This section describes the specific actions you as a cloud operator need to complete before or after your Container Cloud cluster update to the Cluster releases 17.0.0, 16.0.0, or 14.1.0.
Upgrade to Ubuntu 20.04 on baremetal-based clusters¶
The Cluster release series 14.x and 15.x are the last ones where Ubuntu 18.04 is supported on existing clusters. A Cluster release update to 17.0.0 or 16.0.0 is impossible for a cluster running on Ubuntu 18.04.
Therefore, if your cluster update is blocked, make sure that the operating system on all cluster nodes is upgraded to Ubuntu 20.04 as described in Operations Guide: Upgrade an operating system distribution.
Configure managed clusters with the etcd storage quota set¶
If your cluster has custom etcd storage quota set as described in
Increase storage quota for etcd, before the management cluster upgrade to 2.25.0,
Manually set the
LCMMachineresources of the cluster controller nodes:
spec: stateItemsOverwrites: deploy: ucp_etcd_storage_quota: "<custom_etcd_storage_quota_value>"
stateItemsOverwrites.deploysection is already set, append
ucp_etcd_storage_quotato the existing parameters.
To obtain the list of the cluster
kubectl -n <cluster_namespace> get lcmmachine
To patch the cluster
LCMMachineresources of the
kubectl -n <cluster_namespace> edit lcmmachine <control_lcmmachine_name>
After the management cluster is upgraded to 2.25.0, update your managed cluster to the Cluster release 17.0.0 or 16.0.0.
Manually remove the
ucp_etcd_storage_quotaparameter from the
Migrate Ceph cluster to address storage devices using by-id¶
Container Cloud uses the device
by-id identifier as the default method
of addressing the underlying devices of Ceph OSDs. This is the only persistent
device identifier for a Ceph cluster that remains stable after cluster
upgrade or any other cluster maintenance.
Therefore, if your existing Ceph clusters are still utilizing the device names
by-path symlinks, migrate them to the
by-id format as
described in Migrate Ceph cluster to address storage devices using by-id.
Point DHCP relays on routers to the new dhcp-lb IP address¶
If your managed cluster has multiple L2 segments using DHCP relays on the
border switches, after the related management cluster automatically upgrades
to Container Cloud 2.25.0, manually point the DHCP relays on your network
infrastructure to the new IP address of the
dhcp-lb service of the
Container Cloud managed cluster in order to successfully provision new nodes
or reprovision existing ones.
To obtain the new IP address:
kubectl -n kaas get service dhcp-lb
This change is required after the product has included the resolution of
the issue related to the incorrect handling of DHCP egress traffic. The fix
involves reconfiguring the external traffic policy for the
Kubernetes Service. For details about the issue, refer to the
Kubernetes upstream bug.