Skip to content

Latest commit

 

History

History
157 lines (137 loc) · 9.75 KB

README.md

File metadata and controls

157 lines (137 loc) · 9.75 KB

openshift-data-foundations

Version: 0.3.5 Type: application AppVersion: 1.16.0

A Helm chart to install ODF on Openshift

Notable changes

  • v0.3.5: Add a status: true to podFailure specifications for <= 4.20 compatibility. The field is mandatory in <= 4.20, and still accepted in later versions.

  • v0.3.4: Add a job-wait-for-mcp Job (sync-wave -2, before the node-labeling Job at -1 and the StorageCluster) that waits for the configured MachineConfigPools (default: worker) to finish reconciling before any nodes are labeled and the StorageCluster is created. Labeling nodes for ODF can trigger an MCO rolling reboot, and forming Ceph monitors mid-rollout can strand monitor PVCs in the wrong availability zone. Controlled by waitForMcp.enabled (default true), waitForMcp.pools, and waitForMcp.timeout. All Jobs (and the node-labeling CronJob) are now drain-cancelable: they trap SIGTERM and exit nonzero so a cancelled run is never marked successful, and podFailurePolicy ignores drain evictions (DisruptionTarget) so a replacement pod resumes the wait instead of burning backoffLimit. Job grace periods reduced 400s to 60s.

  • v0.3.3: Only emit preparePlacement when overridden. Otherwise the application may go out-of-sync on ArgoCD.

  • v0.3.2: Make preparePlacement configurable.

  • v0.3.1: Make objectStorage.enable: false cover more resources.

  • v0.3.0: Replace global.datacenter.storageClassName with odf.osd.pvc.storageClassName. When unset, the OSD storage class is derived from global.clusterPlatform (AWS: gp3-csi, Azure: managed-csi, GCP: standard-csi). This is a backwards-incompatible change. Also add a CronJob that periodically re-runs the same node-labeling logic as the bootstrap Job (same useSpecificNodes / selector conditions).

  • v0.2.3: Allow passing of label selector to labelling job to avoid, for example, labelling submariner nodes. Also ensure at least 3 nodes are labelled when using label selector.

  • v0.2.2: Reconfigure sync ordering to ensure label job runs before attempts to create storagecluster/storagesystem

  • v0.2.1: Introduce boolean to configure where mirroring is enabled. Needed for RamenDR. Defaults to false

Homepage: https://github.com/validatedpatterns/openshift-data-foundations-chart

Maintainers

Name Email Url
Validated Patterns Team validatedpatterns@googlegroups.com

Notes

This branch currently tracks the v0.3.x releases which use the host as a default failure domain for objectStorage.

Values

Key Type Default Description
global.clusterPlatform string "" OpenShift cluster platform (AWS, Azure, GCP). Used to select the default OSD storage class when odf.osd.pvc.storageClassName is empty.
job.failedJobsHistoryLimit int 1 failedJobsHistoryLimit for the label-storage-nodes CronJob.
job.image string "image-registry.openshift-image-registry.svc:5000/openshift/cli:latest"
job.schedule string "*/15 * * * *" Cron schedule for re-labeling storage nodes after the initial Job (UTC).
job.successfulJobsHistoryLimit int 3 successfulJobsHistoryLimit for the label-storage-nodes CronJob.
objectStorage.dataPool.failureDomain string "host" Failuredomain for the dataPool
objectStorage.dataPool.replicas int 3
objectStorage.enable bool true Deploy object storage (Ceph RGW, NooBaa MCG, and related StorageClasses/Routes). When false, NooBaa reconciliation is disabled.
objectStorage.gateway.instances int 2
objectStorage.metadataPool.failureDomain string "host" Failuredomain for the metadataPool
objectStorage.resources.limits.cpu string "2"
objectStorage.resources.limits.memory string "6Gi"
objectStorage.resources.requests.cpu string "1"
objectStorage.resources.requests.memory string "4Gi"
odf.enable_mirroring bool false
odf.externalUrl string "https://s3-rgw-openshift-storage"
odf.mds.requests.cpu int 3
odf.mds.requests.memory string "8Gi"
odf.mgr.requests.cpu int 1
odf.mgr.requests.memory string "3Gi"
odf.mon.requests.cpu int 1
odf.mon.requests.memory string "2Gi"
odf.namespace string "openshift-storage"
odf.noobaacore.requests.cpu int 1
odf.noobaacore.requests.memory string "4Gi"
odf.noobaadb.requests.cpu int 1
odf.noobaadb.requests.memory string "4Gi"
odf.osd.pvc.storage string "2Ti"
odf.osd.pvc.storageClassName string "" Storage class for ODF OSD volumes. Empty selects a platform default from global.clusterPlatform (AWS: gp3-csi, Azure: managed-csi, GCP: standard-csi).
odf.osd.requests.cpu int 2
odf.osd.requests.memory string "5Gi"
odf.preparePlacement object {} Optional preparePlacement for OSD device sets. When set, overrides the default empty preparePlacement on the StorageCluster storageDeviceSets, allowing topology spread constraints or affinity rules to control OSD prepare pod scheduling.
odf.serviceUrl string "http://rook-ceph-rgw-ocs-storagecluster-cephobjectstore.openshift-storage.svc.cluster.local"
odf.storageClass.name string "ocs-storagecluster-ceph-rgw"
odf.storageClass.objectStoreName string "ocs-storagecluster-cephobjectstore"
rbac.roleBindings[0].createBinding bool true
rbac.roleBindings[0].name string "label-storage-nodes"
rbac.roleBindings[0].roleRef.kind string "ClusterRole"
rbac.roleBindings[0].roleRef.name string "label-storage-nodes"
rbac.roleBindings[0].scope.cluster bool true
rbac.roleBindings[0].scope.namespace string ""
rbac.roleBindings[0].subjects.apiGroup string ""
rbac.roleBindings[0].subjects.kind string "ServiceAccount"
rbac.roleBindings[0].subjects.name string "odf-node-label-sa"
rbac.roleBindings[0].subjects.namespace string "openshift-storage"
rbac.roleBindings[1].createBinding bool true
rbac.roleBindings[1].name string "wait-for-mcp"
rbac.roleBindings[1].roleRef.kind string "ClusterRole"
rbac.roleBindings[1].roleRef.name string "wait-for-mcp"
rbac.roleBindings[1].scope.cluster bool true
rbac.roleBindings[1].scope.namespace string ""
rbac.roleBindings[1].subjects.apiGroup string ""
rbac.roleBindings[1].subjects.kind string "ServiceAccount"
rbac.roleBindings[1].subjects.name string "odf-node-label-sa"
rbac.roleBindings[1].subjects.namespace string "openshift-storage"
rbac.roles[0].apiGroups[0] string "\"\""
rbac.roles[0].createRole bool true
rbac.roles[0].name string "label-storage-nodes"
rbac.roles[0].resources[0] string "nodes"
rbac.roles[0].scope.cluster bool true
rbac.roles[0].verbs[0] string "get"
rbac.roles[0].verbs[1] string "list"
rbac.roles[0].verbs[2] string "patch"
rbac.roles[0].verbs[3] string "update"
rbac.roles[1].apiGroups[0] string "machineconfiguration.openshift.io"
rbac.roles[1].createRole bool true
rbac.roles[1].name string "wait-for-mcp"
rbac.roles[1].resources[0] string "machineconfigpools"
rbac.roles[1].scope.cluster bool true
rbac.roles[1].verbs[0] string "get"
rbac.roles[1].verbs[1] string "list"
rbac.roles[1].verbs[2] string "watch"
route.name string "s3-rgw"
route.port.targetPort string "http"
route.service.name string "rook-ceph-rgw-ocs-storagecluster-cephobjectstore"
route.service.weight int 100
serviceAccountName string "odf-node-label-sa"
storageSystem.deploy bool false
storageSystem.inventory.nodeJobLabelSelector string "node-role.kubernetes.io/worker="
storageSystem.inventory.nodes[0] string "nodeA"
storageSystem.inventory.nodes[1] string "nodeB"
storageSystem.inventory.nodes[2] string "nodeC"
storageSystem.inventory.useSpecificNodes bool true
storageSystem.name string "ocs-storagecluster-storagesystem"
storageSystem.namespace string "openshift-storage"
storageSystem.spec.kind string "storagecluster.ocs.openshift.io/v1"
storageSystem.spec.name string "ocs-storagecluster"
storageSystem.spec.namespace string ""
waitForMcp object {"enabled":true,"pools":["worker"],"timeout":2700} Wait for MachineConfigPools to finish reconciling (Updated=True, Degraded=False) before the StorageCluster is created. Labeling nodes for ODF can trigger an MCO rolling reboot; forming Ceph monitors mid-rollout can strand monitor PVCs in the wrong availability zone.
waitForMcp.pools list ["worker"] MachineConfigPools to wait on before creating the StorageCluster.
waitForMcp.timeout int 2700 Per-pool wait timeout in seconds (45m covers a 3-node rolling reboot).

Autogenerated from chart metadata using helm-docs v1.14.2