UCP 3.0 release notesEstimated reading time: 6 minutes
Here you can learn about new features, bug fixes, breaking changes, and known issues for the latest UCP version. You can then use the upgrade instructions, to upgrade your installation to the latest release.
Version 3.0.0 (2018-04-17)
The UCP system requirements were updated with 3.0.0. Make sure to check the system requirements before upgrading.
- UCP now supports Kubernetes as an orchestrator, in addition to the existing Swarmkit and “classic” Swarm orchestrators. Kubernetes system components are automatically deployed on all manager and Linux worker nodes managed by UCP. Learn more about Kubernetes support.
- Worker nodes running Linux on amd64 architectures can be configured to run only Swarm workloads, only Kubernetes workloads, or mixed workloads. Manager nodes are by default Mixed in order to support Swarm and Kubernetes system components. However, it is not recommended to run Worker nodes as Mixed due to potential resource contention issues.
- Users can deploy Kubernetes workloads through the web UI, and the CLI using
a UCP client bundle and
kubectl. Learn more.
- Users can now use Compose to deploy Kubernetes workloads from the web UI. Lean more.
- UCP includes Calico as the default CNI plugin for networking of Kubernetes
applications. Learn more.
The following Calico features are supported:
- L3 IP-IP Overlay Data Path.
- BGP control plane.
- Calico IPAM.
- Management of Calico CNI plugin lifecycle.
- Kubernetes Network Policy. This is experimental in 3.0.0.
- You can now use layer 7 routing in your Kubernetes workloads by using an NGINX-based ingress controller. Learn more.
- Layer 7 routing for Swarmkit applications has been upgraded to use Interlock backend. This adds increased performance, stability, and new features including SSL Termination, Contextual Path-based Routing, Websocket Support, and Canary Application Instance deployments. Existing Hostname Routing Mesh (HRM) labels (and newly added labels with the old format) will automatically migrate to the new format. It is strongly recommended to use the new format for new applications in order to take advantage of the new features. Learn more.
- Support for NFS-based Kubernetes persistent volumes. Additional volume plugins will be available in future releases.
- Role-based access control now supports Kubernetes resources.
- In addition to users, teams, organizations, and grants you can now use Kubernetes Service Accounts as a subject type. Learn more.
- You can now create custom roles with Kubernetes API permissions. Default roles include Kubernetes API permissions based on their access type. As an example, View-Only contains Swarm and Kubernetes read-only API permissions.
- In addition to collections, grants can now use Kubernetes Namespaces as a resource set type.
- Admins can now link a Kubernetes namespace to a collection of nodes in order to isolate users and workloads between different nodes.
- Administrators can now enforce only running trusted images for both swarm and Kubernetes applications. Learn more
- API support for registering multiple UCP clusters to a single DTR for the purposes of signed image enforcement. Learn more.
Restricted Controlrole includes the
User ImpersonationKubernetes action, which can allow a user to escalate to admin privileges if the role is granted against
All Kubernetes Namespaces. For this version, we recommend that administrators do not grant the
Restricted Controlrole against Kubernetes namespaces, and use custom roles instead. This issue does not affect any other roles in the system, or any of the grants using
Restricted Controlagainst collections.
- For increased security UCP now requires clients to use TLS version 1.2.
- Platform support
- Kubernetes is not yet supported for Windows based workloads. Use Swarmkit for Windows based workloads instead.
- EE 2.0 is not yet supported in IBM Z platforms.
- Both Docker and
kubectlCLIs report that UCP is running Kubernetes 1.8.2, when in fact it is running 1.8.9.
- Both Docker and
- Swarm encrypted overlay networks might not work as expected because default Kubernetes firewall rules are interfering with them. Learn more.
- Calico networking for Kubernetes is not supported for Microsoft Azure. UCP leverages Azure networking and IPAM for control-plane and connectivity. Learn how to deploy EE 2.0 on Azure.
- Azure IPAM will fail if nodes in the cluster are connected to different subnets. As a workaround ensure network setup avoids multiple subnets. This will be rectified in an upcoming patch release (#12894).
- UCP Calico control-plane supports full-mesh BGP peering only at release-time. Calico control-plane may cause high CPU on nodes in clusters above 100 nodes. A route reflector based partial-mesh BGP control-plane will reduce CPU consumption when scaling past 100 nodes. Route-reflector configurations will be included in a future release.
- In some deployments the
kube-dnscomponent won’t be able to resolve external domain names. Deploy a ConfigMap to work around this.
- If you upgrade from UCP 2.x to UCP 3.x on Azure, Kubernetes networking doesn’t work. The cluster upgrade completes, and Swarm workloads work, but Kubernetes networking will be down.
- If upgrading UCP through the UI, UCP will not check to ensure the manager node has the minimum memory required of 4 GB. Upgrading through the CLI does check for this requirement.
- Putting a node in
drainmode currently removes only Swarm workloads, and not Kubernetes workloads. This will be fixed in a future release.
- Kubernetes base image layer uses Ubuntu 16.04 which contains some known CVE vulnerabilities. These will be removed when the base image layer is updated.
docker system prune -adirectly on individual worker nodes in the cluster will potentially delete UCP system images. This behavior will not occur if the prune command is run using a UCP client bundle.
- Compose for Kubernetes only supports v3 or higher YAML files. Any older version YAML files will silently fail without errors.
- Linking Kubernetes namespace to a collection of nodes in order to isolate Kubernetes workloads between different nodes is not working as expected. You can use this workaround.
kubectl get csmight show some internal UCP components as unhealthy when that’s not the case.
- UCP does not yet support dynamic volume provisioning (NFS volumes do not support this). This will change in future releases when more volume types are available.
The following functionality has been deprecated with UCP 3.0.0 and will be unavailable in the next UCP feature release.
- The web UI is going to stop supporting users to deploy stacks with basic containers. You should update your Compose files to version 3, and deploy your stack as a Swarm service or Kubernetes workload.
- The option to integrate with a remote Syslog system is going to be removed from the UCP web UI. You can configure Docker Engine for this.
- The option to configure a rescheduling policy for basic containers is deprecated. Deploy your applications as Swarm services or Kubernetes workloads.