SUSE Harvester - First Impressions

Suse Harvester First Impressions

0:00

/656.80551

1×

A sudden failure occurred recently in my home lab (admittedly foreseeable and of my own making) which opened my availability to try a new hypervisor. For some context I have been a longtime user of XCP-NG for my virtualization needs and prior to that ran purely bare metal in the realm of ancient and slow hardware. I have dabbled into other areas such as Proxmox, VMware, and libvirt manager, but always returned back to XCP-NG with Xen Orchestra for its simple, well documented, and high quality implementation for high availability clustered systems.

This has of course come with drawbacks as no one in the hypervisor space has ever taken convenient feature management and pleasant UI design to be be anything but mutually exclusive. As such, when I was presented with SUSE's Harvester, Longhorn, and Rancher ecosystem I was pleasantly intrigued. The UI looks nice, the responsiveness is tolerable, and the promise, while baffling to anyone but the developers was worthwhile.

As such I've taken the leap into Harvester and have discovered what is most certain to be my new choice for a hypervisor for at least the next few years. Understanding why this may be the right choice for you as well will depend on the following circumstances:

The Impossible Explanation

SUSE has sent more emails, and hosted more conferences on Harvester, Rancher, and Longhorn, than I believe should ever be necessary given the volume of users currently fleeing from Red Hat and VMware. The majority of these informative emails are inviting users to experience a demonstration of this incredible new software. The demonstration of this admitted boring class of software product is necessary, because even when prompted they cannot give a clear explanation of what their product is doing differently besides spouting techno-babble and proclaiming AI Kubernetes.

There is nothing wrong with this exclamation as it is a wonderful ecosystem for managing Kubernetes, but it is rather useless in helping the majority of users running home labs or assisting small companies still scared of containers. These people (often myself included) are left entirely in the dark about this class of product and justifiably resign ourselves to neither needing Kubernetes, nor wanting anything to do with a system proclaiming to be run by it. With that said, SUSE I'll do you job quickly and provide you a meaningful explanation for us lowly administrators of the world.

Harvester is a Type 1 KVM based hypervisor running on SUSE Linux Enterprise Server Micro, with Longhorn as its storage manager, that uses a Kubernetes layer to provide advanced features such as:

1. Two Node High Availability Clustering
2. Easy Internal Networking, as well as VLAN integrations
3. Kubernetes tagging (json based tags)
4. Very easy backup management
5. Cloud Init/Cloud Config
6. Easy Migration Between clusters and machines
7. Complete compatibility and integration with Longhorn and Rancher for integrating Kubernetes and redundant storage

This means for a small virtualized workload it can be relatively trivial to go from a 5 or 6 machine layout with 2-3 SAN servers (cephfs, etc), and 3 nodes for a high availability cluster, to a 4 machine layout with 2 compute nodes on Harvester, and 2 storage nodes on Longhorn. With this design even being recommended on the home page of each projects website.

The Documentation

As with any project related to SUSE, the documentation of Harvester is a disjointed constant mix of technical brilliance and nonsensical or outdated confusion. That being said the necessary information to operate Harvester well does exist and the docs page at docs.harvesterhci.io is technically excellent and fairly complete, and with enough patience you can get fairly proficient just by reading the documentation.

I'd like to say that is enough to build your business or home lab around, but I know that most users need substantially more information to assist with understanding and recommended configurations that SUSE projects as a general rule are adverse to providing to their users. Additionally the documentation is strictly to the point without much care given to concepts or practical implementation. This lack of practical implementation documentation goes so far as to make asking for a recommended minimum deployment at a SUSE event for the product, lead to politician levels of talking in circles non answers.

So should this dissuade you from setting up Harvester? If you struggle to RTFM or even just prefer to have good examples and practical implementation advice, then yes choosing Harvester will probably hurt you. If you are confident in reading very specific documentation, and searching for 10 articles for every feature you want to use, then no Harvester will serve you just fine.

Features Forever

By mixing Kubernetes with KVM Harvester unlocks tooling that is absolutely incredible in terms of value to dollar, and time to effort. This is the greatest strength of SUSE related products in every area they have touched. For example the value of YAST to system administrators is incalculable, the value of Uyuni-Project after the loss of Spacewalk has been nothing short of a remarkable improvement, and the clean integration of Longhorn, Harvester, and Rancher together are truly capable of changing our interaction with virtualization as a whole. To make this clear I'll list my personal favorite features from Harvester.

1. Cloud Config - Setting up VMs even from templates can be a tremendous time waster. Easy and directly integrated cloud config makes my life easier even in setting up a few simple DNS servers. After having this immediately on hand and built in I cannot picture myself going back to a hypervisor without this built in.
2. SSH Key injection to the VM - Being able to have secrets on the hypervisor level and integrate them for things like setting ssh keys, or even partnering with cloud config to auto setup salt configurations with Uyuni is invaluable. These types of steps are often overlooked because they take so much time, and integrating it like this makes running things well an easier process.
3. Simple Backup - VEAM sells a nice backup product for Harvester that I will certainly investigate in the future, but while I have certainly bemoaned SUSE's over complication in documentation their approach to backups is first rate, because it is simple. Just attach an NFS or S3 bucket in the Settings, and then backup your VMs to that storage. For home or small business use, imagine replacing a few levels of storage systems with "Here's my backup bucket, it costs $1 per Terabyte per month."
4. PCI Pass Though/GPU - Passing through PCI and vGPU pass through is a fickle thing even in mature and expensive systems. XCP-NG is making this seamless in another 6-18 months, and Proxmox has some interesting (but admittedly questionable) tooling. However the pass through in Harvester is well developed with PCI devices I barely know exist showing up as pass through devices, and SR-IOV GPU and vGPU having logical separation for ease of management.
5. It works like it should mentality - Hypervisors are a tool that demands a good UI to be effective, and yet users are constantly driven back to the command line to get their work done. In Harvester the developers place extreme focus on not having to use the cli pretty much at all, and this ensures that when I hand off a project on Harvester to another technician they will be able to see the configuration and not need to know the double secret handshake for setting up another node (cough... cough... Proxmox's infinite configurability, cough... cough... XCP-NG device pass through).
6. Web VNC & Serial Console - As an enthusiast of passwords that extend past the AI token limit I can truthfully say that remote console without ssh will more than likely save my home lab more than once in the next year. I know its available from other vendors, but the integration in Harvester strikes me as pleasantly appealing and clinically useful. This is something that has driven me mad at times in the XCP-NG and Xen Orchestra web ui for years.

The Decision

So, should you? Shouldn't you? Could you? Would you? Really, yes you should. Harvester is ahead of the curve in many ways that provide real world tangible benefits to users, and I don't see other hypervisors joining in anytime soon. It's not a perfect solution and I can certainly complain about documentation woes until the day I die, but as with almost anything with the SUSE name attached to it, it works, it works well, and it does what you expect of it.

So if you're in the market for a new hypervisor, it is worth the time to take a look at Harvester, and if you're a hopeful Kubernetes enthusiast like me, then it really is the best option you could ask for.