Kubermatic Kubernetes Platform (KKP) 2.11 adds support to expose user clusters by creating one service of type
LoadBalancer per user
cluster. Check out the Expose Strategy documentation for more details.
A new flag
service-account-signing-key was added to the KKP API. It is used to sign service account tokens via
HMAC. It should be unique per KKP installation and can be generated with the command:
base64 -w0 /dev/urandom | head -c 100
The value for this flag must be stored in
auth section for
kubermatic: auth: serviceAccountKey: "xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"
KKP 2.11 ships with Elasticsearch 7.0. Version 7 can do rolling upgrades only if the Elasticsearch cluster is already running on version 6.7 (the version shipped in KKP 2.10).
If you are already running KKP 2.10, you can simply deploy the updated Helm chart and the nodes will update one by one to Elasticsearch 7.
If you are running an older KKP release, you must either upgrade to KKP 2.10 first or shutdown the entire Elasticsearch cluster before doing the upgrade (e.g. by purging the chart before installing the new chart).
Note that because Elasticsearch 7.0 changed its database fields, you cannot downgrade to any previous release after the upgrade.
The backwards compatibility for the renamed configuration keys (
replicas etc.) introduced in KKP
2.10 has been removed, so make sure to update your
values.yaml to use the new config keys as documented in the 2.10
Both charts have had their naming scheme refactored in order to seamlessly install multiple copies along each other. This lead to the hostnames and PersistentVolumeClaim names changing.
alertmanager, but a service with the old name is installed to give user clusters time to reconcile. The fallback service will be removed in a later bugfix release.
In order to migrate existing data you must deploy both charts with the
migration.enabled flag set to
prometheus: migration: enabled: true alertmanager: migration: enabled: true
This will update the resources and deploy a one-time job to copy the existing data into the new volumes.
Wait until the job’s pod has completed and then deploy the charts again, this time disabling the migration again by
migration configuration key from your
values.yaml. This will remove the one-time job and instead
deploy the new StatefulSet.
Once you have verified that both applications work, you can safely remove the old PVCs:
kubectl -n monitoring delete pvc \ prometheus-kubermatic-db-prometheus-kubermatic-0 \ prometheus-kubermatic-db-prometheus-kubermatic-1 \ alertmanager-kubermatic-db-alertmanager-kubermatic-0 \ alertmanager-kubermatic-db-alertmanager-kubermatic-1 \ alertmanager-kubermatic-db-alertmanager-kubermatic-2
Alertmanager 0.17 changed the included
amtool to be incompatible with previous versions. This only affects you
if you ever exported silences and wanted to re-import them. Please consult the
release notes for more information.
alertmanager.version field in the Chart’s
values.yaml was deprecated and replaced with the more common
alertmanager.image.tag field. Update your Helm values if you have ever set the Alertmanager’s version explicitly.
migrate-values command of the KKP-Installer can automate this migration step for you.
alertmanager.resources.storage has been renamed to
alertmanager.storageSize and the old key has
been deprecated. Please update your
values.yaml if you’ve ever increased the volume size for Alertmanager.
The top-level key in the
values.yaml has been fixed and is now called
nodePortProxy. Please update your Helm
values as the old name (
nodePortPoxy) is now deprecated.
KKP 2.11 ships with cert-manager version 0.8, which slightly changed how certificates and issuers are configured. The 0.8 release is backwards compatible, but this is going to be removed until version 1.0. It’s recommended that KKP customers upgrade their certificates already to be future-proof.
Please consult the upgrade notes
for more information. In most cases it should be enough to simply remove the
spec.acme field from all
Certificate resources. As noted in the documentation, you can use
kubectl get certificates --all-namespaces
to see which are still using the pre-0.8 syntax.
Action required: Because Helm cannot update CRDs, you must manually update the CRDs for this release by running
kubectl apply -f config/cert-manager/templates/crd.yaml. Failure to do so will break certificate renewal because
of spec validation failures. If you notice that the cert-manager is logging that orders are invalid, make sure the
CRDs are up-to-date and you restarted the cert-manager pods.