changelogs/CHANGELOG-1.15.md
https://github.com/vmware-tanzu/velero/releases/tag/v1.15.0
velero/velero:v1.15.0
https://velero.io/docs/v1.15/upgrade-to-1.15/
Data transfer activities for CSI Snapshot Data Movement are moved from node-agent pods to dedicate backupPods or restorePods. This brings many benefits such as:
Item Block concepts are introduced for resource backups to help to achieve multiple thread backups. Specifically, correlated resources are categorized in the same item block and item blocks could be processed concurrently in multiple threads.
ItemBlockAction plugin is introduced to help Velero to categorize resources into item blocks. At present, Velero provides built-in IBAs for pods and PVCs and Velero also supports customized IBAs for any resources.
In v1.15, Velero doesn't support multiple thread process of item blocks though item block concepts and IBA plugins are fully supported. The multiple thread support will be delivered in future releases.
For more information, check the design https://github.com/vmware-tanzu/velero/blob/main/design/backup-performance-improvements.md.
Repository maintenance are resource consuming tasks, Velero now allows you to configure the nodes to run repository maintenance jobs, so that you can run repository maintenance jobs in idle nodes or avoid them to run in nodes hosting critical workloads.
To support the configuration, a new repository maintenance configuration configMap is introduced.
For more information, check the document https://velero.io/docs/v1.15/repository-maintenance/.
In 1.15, Velero allows you to configure the data mover backupPods to read-only mount the backupPVCs. In this way, the data mover expose process could be significantly accelerated for some storages (i.e., ceph).
To support the configuration, a new backup PVC configuration configMap is introduced.
For more information, check the document https://velero.io/docs/v1.15/data-movement-backup-pvc-configuration/.
In 1.15, Velero allows you to configure the storageclass used by the data mover backupPods. In this way, the provision of backupPVCs don't need to adhere to the same pattern as workload PVCs, e.g., for a backupPVC, it only needs one replica, whereas, the a workload PVC may have multiple replicas.
To support the configuration, the same backup PVC configuration configMap is used.
For more information, check the document https://velero.io/docs/v1.15/data-movement-backup-pvc-configuration/.
The backup repository may need to cache data on the client side during various repository operations, i.e., read, write, maintenance, etc. The cache consumes the root file system space of the pod where the repository access happens.
In 1.15, Velero allows you to configure the total size of the cache per repository. In this way, if your pod doesn't have enough space in its root file system, the pod won't be evicted due to running out of ephemeral storage.
To support the configuration, a new backup repository configuration configMap is introduced.
For more information, check the document https://velero.io/docs/v1.15/backup-repository-configuration/.
In 1.15, several performance related issues/enhancements are included, which makes significant performance improvements in specific scenarios:
client-burst/client-qps parameters are automatically inherited to plugins, so that you can use the same velero server parameters to accelerate the plugin executions when large number of API server calls happen, see issue https://github.com/vmware-tanzu/velero/issues/7806Golang runtime: v1.22.8
kopia: v0.17.0
Due to an issue of Kubernetes upstream, if a volume is mounted as read-only in SELinux environments, the read privilege is not granted to any user, as a result, the data mover backup will fail. On the other hand, the backupPVC must be mounted as read-only in order to accelerate the data mover expose process.
Therefore, a user option is added in the same backup PVC configuration configMap, once the option is enabled, the backupPod container will run as a super privileged container and disable SELinux access control. If you have concern in this super privileged container or you have configured pod security admissions and don't allow super privileged containers, you will not be able to use this read-only backupPVC feature and lose the benefit to accelerate the data mover expose process.
Restic path for fs-backup is in deprecation process starting from 1.15. According to Velero deprecation policy, for 1.15, if Restic path is used the backup/restore of fs-backup still creates and succeeds, but you will see warnings in below scenarios:
--uploader-type=restic is used in Velero installationPreviously, a fixed name is searched for node-agent configuration configMap. Now in 1.15, Velero allows you to customize the name of the configMap, on the other hand, the name must be specified by node-agent server parameter node-agent-configmap.
In 1.15, below Velero server parameters for repository maintenance jobs are moved to the repository maintenance job configuration configMap. While for back compatibility reason, the same Velero sever parameters are preserved as is. But the configMap is recommended and the same values in the configMap take preference if they exist in both places:
--keep-latest-maintenance-jobs
--maintenance-job-cpu-request
--maintenance-job-mem-request
--maintenance-job-cpu-limit
--maintenance-job-mem-limit
In 1.15, the Changing PVC selected-node feature enters deprecation process and will be removed in future releases according to Velero deprecation policy. Usage of this feature for any purpose is not recommended.