Repository navigation
[occm] Reduce ClusterRole scope - #3165
Merged
kubernetes-prow[bot] merged 4 commits intoAug 27, 2026
Merged
Conversation
kubernetes-prow
Bot
added
release-note
Contributor
|
/lgtm looks like this needs a bump to a new chart version |
Contributor
|
@winiciusallan: changing LGTM is restricted to collaborators DetailsIn response to this:
Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes-sigs/prow repository. |
mandre
requested changes
Aug 27, 2026
OCCM does not need to create or delete nodes Signed-off-by: Stephen Finucane <stephenfin@redhat.com>
OCCM does not need this access. Signed-off-by: Stephen Finucane <stephenfin@redhat.com>
OCCM does not query Kubernetes secrets via the Kubernetes API in any of
its controllers. The only secret is needs access to is the
'cloud-config' secret mounted as a volume into the container. This
exists in the 'kube-system' namespace for standard manifests and the
release namespace when deployed via Helm. Therefore it's unnecessary to
grand cluster-wide get, list and watch access to all secrets via the
system:cloud-controller-manager ClusterRole.
Fix this by introducing a new
'system:cloud-controller-manager:secret-reader' Role in the standard
manifests and a '{{ .Values.clusterRoleName }}:secret-reader' Role in
the Helm charts.
Signed-off-by: Stephen Finucane <stephenfin@redhat.com>
stephenfin
force-pushed
the
reduce-clusterrole-scope
branch
from
August 27, 2026 15:15
3fcab16 to
d8e9a5d
Compare
Signed-off-by: Stephen Finucane <stephenfin@redhat.com>
kubernetes-prow
Bot
added
size/L
kubernetes-prow
Bot
added
the
lgtm
Contributor
|
[APPROVALNOTIFIER] This PR is APPROVED This pull-request has been approved by: mandre The full list of commands accepted by this bot can be found here. The pull request process is described here DetailsNeeds approval from an approver in each of these files:
Approvers can indicate their approval by writing |
kubernetes-prow
Bot
added
the
approved
Contributor
|
/cherry-pick release-1.36 release-1.35 release-1.34 |
|
@mandre: #3165 failed to apply on top of branch "release-1.36": DetailsIn response to this:
Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes-sigs/prow repository. |
kubernetes-prow Bot
pushed a commit
that referenced
this pull request
Sep 9, 2026
* [occm] Include cloudConfigContents in DaemonSet checksum annotation (#3080) * occm: include cloudConfigContents in DaemonSet checksum annotation When cloud config is supplied via cloudConfigContents rather than the structured cloudConfig values, changes to it do not alter the checksum annotation on the DaemonSet pod template. Pods are not restarted after credential rotation or endpoint changes. * Bump openstack-cloud-controller-manager chart to 2.36.1 --------- Co-authored-by: 234u34k <234u34k@users.noreply.github.com> (cherry picked from commit 231a42c) * [manila-csi-plugin, cinder-csi-plugin]: Allow cluster-id to be populated from configmap / secret (#3152) * feat(manila-csi): allow cluster-id to be pulled from env variables Signed-off-by: Jack Charlie Munday <1602077@users.noreply.github.com> * feat(cinder-csi): allow cluster-id to be pulled from env variables Signed-off-by: Jack Charlie Munday <1602077@users.noreply.github.com> * chore: bump chart versions for manila & cinder Signed-off-by: Jack Charlie Munday <1602077@users.noreply.github.com> --------- Signed-off-by: Jack Charlie Munday <1602077@users.noreply.github.com> Co-authored-by: 1602077 <1602077@users.noreply.github.com> (cherry picked from commit 78fca23) * Fix config drive mount and add flag to control the search order in which the driver retrieves instance metadata (#3141) * add mount for config drive and add search order flag * rebase and fix conflict * remove duplicate searchorder flag * remove trail newline * Apply suggestion from @kayrus * mount /dev/disk instead of /dev * bump chart version * mount /dev to get access to the config-drive --------- Co-authored-by: pýrus <kayrus@users.noreply.github.com> (cherry picked from commit 32a5622) * cinder-csi: make csi-snapshotter sidecar optional (#3078) Co-authored-by: 234u34k <234u34k@users.noreply.github.com> (cherry picked from commit ff04631) * [occm] Reduce ClusterRole scope (#3165) * [occm] Limit verbs for node access OCCM does not need to create or delete nodes Signed-off-by: Stephen Finucane <stephenfin@redhat.com> * [occm] Prevent access to PVs OCCM does not need this access. Signed-off-by: Stephen Finucane <stephenfin@redhat.com> * [occm] Add new Role for secrets OCCM does not query Kubernetes secrets via the Kubernetes API in any of its controllers. The only secret is needs access to is the 'cloud-config' secret mounted as a volume into the container. This exists in the 'kube-system' namespace for standard manifests and the release namespace when deployed via Helm. Therefore it's unnecessary to grand cluster-wide get, list and watch access to all secrets via the system:cloud-controller-manager ClusterRole. Fix this by introducing a new 'system:cloud-controller-manager:secret-reader' Role in the standard manifests and a '{{ .Values.clusterRoleName }}:secret-reader' Role in the Helm charts. Signed-off-by: Stephen Finucane <stephenfin@redhat.com> * occm: Bump openstack-cloud-controller-manager chart to 2.36.2 Signed-off-by: Stephen Finucane <stephenfin@redhat.com> --------- Signed-off-by: Stephen Finucane <stephenfin@redhat.com> (cherry picked from commit cc53644) * [occm] Fix serviceMonitor labelSelector and remove deprecated bearerTokenFile auth (#3039) * servicemonitor changes * add always allow path "/metrics" * chart version bump * chart version bump * Fix indentation for labels in servicemonitor.yaml * Move flag placement to inside serviceMonitor.enabled block Co-authored-by: Jihyun Hur <huhji.elha@gmail.com> * Fix indentation and duplication * Chart version bump and rebase * Chart version bump --------- Co-authored-by: Matt Anson <matta@stackhpc.com> Co-authored-by: Jihyun Hur <huhji.elha@gmail.com> (cherry picked from commit e061006) * cinder-csi: configure topology with --with-topology (#3155) Co-authored-by: 234u34k <234u34k@users.noreply.github.com> (cherry picked from commit 16c3b98) * [occm] Make hostNetwork configurable in the Helm chart (#3097) * [occm] Make hostNetwork configurable in the Helm chart * Bump version to 2.36.1 in Chart.yaml * Update Chart.yaml * [occm] Drop dnsPolicy override logic, change default in values.yaml Per review, the template no longer derives dnsPolicy from hostNetwork. The default dnsPolicy is now ClusterFirstWithHostNet in values.yaml, matching the default hostNetwork: true. Users disabling hostNetwork should set dnsPolicy: ClusterFirst. Bump chart to 2.36.4 since master already moved to 2.36.3. (cherry picked from commit e87f547) * [cinder-csi-plugin] Reduce caps of manifests, charts (#3181) * cinder-csi-plugin: Add note on why hostNetwork is required Signed-off-by: Stephen Finucane <stephenfin@redhat.com> * cinder-csi-plugin: Drop redundant capabilities add `privileged: true` already implies this capability. Signed-off-by: Stephen Finucane <stephenfin@redhat.com> * cinder-csi-plugin: Add securityContext for node plugin sidecars Neither sidecar requires root privileges nor a writable filesystem. They only talk to the CSI socket. securityContext was already present in the charts but was unset. It was missing from the manifests. We now set it in both places. Signed-off-by: Stephen Finucane <stephenfin@redhat.com> * cinder-csi-plugin: Add securityContext for all controller plugin containers Unlike the node plugin, which requires the ability to e.g. mount things, the controller plugin only talks to the k8s and OpenStack APIs. Thus, it is possible to add a strict policy check for all the containers here. Like the node plugin though, securityContext was already present in the charts but was unset, while the configuration was missing from the manifests. We set it in both places. Signed-off-by: Stephen Finucane <stephenfin@redhat.com> * cinder-csi-plugin: Set dnsPolicy This was already set in the charts but was missing in the static manifests. We want to use the cluster DNS, not the hosts (which would otherwise happen due to hostNetwork being true). Signed-off-by: Stephen Finucane <stephenfin@redhat.com> * cinder-csi-plugin: Set priorityClassName These pods handle mounts. If they get evicted, nodes cannot unmount volumes which in turn prevents node drain. Other providers already set this [1][2]. Note that we currently have a priorityClassName variable in the charts, but this is a single global value meaning the you couldn't set it to anything but system-node-critical, which is the default, since system-cluster-critical would under-privilege the node DaemonSet. We simply ignore this old, unusable value if set. [1] https://github.com/kubernetes-sigs/aws-ebs-csi-driver/blob/36d7fd880c3c4ff3b29212e2c5df8aca6628eda6/charts/aws-ebs-csi-driver/templates/_node.tpl#L48 [2] https://github.com/kubernetes-sigs/azuredisk-csi-driver/blob/6bc2e5c4608b9874f093620eaf1865ac80b93e1c/charts/v1.28.7/azuredisk-csi-driver/templates/csi-azuredisk-node.yaml#L56 Signed-off-by: Stephen Finucane <stephenfin@redhat.com> * cinder-csi-plugin: Bump cinder-csi-plugin chart to 2.36.4 Signed-off-by: Stephen Finucane <stephenfin@redhat.com> * tests: Report kernel features on hosts So we can determine whether things like seccomp are supported. Signed-off-by: Stephen Finucane <stephenfin@redhat.com> --------- Signed-off-by: Stephen Finucane <stephenfin@redhat.com> (cherry picked from commit f16c52a) * [manila-csi-plugin] Reduce caps of manifests, charts (#3182) * manila-csi-plugin: Add note on why hostNetwork is required Signed-off-by: Stephen Finucane <stephenfin@redhat.com> * manila-csi-plugin: Drop redundant capabilities add `privileged: true` already implies this capability. Signed-off-by: Stephen Finucane <stephenfin@redhat.com> * manila-csi-plugin: Add securityContext for node plugin sidecars The registrar sidecar requires neither root privileges nor a writable filesystem: it only talks to the CSI socket. Add a relevant securityContext. Signed-off-by: Stephen Finucane <stephenfin@redhat.com> * manila-csi-plugin: Add securityContext for all controller plugin containers Unlike the node plugin, which requires the ability to e.g. mount things, the controller plugin only talks to the k8s and OpenStack APIs. Thus, it is possible to add a strict policy check for all the containers here. Signed-off-by: Stephen Finucane <stephenfin@redhat.com> * manila-csi-plugin: Set priorityClassName These pods handle mounts. If they get evicted, nodes cannot unmount volumes which in turn prevents node drain. Other providers already set this [1][2]. [1] https://github.com/kubernetes-sigs/aws-ebs-csi-driver/blob/36d7fd880c3c4ff3b29212e2c5df8aca6628eda6/charts/aws-ebs-csi-driver/templates/_node.tpl#L48 [2] https://github.com/kubernetes-sigs/azuredisk-csi-driver/blob/6bc2e5c4608b9874f093620eaf1865ac80b93e1c/charts/v1.28.7/azuredisk-csi-driver/templates/csi-azuredisk-node.yaml#L56 Signed-off-by: Stephen Finucane <stephenfin@redhat.com> * manila-csi-plugin: Bump manila-csi-plugin chart to 2.36.3 Signed-off-by: Stephen Finucane <stephenfin@redhat.com> --------- Signed-off-by: Stephen Finucane <stephenfin@redhat.com> (cherry picked from commit 3db62bf) * [cinder-csi-plugin] Allow setting parameters on chart-managed StorageClasses (#3173) Add optional storageClass.delete.parameters and storageClass.retain.parameters values to the Helm chart, rendered into the respective StorageClass. Defaults to {}, rendered output is unchanged when unset. Signed-off-by: Dean Chen <862469039@qq.com> (cherry picked from commit c36c84d) * [occm] Reduce caps of manifests, charts (#3191) * openstack-cloud-controller-manager: Add note on why hostNetwork is required Signed-off-by: Stephen Finucane <stephenfin@redhat.com> * openstack-cloud-controller-manager: Add securityContext for controller containers The OCCM only talks to the Kubernetes and OpenStack APIs and does not need write access to the root filesystem. Unlike the CSI sidecar containers, the pod already runs as a non-root user (runAsUser: 1001), so we can additionally set allowPrivilegeEscalation and runAsNonRoot. Signed-off-by: Stephen Finucane <stephenfin@redhat.com> * openstack-cloud-controller-manager: Set dnsPolicy We want to use the cluster DNS, not the host's, which would otherwise happen due to hostNetwork being true. Signed-off-by: Stephen Finucane <stephenfin@redhat.com> * openstack-cloud-controller-manager: Set priorityClassName If the OCCM is evicted, new nodes cannot be initialized and will remain in the NotReady state. Other providers already set this. Signed-off-by: Stephen Finucane <stephenfin@redhat.com> * openstack-cloud-controller-manager: Drop Bitnami common chart dependency The dependency was used solely for the common.labels.standard helper, which generates standard Kubernetes recommended labels. This is trivially replicated inline in _helpers.tpl, and the other CSI charts in this repo already manage without it. Signed-off-by: Stephen Finucane <stephenfin@redhat.com> * openstack-cloud-controller-manager: Bump openstack-cloud-controller-manager chart to 2.36.5 We also change the maintainer since Frank has not been active in CPO for many years now. Signed-off-by: Stephen Finucane <stephenfin@redhat.com> * openstack-cloud-controller-manager: Drop all capabilities The OCCM only makes API calls to Kubernetes and OpenStack and requires no Linux capabilities. Signed-off-by: Stephen Finucane <stephenfin@redhat.com> * openstack-cloud-controller-manager: Use stable labels Do not match on 'helm.sh/chart' and 'app.kubernetes.io/version' in the ServiceMonitor selector since they change on chart bump or app bumps respectively and would cause Prometheus to lose its scrape target after an upgrade. Instead use 'app.kubernetes.io/name' and 'app.kubernetes.io/instance', which are stable across upgrades and sufficient to uniquely identify the Service. These are extracted into a new 'occm.labels.matchLabels' helper for reuse. Note that we only change things for ServiceMonitor. While both the Service and DaemonSet use old-style app/release/component labels, we can't update the DaemonSet without a major version change since this would be a breaking change and we can't update the Service without updating the DaemonSet since doing so would stop traffic getting routed correctly. We'll need to address these in a follow-up that is only included in the new release. Signed-off-by: Stephen Finucane <stephenfin@redhat.com> --------- Signed-off-by: Stephen Finucane <stephenfin@redhat.com> (cherry picked from commit f3ef836) --------- Signed-off-by: Jack Charlie Munday <1602077@users.noreply.github.com> Signed-off-by: Stephen Finucane <stephenfin@redhat.com> Signed-off-by: Dean Chen <862469039@qq.com> Co-authored-by: 234u34k <ashburn.tech+github@gmail.com> Co-authored-by: 234u34k <234u34k@users.noreply.github.com> Co-authored-by: jack munday <62025739+1602077@users.noreply.github.com> Co-authored-by: 1602077 <1602077@users.noreply.github.com> Co-authored-by: Diana Gaponcic <37191907+dgaponcic@users.noreply.github.com> Co-authored-by: pýrus <kayrus@users.noreply.github.com> Co-authored-by: Claudia Watson <claudia.lola.watson@gmail.com> Co-authored-by: Matt Anson <matta@stackhpc.com> Co-authored-by: Jihyun Hur <huhji.elha@gmail.com> Co-authored-by: Denis GERMAIN <dt.germain@gmail.com> Co-authored-by: Dean Chen <51218137+locker95@users.noreply.github.com>
kubernetes-prow Bot
pushed a commit
that referenced
this pull request
Sep 22, 2026
#3165 dropped create/delete on nodes. the node lifecycle controller still needs delete so OCCM can remove nodes whose OpenStack VMs are already gone. without it, those nodes stay NotReady forever. create stays omitted. Signed-off-by: Dean Chen <862469039@qq.com>
shiftstack-merge-bot-cloner Bot
pushed a commit
to shiftstack/cloud-provider-openstack
that referenced
this pull request
Oct 15, 2026
* [occm] Limit verbs for node access
OCCM does not need to create or delete nodes
Signed-off-by: Stephen Finucane <stephenfin@redhat.com>
* [occm] Prevent access to PVs
OCCM does not need this access.
Signed-off-by: Stephen Finucane <stephenfin@redhat.com>
* [occm] Add new Role for secrets
OCCM does not query Kubernetes secrets via the Kubernetes API in any of
its controllers. The only secret is needs access to is the
'cloud-config' secret mounted as a volume into the container. This
exists in the 'kube-system' namespace for standard manifests and the
release namespace when deployed via Helm. Therefore it's unnecessary to
grand cluster-wide get, list and watch access to all secrets via the
system:cloud-controller-manager ClusterRole.
Fix this by introducing a new
'system:cloud-controller-manager:secret-reader' Role in the standard
manifests and a '{{ .Values.clusterRoleName }}:secret-reader' Role in
the Helm charts.
Signed-off-by: Stephen Finucane <stephenfin@redhat.com>
* occm: Bump openstack-cloud-controller-manager chart to 2.36.2
Signed-off-by: Stephen Finucane <stephenfin@redhat.com>
---------
Signed-off-by: Stephen Finucane <stephenfin@redhat.com>
shiftstack-merge-bot-cloner Bot
pushed a commit
to shiftstack/cloud-provider-openstack
that referenced
this pull request
Oct 15, 2026
kubernetes#3165 dropped create/delete on nodes. the node lifecycle controller still needs delete so OCCM can remove nodes whose OpenStack VMs are already gone. without it, those nodes stay NotReady forever. create stays omitted. Signed-off-by: Dean Chen <862469039@qq.com>
kubernetes-prow Bot
pushed a commit
that referenced
this pull request
Oct 15, 2026
#3165 dropped create/delete on nodes. the node lifecycle controller still needs delete so OCCM can remove nodes whose OpenStack VMs are already gone. without it, those nodes stay NotReady forever. create stays omitted. Signed-off-by: Dean Chen <862469039@qq.com> Co-authored-by: Dean Chen <862469039@qq.com>
shiftstack-merge-bot-cloner Bot
pushed a commit
to shiftstack/cloud-provider-openstack
that referenced
this pull request
Oct 26, 2026
* [occm] Limit verbs for node access
OCCM does not need to create or delete nodes
Signed-off-by: Stephen Finucane <stephenfin@redhat.com>
* [occm] Prevent access to PVs
OCCM does not need this access.
Signed-off-by: Stephen Finucane <stephenfin@redhat.com>
* [occm] Add new Role for secrets
OCCM does not query Kubernetes secrets via the Kubernetes API in any of
its controllers. The only secret is needs access to is the
'cloud-config' secret mounted as a volume into the container. This
exists in the 'kube-system' namespace for standard manifests and the
release namespace when deployed via Helm. Therefore it's unnecessary to
grand cluster-wide get, list and watch access to all secrets via the
system:cloud-controller-manager ClusterRole.
Fix this by introducing a new
'system:cloud-controller-manager:secret-reader' Role in the standard
manifests and a '{{ .Values.clusterRoleName }}:secret-reader' Role in
the Helm charts.
Signed-off-by: Stephen Finucane <stephenfin@redhat.com>
* occm: Bump openstack-cloud-controller-manager chart to 2.36.2
Signed-off-by: Stephen Finucane <stephenfin@redhat.com>
---------
Signed-off-by: Stephen Finucane <stephenfin@redhat.com>
shiftstack-merge-bot-cloner Bot
pushed a commit
to shiftstack/cloud-provider-openstack
that referenced
this pull request
Oct 26, 2026
kubernetes#3165 dropped create/delete on nodes. the node lifecycle controller still needs delete so OCCM can remove nodes whose OpenStack VMs are already gone. without it, those nodes stay NotReady forever. create stays omitted. Signed-off-by: Dean Chen <862469039@qq.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What this PR does / why we need it:
A collection of 3 changes, all intended to reduce the cluster-wide access granted by the
system:cloud-controller-managerClusterRole.[occm] Limit verbs for node access: OCCM does not need to create or delete nodes[occm] Prevent access to PVs: OCCM does not need access to PVs[occm] Add new Role for secrets: OCCM does not need cluster-wide secret access. It only needs access to the configuration secret (kube-system / cloud-configif using the standard manifests).Which issue this PR fixes(if applicable):
(none)
Special notes for reviewers:
Release note: