(Release Date: 2022-03-23)
Main goals for Release 2 (R2) were massive improvements in bare metal deployment and our cluster management layer gaining the ability to handle many clusters independently with a number of optional services.
-
We support the latest Kubernetes 1.22 and 1.23 releases.
-
The Kubernetes Cluster API is now available in a stable v1beta1 release 1.0.x with the corresponding cluster-api-provider-openstack 0.5.x.
-
There are a number of new standard services available for the k8s capi managed clusters, amongst which cert-manager and flux. The clusters have better default settings for the nginx-ingress, anti-affinity for the nodes and the ability to chose cilium over calico and to have stable multi-controller node setups on clouds without low-latency local storage. Please consult the k8s capi provider release notes for more details.
-
- We have also enabled SPICE support in addition to noVNC to access the graphical console of VMs.
-
The base infrastructure is provided by OSISM 3.0.0 which in turn build on top of kolla and kolla-ansible.
-
The Cluster Management Node is now well prepared to manage numerous clusters with independent settings and different feature sets by creating default settings and then keeping track of various workload clusters in own directories. Documentation has been vastly improved.
-
The Cluster Management node now gets its artifacts directly from git, making incremental updates to it a lot easier, thus also avoiding to disrupt workload clusters through redeployed management nodes to roll out updates.
-
Added dashboards for the operators:
- Homer
- Flower
- Grafana dashboards
-
Work is underway to supersede openstack-health-monitor with a solution that is using tempest and rally. The health-monitor has received improvements though and is at this point still fully supported and recommended -- it has surfaced a number of issues with test clouds, especially failed metadata services.
-
User federation has been prepared to be ready for Gaia-X federation integration
- Keystone can consume users from Keycloak via OpenID-Connect
- Keycloak uses the highly-available Galera database cluster now
- mod_oauth2 support for Keystone
-
Vast improvements in the SCS Deployment automation
- Full automation of bare metal deployment with Bifrost and Ironic
- Using NetBox as central source of truth for the complete setup
-
New services available (opt-in)
- ClamAV, dnsdist, cgit, FRRouting, Nexus, Tang
-
Traefik centrally routes the connections to Nexus, NetBox, phpMyAdmin, Homer, Flower, ARA, cgit
We now have scripts that allow us to connect to the workload cluster node network for debugging purposes.
The configuration of the testbed was minimized and the deployment was made more production-oriented.
Further noteworthy improvements to testbed:
- TLS is implemented throughout the services also in testbed
- Virtual BMC in testbed
- Public DNS for testbed (
testbed.osism.xyz)
We have a Zuul CI framework running and started migrating CI testing from github actions to using our Zuul infrastructure.
Renovate is being used to keep the pinned versions up-to-date and consistent across the many repositories.
An overview over the used software versions is available from the OSISM release repository as input for a complete SBOM. This allows to e.g. investigate the contents of the used (v3.0.0) images.
The names of a few settings have changed since R1 -- if you have diverged from the defaults,
this may require adjusting the environment.tfvars or the clusterctl.yaml files.
See k8s-cluster-api-provider Release Notes
for more details.
The updating approach has fundamentally changed: If you were used to deploy fresh management nodes regularly to benefit from the upstream improvements, this need has been vastly reduced now, allowing for long-living management nodes and workload clusters managed by them.
-
Playbook generic-configuration.yml was deprecated. From now on, please use the playbook of the same name in the manager environment (manager-configuration.yml). All configuration parameters from environments/configuration.yml should be moved to environments/manager/configuration.yml.
-
In kolla-ansible the haproxy role was renamed to loadbalancer. Accordingly, loadbalancer must now be used for the deployment of HAProxy.
- OpenStack Victoria images are no longer built and thus no longer kept updated
- Support for Zabbix has been removed, Prometheus will be used as the only monitoring stack in the future
- Heimdall as a service was removed, as an alternative Homer is now available
Deprecations happen according to our deprecation policy.
- Cockpit is deprecated in favor of Boundary by HashiCorp or Teleport
- ceph-ansible is deprecated in preparation for cephadm
- All osism- scripts on the manager are deprecated and will be replaced by the new OSISM CLI. The scripts will be removed in the next release
- The following services are currently not used and are deprecated and scheduled for removal as of now: Falco, Jenkins, Rundeck, Lynis, Trivy
- Heat will no longer be offered by default in the testbed in the future
- The docker-compose CLI will be removed and replaced by the new compose plugin for Docker. docker-compose is then no longer available and docker compose must be used instead
- The Elasticsearch container included in OSISM testbed was exposed to the log4j issue -- new images were provided for addressing this. See the security advisory
-
The nginx-ingress loadbalancer could run into name conflicts before. The loadbalancer now uses a health monitor to avoid routing to the wrong nodes, which typically resulted in 10s delays when connecting to the service behind the ingress controller.
-
cAdvisor has now reduced number of Prometheus metrics and labels exported by default - this will ease the load on the system. This implies that corresponding time series data will no longer be created.
The clusters created with our cluster-API cluster management solution pass the CNCF conformance tests as reported by sonobuoy.
The OpenStack layer passes the OIF trademark tests, so cloud providers leveraging the stack should easily be able to achieve the "OpenStack powered compute" trademark certification.
Our partner PlusServer has achieved a BSI C5 security certification for their SCS implementation pluscloud open.
We are working within Gaia-X to further the power of Gaia-X self-descriptions and are closely working with the GXFS project to jointly deliver a standard toolbox for Gaia-X compliant infrastructure and service offerings.
The SCS standards for flavor naming and image metadata are largely unchanged since R1. We have however made progress in our reference implementation fully implementing them without any further tweaks.
See Release Numbering scheme -- unchanged from R0.
We have added the tag v3.0.0 to the relevant repositories to designate the SCS_RELEASE_R2.
Note that we will release R3 (v4.0.0) in September 2022 and stop providing maintenance updates for R2 at the end of October.
Alongside with R2 we published a blog post on some first thoughts on future directions towards R3.
We appreciate contribution to strategy and implementation, please join our community -- or just leave input on the github issues and PRs. Have a look at our contributor guide. We also have worked on a Code of Conduct to document the expected behavior of contributors and how we deal with cases where individuals fail to meet the expectation.