Harvester HCI · VMs and Kubernetes on bare metal
Virtual machines and Kubernetes, on one Harvester HCI cluster.
Harvester is SUSE's open-source hyperconverged infrastructure. It turns a handful of bare-metal servers into one pool of compute, storage and networking, and runs virtual machines and Kubernetes side by side on it. It's what Baytech Cloud, the platform that hosts our clients' systems, is built on. We can build and run a cluster for you, or host you on ours.
Watch the cluster
How Harvester keeps running when a server doesn't
Every server in a Harvester cluster is compute, storage and network at once, and every disk is kept as copies on several servers. So a server can be taken out for an upgrade, or lost without warning, and the work carries on. Here's what that looks like on a four-server cluster like ours.
running
disk copies
running
disk copies
running
disk copies
running
disk copies
trafficflowing
- databases3/3
- file shares3/3
- vm disks3/3
- app data3/3
planned
Upgrading a server without stopping its VMs
Harvester upgrades one server at a time. Before a server restarts, its virtual machines move to the others while they keep running, so the people using them don't notice.
- ✓Server 2 goes into maintenance mode
- ✓Its VMs live-migrate to the other servers, still running
- ✓The server upgrades and restarts
- ✓It rejoins and brings its disk copies up to date
unplanned
Losing a server without warning
When a server fails, its disk copies go with it, but every disk still has copies on other servers. Harvester restarts the lost server's workloads on the healthy ones, and Longhorn rebuilds the missing copies elsewhere.
- ✓Server 3 stops responding
- ✓Every disk keeps working from its surviving copies
- ✓Its workloads restart on the healthy servers
- ✓Longhorn rebuilds the missing copies; the repaired server rejoins
An illustration of how Harvester and Longhorn handle maintenance and failure, on four servers like Baytech Cloud. Services that run more than one copy keep answering throughout; a VM on a lost server is restarted, not moved.
What Harvester HCI GivesYou
Harvester is built from open-source parts already proven in production: Linux, KVM, Kubernetes, KubeVirt and Longhorn.
Virtual Machines and Containers Together
Windows and Linux virtual machines run next to Kubernetes workloads on the same servers, managed through the same Kubernetes API. KubeVirt runs each VM on KVM, the hypervisor built into Linux.
Storage Built In
Longhorn keeps every disk as copies on several servers, so there's no separate SAN or storage array to buy, cable and look after.
Live Migration
A running VM can move from one server to another for maintenance or an upgrade without being shut down.
Managed with Rancher
Harvester plugs into Rancher, so virtual machines and every Kubernetes cluster are managed from one place, in your data center, at the edge or in the cloud.
Open Source, No License Fees
Harvester is open source. There are no per-socket or per-VM hypervisor licenses, and no vendor lock-in.
Snapshots and Backups
VM snapshots and backups to S3-compatible storage or NFS are part of the platform, so recovery doesn't depend on another product.
Bringing Harvester to you
From separate servers, storage and hypervisors to one cluster
Many teams run virtual machines on one platform, storage on another and containers somewhere else, each with its own licenses and its own specialist. Here's how we move that onto Harvester, or onto Baytech Cloud if you'd rather not run servers at all.
- 01assess
Map what you run today
Every VM, database and application: what it needs, what it costs, and what depends on what.
- 02design
Size the cluster
At least three servers for high availability, with the network and storage laid out for your workloads.
- 03build
Build it as code
Virtual machines are defined in Terraform through Harvester's provider, and Kubernetes apps ship through Argo CD, so the whole setup can be rebuilt from a repository.
- 04move
Move the workloads
VMs are imported and tested before cut-over, and applications move into containers where that makes sense.
- 05protect
Back it up offsite
VM backups go to S3-compatible object storage in another location, and databases get their own schedule on top.
- 06run
Run it, or hand it over
We monitor it, keep it upgraded and answer when something needs attention, or train your team to do it.
the Harvester release Baytech Cloud runs today, upgraded in place from 1.0
Kubernetes clusters on it (development, test, production and GitOps), managed by Rancher
applications deployed to it through Argo CD
client systems we host on it
Figures from Baytech Cloud, our own Harvester cluster.
Baytech Cloud, Built onHarvester
The platform that hosts our clients' systems runs on the same stack we'd build for you.
Bare-Metal Servers
Four dedicated servers in a US data center form one Harvester cluster: 96 CPU cores and 1.5 TB of memory in one pool, on NVMe storage.
VMs Next to Kubernetes
Windows servers, SQL Server, PostgreSQL, Sentry, monitoring and build agents run as virtual machines, alongside the Kubernetes clusters that run our clients' applications.
Everything as Code
Terraform builds the virtual machines through Harvester's provider, and Argo CD keeps every Kubernetes app in step with its Git repository.
Watched Around the Clock
Zabbix, Uptime Kuma, Grafana and Sentry watch the servers, the systems and the errors, and alert our engineers.
Backed Up Offsite
VM backups go to S3-compatible object storage in another location. PostgreSQL also gets weekly full, daily and hourly backups.
Upgraded in Place
We've taken the cluster from Harvester 1.0 to 1.7, backing up Rancher before each upgrade.
Running on Harvester today
Petra Medical College's student, staff and partner portals run on Kubernetes on our Harvester cloud, released through Rancher and Argo CD. We built the platform in five months, after another firm had spent two years on it, and we still host it.
Baytech Consulting is significantly more focused than anyone else I’ve worked with.
The Rest of theStack
Harvester is one part of an open-source stack we use every day, from the servers to the release pipeline.
Rancher
Open-source Kubernetes management. Rancher gives us one place to deploy, monitor and secure clusters across clouds and on-premises servers, with access policies built in, and it makes Kubernetes manageable for teams without deep container experience.
Kubernetes
The standard for running containers. It automates deployment, scaling and recovery, runs the same on-premises, in the cloud or in between, and makes efficient use of the servers underneath. In Harvester, its API manages virtual machines too.
Argo CD
GitOps continuous delivery for Kubernetes. Git is the source of truth, so every change is version controlled, traceable and auditable, deployments are automatic and repeatable, and a bad release rolls back quickly.
Longhorn
Distributed block storage for Kubernetes. Each volume is kept as copies on separate servers and rebuilt automatically when one is lost, with snapshots and offsite backups.
KubeVirt
The part of Harvester that runs virtual machines as Kubernetes workloads, so VMs and containers share one platform and one way of working.
Linux and KVM
Harvester runs on an enterprise Linux base and uses KVM, the kernel-based virtual machine hypervisor that Linux has shipped for years, rather than a proprietary hypervisor.
Thinking about Harvester?
Tell us what you run today and where. We'll tell you plainly whether Harvester fits, what the cluster would look like, and whether it belongs in your own building or on Baytech Cloud.
Talk to an engineerCompany Facts
& Figures
Our Office
Get in Touch
Software is what you depend on to get things done. Baytech is who you depend on for software.
