Harvester, an open-source hyperconverged infrastructure (HCI) software built using Kubernetes, recently announced its v0.2.0 release. Harvester can be used to implement HCI on bare metal servers and is an open source alternative to vSphere and Nutanix. The first release, 0.1.0, was made public earlier this year.
In his blog post, Sheng Yang, senior engineering manager at SUSE, explained the new features in Harvester v0.2.0 and announced its beta availability. These include raw block device support, VM Live migration support, and Rancher Integration.
The below diagram provides the high-level architecture of Harvester.
Source: https://docs.harvesterhci.io/v0.2/
To minimize load on the Harvester management console, removing MinIO dependency, virtual machines (VMs) are now provided with raw block devices. Leveraging the Backing Image feature in Longhorn v1.1.1, the new release allows for instantaneous image import for the users.
VM live migration is now supported in Harvester, enabling VM to migrate from one node to another for performing maintenance work. Yang recommends using the VLAN network, providing further support documentation.
Harvester v0.2.0 adds VM backup support, providing a method for backing up VM images outside of the cluster. One can store backups of VM volumes by creating an S3 compatible endpoint or NFS server. Detailed instructions for VM backup and restore are available here.
Providing ease for populating bare-metal nodes with required operating systems, PXE boot installation support is now available in Harvester. There are examples available here for using iPXE based automatic installation of Harvester on various platforms.
For Kubernetes management, Harvester now comes with a built-in Rancher Server. Calling it the most requested feature, Yang said that engineers could use the built-in Rancher server to create Kubernetes clusters on top of the Harvester bare-metal clusters. Rancher and Harvester share the authentication process; one can follow the instructions here for Rancher integration.
Yang also mentioned that the work is in progress for the snapshot feature for VMs, to revert the state of VM to a previous snapshot. No data will be copied outside the cluster for a snapshot.
Harvester is available at GitHub, and readers can join the conversation on the #harvester Slack channel.