Permissions Required to Use CloudPilot AI
CloudPilot AI uses a two-phase installation. Phase 1 installs the Agent for cluster discovery, reporting, and CloudPilot component management. Most inventory access is read-only, but the installation also grants the write permissions listed below for policy synchronization, CRD upgrades, diagnostics, and component updates. Phase 2 (optional) enables cluster optimization and requires additional Kubernetes RBAC and cloud provider permissions.
The tables below describe the published Node v1.23.1 and Workload Autoscaler v0.9.0 installation artifacts. Permissions depend on the installed version, cloud provider, and enabled components. Review the manifests and rendered charts for the version you install. An image-only upgrade does not update existing RBAC; permission changes require applying the corresponding manifests or Helm release.
CloudPilot AI follows the principle of least privilege. You only need to grant CloudPilot AI the minimal permissions required to perform its operations — no more. Here is the overall architecture:
Note:: All credentials are mounted in your local cluster components and will not be stored or synced to our servers or the public internet.
Phase 1: Agent Permissions
The Phase 1 manifest binds the cloudpilot-agent service account to a cluster role and a role in the cloudpilot namespace. Cluster scope below means access across namespaces for namespaced resources. The base Phase 1 manifest grants no ConfigMap access.
Agent Cluster Permissions
| API Group | Resources | Verbs | Purpose |
|---|---|---|---|
batch | jobs, cronjobs | get, list, watch | Discover batch workloads and their ownership. |
batch | cronjobs | patch | Synchronize the proactive-update label on CronJobs. |
karpenter.k8s.aws | ec2nodeclasses | get, list, watch | Read AWS NodeClass metadata and status. |
karpenter.k8s.gcp | gcenodeclasses | get, list, watch | Read GCP NodeClass metadata and status. |
karpenter.azure.com | aksnodeclasses | get, list, watch | Read Azure NodeClass metadata and status. |
karpenter.sh | nodeclaims, nodepools | get, list, watch | Read node provisioning, readiness, and pool configuration. |
storage.k8s.io | csinodes | get, list, watch | Discover CSI/node storage topology. |
"" (core) | namespaces, nodes, pods, events, persistentvolumeclaims, persistentvolumes | get, list, watch | Collect cluster, workload, event, and storage inventory. |
"" (core) | nodes/proxy | get | Read disk and PVC usage through the kubelet stats proxy. |
"" (core) | events | create | Publish Kubernetes Events. |
apps | deployments, daemonsets, statefulsets, replicasets | get, list, watch | Read workload controllers for optimization analysis. |
policy | poddisruptionbudgets | get, list, watch | Evaluate disruption constraints. |
metrics.k8s.io | pods, nodes | get, list, watch | Read runtime resource metrics. |
apiextensions.k8s.io | customresourcedefinitions | get, list, watch, create, patch | Discover and apply CRDs during component upgrades. |
evpa.cloudpilot.ai | autoscalingpolicyconfigurations, autoscalingpolicyconfigurations/status | get, list, watch | Read autoscaling configuration. |
evpa.cloudpilot.ai | autoscalingpolicies, autoscalingpolicies/status, recommendationpolicies | create, update, list, watch, patch, delete | Synchronize autoscaling and recommendation policies. |
Agent Namespace Permissions
All permissions in this table apply only in the cloudpilot namespace. Component update permissions are further restricted to the listed resource names.
| API Group | Resources | Verbs | Purpose |
|---|---|---|---|
"" (core) | pods/log | get | Read CloudPilot component logs for warning diagnostics. |
coordination.k8s.io | leases | get, list, create, update, patch, watch | Perform leader election and coordination. |
apps | deployments (resourceNames: cloudpilot-agent, cloudpilot-workload-autoscaler, cloudpilot-controller, cloudpilot-webhook, cloudpilot-workload-log-collector, cloudpilot-aws-optimizer, cloudpilot-alibabacloud-optimizer, cloudpilot-gcp-optimizer, cloudpilot-azure-optimizer) | get, update, patch | Update named CloudPilot Deployments. |
apps | daemonsets (resourceNames: cloudpilot-node-agent, cloudpilot-nri-plugin, cloudpilot-spot-handler, image-accelerator) | get, update, patch | Update named CloudPilot DaemonSets. |
apps | statefulsets (resourceNames: cloudpilot-victoria-metrics) | get, update, patch | Update the CloudPilot metrics StatefulSet. |
"" (core) | pods, events | get, list, watch | Observe CloudPilot runtime Pods and namespace Events. |
apps | replicasets | get, list, watch | Observe Deployment rollouts. |
The Agent’s log permission is namespace-scoped and only grants get; its node-proxy permission also only grants get. Additional Agent permissions installed with Workload Autoscaler are listed below.
Phase 2: Additional Permissions for Cluster Optimization
If you proceed to Phase 2, CloudPilot AI will deploy components that actively optimize your cluster. This includes managing Kubernetes workloads and interacting with your cloud provider’s infrastructure APIs (for example, EC2, Auto Scaling Groups, GKE / Compute Engine resources, or AKS / Azure infrastructure resources).
Kubernetes RBAC Permissions
The standard Phase 2 installation runs the optimizer and shared CloudPilot components as cloudpilot/cloudpilot-admin. The base chart binds this account to both Phase 1 Agent roles, so its effective permissions include the entire Phase 1 table above. Kubernetes adds the permissions from all bindings; a narrower rule in one role does not remove access granted by another role.
The following tables describe the additional rules in the published Node Autoscaler v1.23.1 optimizer and base charts. They describe the default installation; custom Helm RBAC rules or a different service account can change the result.
Shared optimizer and base permissions
| Scope | API Group | Resources | Verbs | Purpose |
|---|---|---|---|---|
| Cluster | "" (core) | pods | get, list, watch, delete | Observe workloads and remove Pods during optimization. |
| Cluster | "" (core) | nodes | get, list, watch, update, patch, delete | Inspect and manage node lifecycle. |
| Cluster | "" (core) | persistentvolumes, persistentvolumeclaims, replicationcontrollers, namespaces | get, list, watch | Read workload, storage, and namespace inventory. |
| Cluster | "" (core) | configmaps | get, list, watch | Read configuration through the base chart’s dedicated ConfigMap role. |
| Cluster | "" (core) | pods/eviction | create | Evict workloads during node changes. |
| Cluster | "" (core) | events | create, patch | Publish operational events. |
| Cluster | storage.k8s.io | storageclasses, csinodes, volumeattachments | get, list, watch | Discover storage capabilities and attachments. |
| Cluster | apps | deployments, statefulsets | get, list, watch, update, patch | Read and adjust supported workload controllers. |
| Cluster | apps | daemonsets, replicasets | list, watch | Observe workload controllers; Phase 1 also supplies get. |
| Cluster | policy | poddisruptionbudgets | get, list, watch | Respect disruption constraints. |
| Cluster | batch | jobs | get, list, watch | Observe batch workloads. |
Namespace: cloudpilot | "" (core) | configmaps | create | Create runtime configuration. |
Namespace: cloudpilot | "" (core) | configmaps (resourceName: cloudpilot-setting) | get, update, patch | Persist cluster settings. |
Namespace: kube-node-lease | coordination.k8s.io | leases | get, list, watch, delete | Observe and clean up node leases. |
Phase 1 already supplies CRD discovery and creation/patching, Agent diagnostics, policy access, and namespaced CloudPilot component management. Those inherited permissions remain available to cloudpilot-admin in addition to the rules below.
Provider-specific cluster permissions
These rows add to the shared table. * means all verbs or all resources, as specified; it is not shorthand for a smaller set of operations.
| Provider | API Group | Resources | Additional verbs | Scope or condition |
|---|---|---|---|---|
| AWS | karpenter.k8s.aws | ec2nodeclasses | get, list, watch, create, delete, patch, update | Cluster. |
| AWS | karpenter.k8s.aws | ec2nodeclasses/status | patch, update | Cluster. |
| AWS | karpenter.sh | nodepools, nodepools/status, nodeclaims, nodeclaims/status | get, list, watch, create, delete, patch, update | Combined optimizer core and directly bound aggregate role. |
| AWS | apiextensions.k8s.io | customresourcedefinitions; customresourcedefinitions/status | update for CRDs; patch for status | Restricted to ec2nodeclasses.karpenter.k8s.aws, nodepools.karpenter.sh, and nodeclaims.karpenter.sh. |
| AWS, Azure, Alibaba Cloud | "" (core) | pods, persistentvolumes, persistentvolumeclaims, replicationcontrollers, namespaces | update, patch | Cluster. |
| AWS, GCP, Azure | "" (core) | configmaps | update, patch | Cluster. |
| AWS, Azure, Alibaba Cloud | apps | daemonsets, replicasets | get, update, patch | Cluster. |
| AWS, Alibaba Cloud | "" (core) | pods/log | get, list, watch, update, patch | Cluster; these Phase 2 roles are broader than the Phase 1 log rule. |
| AWS, Alibaba Cloud | admissionregistration.k8s.io | mutatingwebhookconfigurations, validatingwebhookconfigurations | get, list, watch, update | Cluster. |
| AWS, Alibaba Cloud | certificates.k8s.io | * | get, create, update, patch, delete, approve | Current optimizer certificate rules; the Workload Autoscaler CSR removal does not remove these rules. |
| AWS, Alibaba Cloud | agent.cloudpilot.ai | podmutations | get, list, watch, create, update, patch, delete | Cluster. |
| GCP | karpenter.k8s.gcp | gcenodeclasses, gcenodeclasses/status | * | Directly bound aggregate role. |
| GCP | karpenter.sh | nodepools, nodepools/status, nodeclaims, nodeclaims/status, nodeoverlays, nodeoverlays/status | * | Directly bound aggregate role. |
| GCP | "" (core) | pods/log | get | Cluster. |
| GCP | admissionregistration.k8s.io | mutatingwebhookconfigurations, validatingwebhookconfigurations | get, list, watch, update, patch | Cluster. |
| GCP | agent.cloudpilot.ai | podmutations | get, list, watch | Cluster. |
| GCP | certificates.k8s.io | certificatesigningrequests, certificatesigningrequests/status; certificatesigningrequests/approval | get, list, watch, create, update, patch, delete for requests and status; update, patch for approval | Cluster. |
| GCP | certificates.k8s.io | signers | approve | ResourceNames: kubernetes.io/kube-apiserver-client-kubelet, kubernetes.io/kubelet-serving. |
| GCP, Alibaba Cloud | coordination.k8s.io | leases | get, list, watch; update, patch for karpenter-leader-election | Cluster. Alibaba Cloud also grants cluster-scoped create. |
| Azure | karpenter.azure.com | aksnodeclasses; aksnodeclasses/status | get, list, watch, create, patch, update for objects; create, patch, update for status | Cluster. |
| Azure | karpenter.sh | nodepools, nodepools/status; nodeclaims, nodeclaims/status | get, list, watch, create, update, patch; NodeClaims and their status also allow delete | Cluster. |
| Azure | karpenter.sh | nodeoverlays; nodeoverlays/status | get, list, watch for objects; get, list, watch, create, update, patch for status | Cluster. |
| Azure | admissionregistration.k8s.io | mutatingwebhookconfigurations, validatingwebhookconfigurations | get, list, watch, update, patch | When the optimizer webhook is enabled (default). |
| Azure | agent.cloudpilot.ai | podmutations | get, list, watch | When the optimizer webhook is enabled (default). |
| Azure | certificates.k8s.io | certificatesigningrequests; certificatesigningrequests/approval; signers | get, list, watch, create, update, patch for requests; update, patch for approval; approve for signer kubernetes.io/kubelet-serving | When the optimizer webhook is enabled (default). |
| Azure | "" (core) | secrets | get, create, update, patch | Cluster-wide when the optimizer webhook is enabled (default). |
| Azure | apiextensions.k8s.io | customresourcedefinitions; customresourcedefinitions/status | list, watch, update for CRDs; patch for status | Webhook-enabled CRD status patches are limited to aksnodeclasses.karpenter.azure.com, nodepools.karpenter.sh, and nodeclaims.karpenter.sh; CRD update is not resource-name restricted. |
| Alibaba Cloud | karpenter.k8s.alibabacloud | ecsnodeclasses; ecsnodeclasses/status | * for objects; patch, update for status | Directly bound aggregate role supplies all object verbs. |
| Alibaba Cloud | karpenter.sh | nodepools, nodepools/status, nodeclaims, nodeclaims/status | * | Directly bound aggregate role. |
GCP and Alibaba Cloud add node metric reads through their optimizer core roles; Phase 1 already grants metrics.k8s.io reads for both Pods and Nodes. AWS, GCP, and Alibaba Cloud directly bind their aggregate optimizer role to the optimizer service account. Azure creates an aggregate-to-admin role but does not directly bind that role to the optimizer account, so that aggregate role is not included in the Azure access listed here.
Additional namespace permissions
All rows below use the cloudpilot namespace unless another namespace is shown. The Phase 1 role inherited by cloudpilot-admin already grants get, list, watch, create, update, and patch on cloudpilot leases, including the controller and workload-log-collector leader-election objects.
| Provider | API Group | Resources | Verbs | Scope or condition |
|---|---|---|---|---|
| AWS, GCP, Alibaba Cloud | "" (core) | configmaps | get, list, watch, create, update, patch | Namespace only; cluster ConfigMap access is listed separately above. |
| AWS, GCP, Alibaba Cloud | "" (core) | secrets | get, list, watch | Namespace only. |
| AWS, Alibaba Cloud | "" (core) | secrets | update | ResourceNames: <optimizer-fullname>-cert, cloudpilot-webhook. With the standard release name, these certificate Secrets are cloudpilot-aws-optimizer-cert or cloudpilot-alibabacloud-optimizer-cert. |
| GCP | "" (core) | secrets | update, patch | ResourceName: cloudpilot-webhook. |
| Azure | "" (core) | configmaps, secrets | get, list, watch | Optimizer webhook enabled (default). Cluster-wide Secret writes are listed separately above. |
| Azure | "" (core) | secrets | update | ResourceName: cloudpilot-azure-optimizer-cert; optimizer webhook enabled (default). |
| AWS, Azure | "" (core) | services | get | Namespace: kube-system; resourceName: kube-dns. |
Optional NRI plugin
On supported EKS Kubernetes 1.34 and later, the standard installer enables the NRI plugin unless explicitly disabled. The base chart gives its separate cloudpilot/cloudpilot-nri-plugin service account cluster-scoped get on core nodes and pods. This account does not inherit the optimizer or Agent role bindings. The plugin runs only on nodes enabled for NRI Memory QoS.
These are the runtime chart grants. The account running the installation also needs permission to create or update the chart’s Kubernetes objects, CRDs, Roles, ClusterRoles, and bindings. Workload Autoscaler uses its own permission sets described later on this page.
Cloud Provider Permissions
AWS IAM Permissions
For EKS, the optimizer controller uses an IAM role through the cluster’s OIDC provider (IRSA). Its trust policy permits sts:AssumeRoleWithWebIdentity for system:serviceaccount:cloudpilot:cloudpilot-admin, with audience sts.amazonaws.com.
The following table describes the controller policy installed by the current Phase 2 script. These are runtime permissions for the controller role; they are separate from the permissions held by the person running the installer.
| Category | Actions | Resource Scope | Condition/Note |
|---|---|---|---|
| ASG discovery | autoscaling:DescribeAutoScalingGroups, autoscaling:DescribeAutoScalingInstances, autoscaling:DescribeLaunchConfigurations, autoscaling:DescribeScalingActivities, autoscaling:DescribeTags | * | Discover ASG topology and state. |
| ASG management | autoscaling:SetDesiredCapacity, autoscaling:TerminateInstanceInAutoScalingGroup, autoscaling:UpdateAutoScalingGroup | * | Scale and rebalance existing ASGs. |
| EC2 discovery | ec2:DescribeImages, ec2:DescribeInstances, ec2:DescribeInstanceTypes, ec2:DescribeInstanceTypeOfferings, ec2:DescribeAvailabilityZones, ec2:DescribeSubnets, ec2:DescribeSecurityGroups, ec2:DescribeLaunchTemplates, ec2:DescribeLaunchTemplateVersions, ec2:DescribeSpotPriceHistory, ec2:DescribeRegions, ec2:GetInstanceTypesFromInstanceRequirements | * | Discover capacity, images, networks, launch templates, and instance requirements. This is an explicit action list; the installer does not grant ec2:Describe*. |
| EC2 provisioning | ec2:RunInstances, ec2:CreateFleet, ec2:CreateLaunchTemplate, ec2:DeleteLaunchTemplate, ec2:CreateTags | * | Provision nodes and manage their launch templates and tags. |
| Images and pricing | ssm:GetParameter, pricing:GetProducts, savingsplans:DescribeSavingsPlans | * | Resolve image parameters and obtain pricing and Savings Plans information. |
| Conditional EC2 termination | ec2:TerminateInstances | * | Requires ec2:ResourceTag/kubernetes.io/cluster/${CLUSTER_NAME} to equal owned or shared. |
| Pass the node IAM role | iam:PassRole | The configured node role ARN | Restricted to the effective node role selected by the installer, including CUSTOM_NODE_ROLE when supplied. |
| Cluster metadata | eks:DescribeCluster | arn:${AWS_PARTITION}:eks:${CLUSTER_REGION}:${AWS_ACCOUNT_ID}:cluster/${CLUSTER_NAME} | Read the target cluster endpoint and metadata. |
| Node group metadata | eks:DescribeNodegroup | * | Discover managed node group configuration. |
| Instance profile creation | iam:CreateInstanceProfile | * | Requires request tags kubernetes.io/cluster/${CLUSTER_NAME}=owned, topology.kubernetes.io/region=${CLUSTER_REGION}, and karpenter.k8s.aws/ec2nodeclass. |
| Instance profile tagging | iam:TagInstanceProfile | * | Requires matching cluster and region resource/request tags, plus the karpenter.k8s.aws/ec2nodeclass resource/request tags. |
| Instance profile management | iam:AddRoleToInstanceProfile, iam:RemoveRoleFromInstanceProfile, iam:DeleteInstanceProfile | * | Requires cluster and region resource tags and the karpenter.k8s.aws/ec2nodeclass resource tag. |
| Instance profile lookup | iam:GetInstanceProfile | * | Read instance profiles used by provisioned nodes. |
The provisioned EC2 nodes use a separate node role. For an installer-managed node role, the script attaches AmazonEKSWorkerNodePolicy, AmazonEKS_CNI_Policy, AmazonEC2ContainerRegistryReadOnly, and service-role/AmazonEBSCSIDriverPolicy. When node group policy inheritance is enabled, it can also copy policies from an existing node group role. Custom node and controller roles are validated rather than rewritten by the installer.
GCP Workload Identity and IAM Permissions
For GKE, the current Phase 2 install flow requires GKE Workload Identity. CloudPilot AI binds the in-cluster cloudpilot/cloudpilot-admin service account to a Google service account (GSA) and reconciles a project-scoped custom IAM role for the controller. No downloaded JSON key is required by the default flow.
| Category | Actions | Resource Scope | Condition/Note |
|---|---|---|---|
| GKE metadata and cluster-side updates | container.clusters.get, container.clusters.list, container.clusters.update | Project | Read cluster metadata, enumerate cluster release/channel details, and apply cluster-side optimizer updates. |
| Region and zonal inventory | compute.zones.list, compute.regions.get, compute.machineTypes.list, compute.projects.get | Project | Discover zones, regions, machine type choices, and the project-level default Compute Engine service account. |
| MIG and template discovery | compute.instanceGroupManagers.get, compute.instanceGroupManagers.update, compute.instanceTemplates.get, compute.instanceTemplates.list, compute.networks.get, compute.subnetworks.get | Project | Discover the GKE-managed MIG / instance-template chain and validate network/subnetwork inputs when creating replacement nodes. |
| Image lookup | compute.images.get, compute.images.list, compute.images.useReadOnly | Project | Resolve image selector terms and template-derived boot images. |
| VM lifecycle | compute.zoneOperations.get, compute.instances.get, compute.instances.list, compute.instances.create, compute.instances.delete, compute.instances.setLabels, compute.instances.setMetadata, compute.instances.setTags, compute.instances.setServiceAccount | Project | Launch, inspect, and clean up CloudPilot-managed Compute Engine instances. |
| Node-creation dependencies | compute.disks.create, compute.subnetworks.use, compute.subnetworks.useExternalIp, iam.serviceAccounts.actAs | Project | Create boot disks, attach subnets, and let instances assume the configured service account. |
These permissions match the least-privilege custom role currently reconciled by the GKE Phase 2 install script.
On GKE, the installer grants roles/iam.workloadIdentityUser on the controller GSA to serviceAccount:<project-id>.svc.id.goog[cloudpilot/cloudpilot-admin]. The default Phase 2 flow annotates the in-cluster cloudpilot/cloudpilot-admin service account with iam.gke.io/gcp-service-account=<controller GSA>. If you provide CUSTOM_CONTROLLER_GSA, CloudPilot reuses that GSA instead of creating and deleting its own controller GSA.
Azure Workload Identity and Azure RBAC Permissions
For AKS, the current Phase 2 install flow requires OIDC issuer and Microsoft Entra Workload ID on the cluster. CloudPilot AI uses a user-assigned managed identity, creates a federated credential for the in-cluster cloudpilot-admin service account, and assigns Azure roles on both the AKS control-plane resource group and the managed node resource group.
| Scope | Required Role | Applied To | Purpose |
|---|---|---|---|
| AKS control-plane resource group | Contributor | /subscriptions/<subscription>/resourceGroups/<AZURE_RESOURCE_GROUP> | Read and manage AKS-side resources used by the optimizer. |
| AKS managed node resource group | Contributor | /subscriptions/<subscription>/resourceGroups/<AZURE_RESOURCE_GROUP_MC> | Create and manage CloudPilot-provisioned Azure resources for replacement nodes. |
| AKS managed node resource group | Network Contributor | /subscriptions/<subscription>/resourceGroups/<AZURE_RESOURCE_GROUP_MC> | Manage NIC, subnet, and related network resources for provisioned nodes. |
| AKS managed node resource group | Managed Identity Operator | /subscriptions/<subscription>/resourceGroups/<AZURE_RESOURCE_GROUP_MC> | Allow CloudPilot-managed nodes to attach and use the configured managed identity. |
These are the Azure roles assigned to the selected managed identity by the current AKS Phase 2 install script. The person running the installer separately needs permission to create the federated credential and these role assignments; those installation privileges are not granted to the optimizer identity by this table.
All cloud provider permissions align closely with those used by Karpenter, with the addition of access to scaling group APIs (AWS/AutoScalingGroup, AlibabaCloud/ScalingGroup) to enable existing node optimization.
Workload Autoscaler Additional Permissions
Installing Workload Autoscaler adds the permissions below to the Phase 1 installation. The controller, certificate Job, Node Agent, and metrics components use the Workload Autoscaler service account (workload-autoscaler in the default installation). Namespaced rules apply to the installation namespace, cloudpilot by default.
These permissions allow the components to:
- Read and patch workloads and related Pods, and evict or resize Pods when applying recommendations
- Read Jobs, CronJobs, Workload Groups, and rollout revisions, and update Workload Group status
- Manage autoscaling configuration, policies, and the Workload Activity timeline
- Maintain webhook registrations and the TLS Secret using locally generated certificates
- Discover metrics endpoints, collect metrics, and perform leader election
Workload Autoscaler Service Account
The following rules are installed by the Workload Autoscaler chart’s ClusterRole and namespaced Role.
| Scope | API Group | Resources | Verbs | Purpose |
|---|---|---|---|---|
| Cluster | "" (core) | pods, pods/eviction, pods/resize, pods/status | create, update, get, list, watch, patch, delete | Apply Pod recommendations, including eviction and in-place resize. |
| Cluster | "" (core) | pods/log | get | Read a bounded current or previous log tail to classify a detected Java OOM. |
| Cluster | apps | deployments, statefulsets, daemonsets, replicasets | get, update, list, watch, patch | Read and update workload objects affected by autoscaling. |
| Cluster | apps | controllerrevisions | list | Read DaemonSet rollout revisions when qualifying original JVM heap settings. |
| Cluster | autoscaling | horizontalpodautoscalers | get, list | Observe HPA resource protection without modifying HPAs. |
| Cluster | "" (core) | limitranges | get, list | Resolve namespace admission defaults before restoring HPA-protected resources. |
| Cluster | batch | jobs, cronjobs | get, list, watch | Resolve native batch workload ownership. |
| Cluster | evpa.cloudpilot.ai | workloadgroups | get, list, watch | Read workload selections and group configuration. |
| Cluster | evpa.cloudpilot.ai | workloadgroups/status | get, update, patch | Report Workload Group reconciliation status. |
| Cluster | evpa.cloudpilot.ai | autoscalingpolicyconfigurations, autoscalingpolicyconfigurations/status | create, update, get, list, watch, patch, delete | Manage autoscaling configuration and status. |
| Cluster | evpa.cloudpilot.ai | workloadautoscaleractivities | create, update, get, list, delete | Maintain the bounded, best-effort Workload Activity timeline. |
| Cluster | evpa.cloudpilot.ai | autoscalingpolicies, autoscalingpolicies/status, recommendationpolicies | get, update, list, watch, patch | Consume and update policy and recommendation resources. |
| Cluster | admissionregistration.k8s.io | mutatingwebhookconfigurations, validatingwebhookconfigurations | get, update, patch, list, watch | Maintain webhook configuration. |
| Cluster | discovery.k8s.io | endpointslices | get, list, watch | Discover metrics scrape endpoints. |
| Cluster | "" (core) | nodes, nodes/proxy, nodes/metrics, services, endpoints | get, list, watch | Discover node and service endpoints and collect metrics. |
| Cluster | networking.k8s.io | ingresses | get, list, watch | Discover ingress endpoints for metrics integrations. |
| Cluster | Non-resource URL | /metrics | get | Read the metrics endpoint. |
| Namespace | coordination.k8s.io | leases | get, create, update, watch | Perform leader election. |
| Namespace | "" (core) | configmaps (resourceNames: kube-root-ca.crt) | get, update | Access the namespace’s cluster CA ConfigMap. |
| Namespace | "" (core) | secrets (resourceNames: workload-autoscaler-webhook by default) | get, update, patch | Maintain the webhook TLS Secret. |
| Namespace | "" (core) | secrets | list, watch | Observe changes to TLS material. |
| Namespace | "" (core) | events | create, patch | Publish Kubernetes Events. |
| Namespace Role | security.openshift.io | securitycontextconstraints (resourceNames: privileged) | use | Support Node Agent execution on OpenShift. This rule is present in the chart’s Role on all providers and only applies where the OpenShift API is available. |
The certificate generator self-signs certificates locally and updates the webhook registrations and TLS Secret. The Workload Autoscaler service account does not require CertificateSigningRequest creation or approval, or signer approval permissions.
For Java OOM classification, Workload Autoscaler requests at most 200 timestamped lines from the selected current or previous log. Each occurrence-level classification attempt has a two-second timeout and a 256 KiB response limit; only the last 10 KiB is retained briefly in memory for parsing. The dedicated client is rate-limited to QPS 2 with burst 12. Only the resulting subtype is stored or uploaded by Audit Log—raw container logs are not.
The HPA and LimitRange permissions are read-only and do not add watch or write access. The Phase 1 Agent does not receive these permissions. If required namespace defaults cannot be read reliably, restoration waits; the raw template baseline and recommendation provenance remain unchanged.
Additional Agent Permissions with Workload Autoscaler
The Workload Autoscaler chart separately binds the existing cloudpilot-agent service account in the installation namespace to the following cluster-wide rules. These permissions are added when Workload Autoscaler is installed; they are separate from the Workload Autoscaler service account above.
| Scope | API Group | Resources | Verbs | Purpose |
|---|---|---|---|---|
| Cluster | batch | jobs, cronjobs | get, list, watch | Discover batch workloads for synchronization and reporting. |
| Cluster | batch | cronjobs | patch | Synchronize the proactive-update label on CronJobs. |
| Cluster | evpa.cloudpilot.ai | workloadgroups | create, update, patch, delete, get, list, watch | Synchronize Workload Group definitions. |
| Cluster | apps | deployments, daemonsets, statefulsets | get, list, watch, patch | Read workload configuration and synchronize optimization labels. |
| Cluster | evpa.cloudpilot.ai | workloadautoscaleractivities | get, list, watch | Collect the Workload Activity timeline for reporting. |
Required Binding
The chart creates a ClusterRoleBinding and a namespaced RoleBinding for the Workload Autoscaler service account, plus a separate ClusterRoleBinding for cloudpilot-agent. Both service accounts reside in the installation namespace. The webhook Secret name follows the chart’s full release name with the -webhook suffix; the default installer uses workload-autoscaler-webhook.
Commitment to Privacy and Security
CloudPilot AI adheres strictly to data protection best practices. We access only the data necessary for operation and optimization, and we never request or retain unnecessary privileges.
By design, we reduce attack surfaces through scoped permissions and continuous review of access requirements. This ensures secure, compliant, and efficient operation within your environment.
For further assistance, feel free to reach out to us through our Slack channel .
