 0.13.2
  • Falco: 0.41.1
  • falco-ctl: 0.11.2
  • gcpaudit: v0.6.0
  • Grafana: 10.4.19
  • Vertical pod autoscaler: 1.4.1
  • dhctl-kube-client: v1.3.1
  • cloud-provider-dynamix dynamix-common: v0.5.0
  • cloud-provider-dynamix capd-controller-manager: v0.5.0
  • cloud-provider-dynamix cloud-controller-manager: v0.4.0
  • cloud-provider-dynamix cloud-data-discoverer: v0.6.0
  • cloud-provider-huaweicloud huaweicloud-common: v0.5.0
  • cloud-provider-huaweicloud caphc-controller-manager: v0.3.0
  • cloud-provider-huaweicloud cloud-data-discoverer: v0.5.0
  • registry-packages-containerdv2: 2.1.3
  • registry-packages-containerdv2-runc: 1.3.0
  • cilium: 1.17.4
  • cilium envoy-bazel: 6.5.0
  • cilium cni-plugins: 1.7.1
  • cilium protoc: 30.2
  • cilium grpc-go: 1.5.1
  • cilium protobuf-go: 1.36.6
  • cilium protoc-gen-go-json: 1.5.0
  • cilium gops: 0.3.27
  • cilium llvm: 18.1.8
  • cilium llvm-build-cache: llvmorg-18.1.8-alt-p11-gcc11-v2-180225
  • User-authn basic-auth-proxy go: 1.23.0
  • Prometheus alerts-reciever go: 1.23.0
  • Prometheus memcached_exporter: 0.15.3
  • Prometheus mimir: 2.14.3
  • Prometheus promxy: 0.0.93
  • Extended-monitoring k8s-image-availability-exporter: 0.13.0
  • Extended-monitoring x509-certificate-exporter: 3.19.1
  • Cilium-hubble hubble-ui: 0.13.2
  • Cilium-hubble hubble-ui-frontend-assets: 0.13.2

Version 1.70

Important

  • The ceph-csi module has been removed. Use the csi-ceph module instead. Deckhouse will not be updated as long as ceph-csi is enabled in the cluster. For csi-ceph migration instructions, refer to the module documentation.

  • Version 1.12 of the NGINX Ingress Controller has been added. The default controller version has been changed to 1.10. All Ingress controllers that do not have an explicitly specified version (via the controllerVersion parameter in the IngressNginxController resource or the defaultControllerVersion parameter in the ingress-nginx module) will be restarted.

  • The falco_events metric (from the runtime-audit-engine module) has been removed. The falco_events metric was considered deprecated since DKP 1.68. Use the falcosecurity_falcosidekick_falco_events_total metric instead. Dashboards and alerts based on the falco_events metric may stop working.

  • All DKP components will be restarted during the update.

Major changes

  • In the Auto update mode, patch version updates (for example, from v1.70.1 to v1.70.2) are now applied taking into account the update windows, if they are set. Previously, in this update mode, only minor version updates (for example, from v1.69.x to v1.70.x) were applied with consideration to update windows, while patch version updates were applied as they appeared on a release channel.
  • A node can now be rebooted if the corresponding Node object has the update.node.deckhouse.io/reboot annotation set.
  • When cleaning up a static node, any local users created by Deckhouse Kubernetes Platform are now also removed.
  • Added synchronization monitoring for Istio in multi-cluster configurations. A new alert D8IstioRemoteClusterNotSynced has been introduced and triggers in the following cases:
    • The remote cluster is offline.
    • The remote API endpoint is not reachable.
    • The remote ServiceAccount token is invalid or expired.
    • There is a TLS or certificate issue between the clusters.
  • The deckhouse-controller collect-debug-info command now also collects debug information for Istio, including:
    • Resources in the d8-istio namespace.
    • CRDs from the istio.io and gateway.networking.k8s.io groups.
    • Istio logs.
    • Sidecar logs of a single randomly selected user application.
  • A new monitoring dashboard has been added to display OpenVPN certificate status. Upon expiration, server certificates will now be reissued, and client certificates will be removed. The following alerts have been added:
  • Monitoring dashboards have been renamed and updated:
    • “L2LoadBalancer” renamed to “MetalLB L2”; pool and column filtering added.
    • “Metallb” renamed to “MetalLB BGP”; pool and column filtering added. The ARP request panel has been removed.
    • “L2LoadBalancer / Pools” renamed to “MetalLB / Pools”.

  • The upmeter module’s PVC size has been increased to accommodate data retention for 13 months. In some cases, the previous PVC size was insufficient.

  • The ModuleSource resource status now includes information about module versions in the source.

  • The Module resource status now includes information about the module’s lifecycle stage. A module can move through the following stages in its lifecycle: Experimental, Preview, General Availability, and Deprecated. For details on module lifecycle stages and how to evaluate its stability, refer to the corresponding section in the documentation.

  • It is now possible to use stronger or more modern encryption algorithms (such as RSA-3072, RSA-4096, or ECDSA-P256) for control plane cluster certificates instead of the default RSA-2048. You can use the encryptionAlgorithm parameter in the ClusterConfiguration resource to configure this.

  • The descheduler module can now be configured to evict pods that are using local storage. Use the evictLocalStoragePods parameter in the module configuration to adjust this.

  • You can now adjust the logging level of the Ingress controller using the controllerLogLevel parameter in the IngressNginxController resource. The default log level is Info. Controlling the logging level can help prevent log collector overload during Ingress controller restarts.

Security

  • The severity level of alerts indicating security policy violations has been raised from 7 to 3.

  • The configuration for Yandex Cloud, Zvirt, and Dynamix providers now uses OpenTofu instead of Terraform. This enables easier provider updates, such as applying fixes for known vulnerabilities (CVEs).

  • CVE vulnerabilities have been fixed in the following modules: chrony, descheduler, dhctl, node-manager, registry-packages-proxy, falco, cni-cilium, and vertical-pod-autoscaler.

Component version updates

The following DKP components have been updated:

  • containerd: 1.7.27
  • runc: 1.2.5
  • go: 1.24.2, 1.23.8
  • golang.org/x/net: v0.38.0
  • mcm: v0.36.0-flant.23
  • ingress-nginx: 1.12.1
  • terraform-provider-aws: 5.83.1
  • Deckhouse CLI: 0.12.1
  • etcd: v3.5.21

Version 1.69

Important

  • Support for Kubernetes 1.32 has been added, while support for Kubernetes 1.27 has been discontinued. The default Kubernetes version has been changed to 1.30. In future DKP releases, support for Kubernetes 1.28 will be removed.

  • All DKP components will be restarted during the update.

Major changes

  • The ceph-csi module is now deprecated. Plan to migrate to the csi-ceph module instead. For details, refer to the Ceph documentation.

  • You can now grant access to Deckhouse web interfaces using user names via the auth.allowedUserEmails field. Access restriction is configured together with the auth.allowedUserGroups parameter in configuration of the following modules with web interfaces: cilium-hubble, dashboard, deckhouse-tools, documentation, istio, openvpn, prometheus, and upmeter (example for prometheus).

  • A new dashboard Cilium Nodes Connectivity Status & Latency has been added to Grafana in the cni-cilium module. It helps monitor node network connectivity issues. The dashboard displays a connectivity matrix similar to the cilium-health status command, using metrics that are already available in Prometheus.

  • A new D8KubernetesStaleTokensDetected alert has been added in the control-plane-manager module that is triggered when stale service account tokens are detected in the cluster.

  • You can now create a Project from an existing namespace and adopt existing objects into it. To do this, annotate the namespace and its resources with projects.deckhouse.io/adopt. This lets you switch to using Projects without recreating cluster resources.

  • A Terminating status has been added to ModuleSource and ModuleRelease resources. The new status will be displayed when an attempt to delete one of them fails.

  • The installer container now automatically configures cluster access after a successful bootstrap. A kubeconfig file is generated in ~/.kube/config, and a local TCP proxy is set up through an SSH tunnel. This allows you to use kubectl locally right away without manually connecting to the control-plane node via SSH.

  • Changes to Kubernetes resources in multi-cluster and federation setups are now tracked directly via Kubernetes API. This enables faster synchronization between clusters and eliminates the use of outdated certificates. In addition, mounting of ConfigMap and Secret resources into Pods has been removed to eliminate family system compromise risks.

  • A new dynamicforward plugin has been added to CoreDNS, improving DNS query processing in the cluster. It integrates with node-local-dns, continuously monitors kube-dns endpoints, and automatically updates the list of DNS forwarders. If the control-plane node is unavailable, DNS queries are still forwarded to available endpoints, improving cluster stability.

  • A new log rotation approach has been introduced in the loki module. Now, old logs are automatically removed when disk usage exceeds a threshold: either 95% of PVC size or PVC size minus the size required to store two minutes of log data at the configured ingestion rate (ingestionRateMB). The retentionPeriodHours parameter no longer controls the data retention and is used for monitoring alerts only. If loki begins removing old logs before the set period is reached, a LokiRetentionPerionViolation alert will be triggered, informing the user that they must reduce the value of retentionPeriodHours or increase the PVC size.

  • A new nodeDrainTimeoutSecond parameter lets you set the maximum timeout when attempting to drain a node (in seconds) for each NodeGroup resource. Previously, you could only use the default value (10 minutes) or reduce it to 5 minutes using the quickShutdown parameter, which is now deprecated.

  • The openvpn module now includes a defaultClientCertExpirationDays parameter, allowing you to define the lifetime of client certificates.

Security

Known vulnerabilities have been addressed in the following modules: ingress-nginx, istio, prometheus, and local-path-provisioner.

Component version updates

The following DKP components have been updated:

  • cert-manager: 1.17.1
  • dashboard: 1.6.1
  • dex: 2.42.0
  • go-vcloud-director: 2.26.1
  • Grafana: 10.4.15
  • Kubernetes control plane: 1.29.14, 1.30.1, 1.31.6, 1.32.2
  • kube-state-metrics (monitoring-kubernetes): 2.15.0
  • local-path-provisioner: 0.0.31
  • machine-controller-manager: v0.36.0-flant.19
  • pod-reloader: 1.2.1
  • prometheus: 2.55.1
  • Terraform providers:
    • OpenStack: 1.54.1
    • vCD: 3.14.1

Version 1.68

Important

  • After the update, the UID will change for all Grafana data sources created using the GrafanaAdditionalDatasource resource. If a data source was referenced by UID, that reference will no longer be valid.

Major changes

  • A new parameter, iamNodeRole, has been introduced for the AWS provider. It lets you specify the name of the IAM role to bind to all AWS instances of cluster nodes. This can come in handy if you need to grant additional permissions (for example, access to ECR, etc.).

  • Creating nodes of the CloudPermanent type now takes less time. Now, CloudPermanent nodes are created in parallel. Previously, they were created in parallel only within a single group.

  • Monitoring changes:
    • Support for monitoring certificates in secrets of the Opaque type has been added.
    • Support for monitoring images in Amazon ECR has been added.
    • A bug that could cause partial loss of metrics when Prometheus instances were restarted has been fixed.

  • When using a multi-cluster Istio configuration or federation, you can now explicitly specify the list of addresses used for inter-cluster requests. Previously, these addresses were determined automatically; however, in some configurations, they could not be resolved.

  • The DexAuthenticator resource now has a highAvailability parameter that controls high availability mode. In high availability mode, multiple replicas of the authenticator are launched. Previously, high availability mode of all authenticators was determined by a global parameter or by the user-authn module. All authenticators deployed by DKP now inherit the high availability mode of the corresponding module.

  • Node labels can now be added, removed, or modified using files stored on the node in the /var/lib/node_labels directory and its subdirectories. The full set of applied labels is stored in the node.deckhouse.io/last-applied-local-labels annotation.

  • Support for the Huawei Cloud provider has been added.

  • The new keepDeletedFilesOpenedFor parameter in the log-shipper module allows you to configure the period to keep the deleted log files open. This way, you can continue reading logs from deleted pods for some time if log storage is temporarily unavailable.

  • TLS encryption for log collectors (Elasticsearch, Vector, Loki, Splunk, Logstash, Socket, Kafka) can now be configured using secrets, rather than by storing certificates in the ClusterLogDestination resources. The secret must reside in the d8-log-shipper namespace and have the log-shipper.deckhouse.io/watch-secret: true label.

  • In the project status under the resources section, you can now see which project resources have been installed. Those resources are marked with installed: true.

  • A new parameter, --tf-resource-management-timeout, has been added to the installer. It controls the resource creation timeout in cloud environments. By default, the timeout is set to 10 minutes. This parameter applies only to the following clouds: AWS, Azure, GCP, OpenStack.

Security

Known vulnerabilities have been addressed in the following modules:

  • admission-policy-engine
  • chrony
  • cloud-provider-azure
  • cloud-provider-gcp
  • cloud-provider-openstack
  • cloud-provider-yandex
  • cloud-provider-zvirt
  • cni-cilium
  • control-plane-manager
  • extended-monitoring
  • descheduler
  • documentation
  • ingress-nginx
  • istio
  • loki
  • metallb
  • monitoring-kubernetes
  • monitoring-ping
  • node-manager
  • operator-trivy
  • pod-reloader
  • prometheus
  • prometheus-metrics-adapter
  • registrypackages
  • runtime-audit-engine
  • terraform-manager
  • user-authn
  • vertical-pod-autoscaler
  • static-routing-manager

Component version updates

The following DKP components have been updated:

  • Kubernetes Control Plane: 1.29.14, 1.30.10, 1.31.6
  • aws-node-termination-handler: 1.22.1
  • capcd-controller-manager: 1.3.2
  • cert-manager: 1.16.2
  • chrony: 4.6.1
  • cni-flannel: 0.26.2
  • docker_auth: 1.13.0
  • flannel-cni: 1.6.0-flannel1
  • gatekeeper: 3.18.1
  • jq: 1.7.1
  • kubernetes-cni: 1.6.2
  • kube-state-metrics: 2.14.0
  • vector (log-shipper): 0.44.0
  • prometheus: 2.55.1
  • snapshot-controller: 8.2.0
  • yq4: 3.45.1

Mandatory component restart

The following components will be restarted after updating DKP to 1.68:

  • Kubernetes Control Plane
  • Ingress controller
  • Prometheus, Grafana
  • admission-policy-engine
  • chrony
  • cloud-provider-azure
  • cloud-provider-gcp
  • cloud-provider-openstack
  • cloud-provider-yandex
  • cloud-provider-zvirt
  • cni-cilium
  • control-plane-manager
  • descheduler
  • documentation
  • extended-monitoring
  • ingress-nginx
  • istio
  • kube-state-metrics
  • log-shipper
  • loki
  • metallb
  • monitoring-kubernetes
  • monitoring-ping
  • node-manager
  • openvpn
  • operator-trivy
  • prometheus
  • prometheus-metrics-adapter
  • pod-reloader
  • registrypackages
  • runtime-audit-engine
  • service-with-healthchecks
  • static-routing-manager
  • terraform-manager
  • user-authn
  • vertical-pod-autoscaler