operator: Change operator management to cluster scope #974
Conversation
|
This PR is stale because it has been open 5 days with no activity. Remove stale label or comment or this will be closed in 5 days. |
RafalKorepta
force-pushed
the
chris/p/k8s-586-cluster-scope
branch
from
fb02a9c to
0441b5d
Compare
RafalKorepta
requested review from
andrewstucki,
chrisseto and
gene-redpanda
as code owners
RafalKorepta
changed the title
RafalKorepta
force-pushed
the
chris/p/k8s-586-cluster-scope
branch
from
0441b5d to
8f3131a
Compare
|
Am I right that after this update the operator will allow deploying the operator itself and redpanda resources into different namespaces? |
|
@he0s Yes, that's the plan |
RafalKorepta
force-pushed
the
chris/p/k8s-586-cluster-scope
branch
from
8f3131a to
8ccedaa
Compare
|
@RafalKorepta that's great! I tried to deploy a cluster with using the operator today and I wanted to have the feature in operator, since it's required to deploy operator and a new cluster into separate namespaces. |
RafalKorepta
force-pushed
the
chris/p/k8s-586-cluster-scope
branch
3 times, most recently
from
2934270 to
d6516dd
Compare
RafalKorepta
changed the title
RafalKorepta
force-pushed
the
chris/p/k8s-586-cluster-scope
branch
2 times, most recently
from
6e857c2 to
0230b93
Compare
RafalKorepta
force-pushed
the
chris/p/k8s-586-cluster-scope
branch
from
0230b93 to
837932a
Compare
andrewstucki
left a comment
andrewstucki
left a comment
There was a problem hiding this comment.
From a basic pass LGTM. Even though the next initial release is light. I would want to verify that the upgrade works smoothly before shipping this though.
|
@andrewstucki https://github.com/redpanda-data/cloudv2/pull/22882 |
RafalKorepta
force-pushed
the
chris/p/k8s-586-cluster-scope
branch
from
837932a to
430d5ad
Compare
|
This PR is stale because it has been open 5 days with no activity. Remove stale label or comment or this will be closed in 5 days. |
chrisseto
force-pushed
the
chris/p/k8s-586-cluster-scope
branch
from
430d5ad to
90a415d
Compare
In default helm deployment the operator will managed its custom resources in multiple namespaces. The Role rules are moved to ClusterRole resource to give enough privileges to operator process.
chrisseto
force-pushed
the
chris/p/k8s-586-cluster-scope
branch
from
90a415d to
67577eb
Compare
|
This PR is in good shape right now. The added upgrade tests will currently fail because they're trying to install CRDs for v25.1.x that don't exist. I don't think it makes sense to keep harpoon working the way it does as it leaves us blind to bugs that might occur from CRD changes. I'm looking into adding vCluster isolation for the upgrade tests. We'll also need to verify that the kustomize installation path used by cloud works as expected or just upgrade them to using the helm chart. (My preference would be for the latter). |
|
This PR is stale because it has been open 5 days with no activity. Remove stale label or comment or this will be closed in 5 days. |
chrisseto
left a comment
chrisseto
left a comment
There was a problem hiding this comment.
Strong preference to use crds.enabled=true so we don't accidentally miss any regressions. We can punt if there's an additional complication I'm overlooking.
| redpandav1alpha2 "github.com/redpanda-data/redpanda-operator/operator/api/redpanda/v1alpha2" | ||
| ) | ||
|
|
||
| // this is a nasty hack due to the fact that we can't disable the linter for typecheck |
There was a problem hiding this comment.
😬 hadn't seen that issue before
There was a problem hiding this comment.
Yeah, it's gross, it happens because the linter freaks out when we're only doing type assertions/casting with an imported package (the places where we're upcasting runtime.Object to client.Object via o := obj.(client.Object)). Output looks like this as a linter failure and is unable to be suppressed:
acceptance/steps/k8s.go:13:2: "sigs.k8s.io/controller-runtime/pkg/client" imported and not used (typecheck)
--
| "sigs.k8s.io/controller-runtime/pkg/client"
this is the only way of getting around it since I can't mute it. We can probably upgrade golangci-lint to see if any newer versions don't have this issue, but it's pretty 🤮 as is.
There was a problem hiding this comment.
upgrading golangci-lint is generally pretty painful punt that to another day / week / month 😛
In default helm deployment the operator will managed its custom resources in
multiple namespaces. The Role rules are moved to ClusterRole resource to give
enough privileges to operator process.
K8S-586