Installation sometimes fails

[ctron]

Sometimes the installation fails with:

error: unable to recognize "operator-lifecycle-manager/deploy/upstream/manifests/latest/0000_30_09-rh-operators.catalogsource.yaml": no matches for kind "CatalogSource" in version "operators.coreos.com/v1alpha1"

This looks like a synchronization issue, where the CRD for the CatalogSource is not properly registered before the installation proceeds

Full log:

namespace/olm created
clusterrole.rbac.authorization.k8s.io/system:controller:operator-lifecycle-manager created
serviceaccount/olm-operator-serviceaccount created
clusterrolebinding.rbac.authorization.k8s.io/olm-operator-binding-olm created
customresourcedefinition.apiextensions.k8s.io/clusterserviceversions.operators.coreos.com created
customresourcedefinition.apiextensions.k8s.io/installplans.operators.coreos.com created
customresourcedefinition.apiextensions.k8s.io/subscriptions.operators.coreos.com created
customresourcedefinition.apiextensions.k8s.io/catalogsources.operators.coreos.com created
configmap/rh-operators created
deployment.apps/olm-operator created
deployment.apps/catalog-operator created
clusterrole.rbac.authorization.k8s.io/aggregate-olm-edit created
clusterrole.rbac.authorization.k8s.io/aggregate-olm-view created
apiservice.apiregistration.k8s.io/v1alpha1.packages.apps.redhat.com created
clusterrolebinding.rbac.authorization.k8s.io/packagemanifest:system:auth-delegator created
rolebinding.rbac.authorization.k8s.io/packagemanifest-auth-reader created
clusterrolebinding.rbac.authorization.k8s.io/packagemanifest-view created
clusterrolebinding.rbac.authorization.k8s.io/package-apiserver-clusterrolebinding created
secret/package-server-certs created
deployment.apps/package-server created
service/package-server created
error: unable to recognize "operator-lifecycle-manager/deploy/upstream/manifests/latest/0000_30_09-rh-operators.catalogsource.yaml": no matches for kind "CatalogSource" in version "operators.coreos.com/v1alpha1"

[ecordell]

Thanks for the report @ctron!

For now you should be able to re-create any objects that failed. Longer term we may want to wrap up installation into a script that waits for the types to be available before continuing.

[ctron]

Yes, re-running works. However, if you want to run an automatic installation, you always have to run twice, with a sleep in the middle, getting lots of error messages, and hoping the sleep is long enough. Not a good solution.

[ecordell]

OLM comes by default in OpenShift 4.0 and is installed by CVO - however, this is still an issue for upstream users, so I'll leave this issue open.

[ecordell]

Hi @ctron. We have now split the crds and other objects into separate artifacts (so you can create the crds and wait for them to register before installing olm) and have provided an install script to help with it. Check it out here: https://github.com/operator-framework/operator-lifecycle-manager/releases/tag/0.10.0