- All pods in the
openshift-monitoring
namespace are now ignored when determining whether a cluster is ready. - All pods in
openshift-pipeline.*
namespaces are now ignored when determining whether a cluster is ready.
- All pods in the
openshift-marketplace
are now ignored when determining whether a cluster is ready. Sometimes those pods can crashloop if quay.io is slow or down but the cluster otherwise functions.
- The
openshift-operators-redhat
,openshift-serverless
, andopenshift-cnv
namespaces are now excluded from pod readiness checks to determine cluster health. None of these namespaces have an impact on the overall cluster availability.
- Fixed OCP 4.3 / K8s 1.16 compatibility in route-helper pod by
explicitly using
deployment.apps
instead ofdeployment
inoc
commands
- Release artifacts now split out the default bundles from the rest of the release. This allows a user to update the operator without overwriting their bundle configurations.
- Add a new
spec.url
field toCrcBundle
resources to specify an optional URL that will be used, if given, when provisioning persistent clusters. This URL should point to a qcow2 or raw virtual machine image over HTTP or HTTPS. This significantly speeds up the initial provisioning time of persistent clusters compared to using a container image. Container images are still used for all non-persistent clusters.
- Added an OpenShift 4.5.0-rc.6 bundle in the default install
- Added an OpenShift 4.5.0-rc.5 bundle in the default install
- The operator will no longer wait for pods in the
openshift-operators
namespace to come up before marking a cluster as ready. These pods are cluster-wide optional operators and not required for core functioning of the cluster.
- Added an OpenShift 4.5.0-rc.4 bundle
- Every cluster gets a unique SSH key generated for the Node when starting the cluster and the default one from the bundle revoked.
- Added an OpenShift 4.4.8 bundle in the default install
- You can now enable monitoring on CrcClusters with the new
spec.enableMonitoring
*bool field. A value of true enables monitoring, a value of false disables it, and an unset value doesn't express an opinion.
- The route helper pod is now cleaned up when stopping a cluster and redeployed when starting that cluster again.
- You can now explicitly stop (but not delete) CrcClusters with the
new
spec.stopped
boolean field. A value of true stops the cluster, a value of false starts the cluster. The default is false, meaning running. There's a correspondingstatus.stopped
that can be read to know when the operator has actually stopped the cluster. This can work for clusters with ephemeral or persistent storage, but ephemeral clusters lose all their data between the stop and start. Persistent clusters retain their data. - Only pods in the
openshift-*
orkube-*
namespaces are considered when checking for cluster readiness. The starting and stopping work above exposed a problem where any random pod in the cluster could cause the operator to wait forever for readiness when it may be that random user pods were not coming up cleanly before stopping the cluster.
- A new
CrcBundle
API controls which CRC bundles are available for use. The default installation installs bundles for recent versions of OCP. - There is now optionally persistent storage with configurable disk
size for CRC clusters. The default is ephemeral (not persistent)
storage. Set
spec.storage.persitent
totrue
in yourCrcCluster
to enable persistence. Setspec.storage.size
to a value like100Gi
to specify the size of the persistent storage. The underlying Node in the CRC cluster will automatically grow its root partition to the specified size.
- You may now specify which CRC bundle (and thus OCP version) to start
with a new
bundleName
field in the CRD spec. Valid bundle names areocp445
for OCP 4.4.5,ocp450rc1
for OCP 4.5.0-rc.1, andocp450rc2
for OCP 4.5.0-rc.2. - All the default OpenShift Routes are now updated to use the real domain. Previously, the image registry, console downloads, and various monitoring routes still had the old *.crc.testing domain.
- The API Server URL as shown in the Console overview and 'Copy Login Command' screens is now correct.
- Don't wait for the community-operator or certified-operator pods in the openshift-marketplace namespace to become Ready before declaring the CrcCluster as Ready. These pods may crashloop for some time after the rest of the cluster is ready if multiple clusters are started up around the same time on a shared parent cluster where each CRC cluster appears as the same IP to quay.io, thus causing rate-limiting.
- Initial release