Stormux CommunitiesSanity Check. Docker vs Incus (LXD)
submitted by lal309
My apologies for the long post.
I have a single server running Unraid with about 12 services (Pihole, Wordpress, Heimdall, Jellyfin, etc.) all running on Docker. This server is also acting as my home lab NAS. Everything runs fine for my use case (at least for right now) but I’ve been thinking about creating some type of compute cluster for my services instead of a single server.
Recently, I saw a discussion about Proxmox, Docker, LXD and Incus where a user felt that Incus was a better option to all the others. Curious, I started reading up on Incus and playing around with it and contemplating switching all my services from Docker in Unraid to an Incus cluster (I’m thinking around 3 nodes) and leaving the Unraid server to serve as a NAS only.
In a nutshell Incus/LXD appear to be (to me) a lightweight version of a VM in which case I would have to manually install and configure each service I have running. Based on the services I have running, that seems like a ton of work to switch to Incus when I could just do 3 physical servers (Debian) in docker swarm mode or a Proxmox cluster with 3 Debian VMs with docker in swarm mode. I’d all possible, I would like to keep my services containerized rather then actual VMs.
What has me thinking that a switch to Incus may be worth it is performance. If the performance of my services is significantly better in Incus/LXDs as compared to Docker, then that’s worth it to me. I have not been able to find any type of performance comparison between Incus/LXD and Docker. I don’t know if there are other reasons as to “Incus over Proxmox and Docker” which is why I’m asking the greater community.
Here’s my question:
Based on your experience and taking into consideration my use case (home lab/home use), do the pros and cons of Incus outweigh accomplishing my goal by creating standalone hosts cluster or Proxmox cluster?
Docker has native compute performance. The processes essentially run on the host kernel with a different set of libs. The only notable overhead is in storing and loading those libs which takes a bit more disk and RAM. This will be true for any container solution and VMs. VMs have a lot of additional overhead. An a cursory glance, Incus seems to provide an interface to run Linux containers or VMs. I wouldn't expect performance differences between containers run through it compared to Docker.
Thank you I appreciate your input!
The overhead is minimal, KVM VMs have near-native performance (type 1 hypervisor). There is some memory overhead as each VM runs its own kernel, but a lot of this is cancelled by KSM [1] which is a memory de-duplication mechanism.
Each VM runs its own system services (think systemd, logging, etc) so there is some memory/disk usage overhead there - but it would be the same with Incus/LXC as they do the same thing (they only share the same kernel).
https://serverfault.com/questions/225719/so-really-what-is-the-overhead-of-virtualization-and-when-should-i-be-concerned
I usually go for bare-metal > on top of that, multiple VMs separated by context (think "tenant", production/testing, public/confidential/secret, etc. VMs provide strong isolation which containers do not. At the very minimum it's good to have at least separate VMs for "serious business" and "lab" contexts) > applications running inside the VMs (containerized or not - service/application isolation through namespaces/systemd has come a long way, see man systemd-analyze security) - for me the benefit of containerization is mostly ease of deployment and... ahem running inscrutable binary images with out-of-date dependencies made by strangers on the Internet)
If you go for a containerization solution on top of your VMs, I suggest looking into podman as a replacement for Docker (less bugs, less attack surface, no single-point-of-failure in the form of a 1-million-lines-of-code daemon running as root, more unix-y, better integration with systemd [2]. But be aware of the maintenance overhead caused by containerization, if you're serious about it you will probably end up maintaining your own images)
No. It isn't going to perform any better then docker.
In theory Incus and LXD by default will be slightly heavier than docker; they run a a lot more bare-metal services (ex. systemd) in container giving them more flexability and a VM like feel, which would 99% of the time be wasted resources in a docker container
They also dont have nearly as much 'out of the box' support as Docker/Podman might, especially for single process containers.
That being said docker used to run on lxc until not too long ago, so there's still many similarities between the 2
Interesting tidbit about the performance. It has been a bit of challenge getting “up to speed” with Incus/LXD from a guide and walkthrough perspective. Although I do find their documentation pretty well organized and useful.
Your workload (a NAS and a handful of services) is going to be a very familiar one to members of the community, so you should get some great answers.
My (I guess slightly wacky) solution for this sort of workload has ended up being a single Docker container inside an LXC container for each service on Proxmox. Docker for ease of management with compose and separate LXCs for each service for ease of snapshots/backups.
Obviously there's some overhead, but it doesn't seem to be significant.
On the subject of clustering, I actually purchased three machines to do this, but have ended up abandoning that idea - I can move a service (or restore it from a snapshot to a different machine) in a couple of minutes which provides all the redundancy I need for a home service. Now I keep the three machines as a production server, a backup (that I swap over to for a week or so every month or two) and a development machine. The NAS is separate to these.
I love Proxmox, but most times it get mentioned here people pop up to boost Incus/LXD so that's something I'd like to investigate, but my skills (and Ansible playbooks) are currently built around Proxmox so I've got a bit on inertia.
Incus is a weird name lol.
But jokes aside, I think Docker and Podman have more adaption?
You are probably right. Judging by their GitHub repo, their first release was in October of 2023. If I understand correctly, Incus is a fork of Canonical LXDs which is not so new??? Idk. Their documentation is quite good but there aren’t a lot of “guides” out there so yea.
You can use LXD guides, it is the same just not under canonical control and with dome new features coming out (I think).
Very true! Thanks.
I am in no way even slightly an "expert" here, but Incus could be considered a lightweight Proxmox..
They're both going to run some VMs and / or containers, but with Proxmox you get the overhead of the fancy GUI.
So if your host(s) aren't running guests at >90% load, then there won't be any difference in performance.
I've recently installed Proxmox because everyone else uses it (and VMware's free ESXi is now dead)... but after pulling my hair out trying to get some things done, I'm seriously looking to move to Incus.
There's another post here somewhere (the one about free ESXi being killed off) with someone explaining more about Incus, which seems like it's the way to go... maybe worth a search.
I believe you are referencing the same post that got my curious about Incus and started playing around with it.
My biggest gripe is the manual installation of all services which I will do if it’s worth it. So far not sure that it is, hence the post to get more opinions.
There’s is a GUI you can install for Incus but it’s optional and not preinstalled.
I appreciate your input.
You should also consider the time you will spend configuring and setting up everything in Incus.
If you do this with educational purpose go for it, otherwise I will advise you not to, as Proxmox has a wider support and probably finding information, etc. for it is going to be easier.
Alternatively, why don't you dedicate one of the hosts to Incus and play around with it and decide if it works for you or not.
Fair point. I’m most familiar with docker and proxmox. Sorta doing it for educational purposes but I also have critical services (critical to me) running that must be available.
I think it is a good way to isolate docker containers from the host without the heavier performance increase of a full VM. Each container can be easily set to an IP address, though the same is probably true for docker idk.
Unrelated, Podman is the a unprivileged implementation of docker with full compatiblity. You can use docker images with it which is great, and the syntax is mostly the same.
Haven’t really looked into Podman as I read somewhere (if I remember correctly) that it takes quite a bit of rewrite (from docker compose to podman). Again, might be speaking out of turn here.
I have no problems thus far. It does have a docker compatiblity mode as well
Depending on your threat model incus/lxd won't add too much security as they generally use the same background software as docker, leaving things kernel exploits as vulerable as just docker
My install is on Arch linux, I just installed incus and cockpit-machines and (from memory) that was enough.
But, yeah, it's definitely a step away from a full GUI...
But I think that's part of the appeal (to me)... there's a lot of things preinstalled with Proxmox, (and XCP-ng, etc), that I'll just never use... so if I could get that old Ras Pi3 as a 2nd node in the cluster without all the fluff... maybe that's a good thing?
That’s another fair point. I do have a couple of pi’s collecting dust. As someone else stated, I need to consider the time it takes me to get up to speed with incus. Can you elaborate on your experience going “from 0 to hero” with incus? Just curious.
Just stumbled across this video which might be of interest to you: https://www.youtube.com/watch?v=85d-HcCZURU
Actually pretty helpful!
I would think of Incus and Proxmox as equivalent - both can run containers and VMs. I like the idea of 3 incus servers each with a VM in Docker Swarm mode for running your docker services. Then if you have additional services that aren't a good fit for docker, you can spin them up as separate containers or VMs in incus as needed
Strictly from a container perspective, wouldn’t this workflow create more overhead? For example, an incus cluster for me it would be Debian hosts (layer 1), incus (layer 2), lxd container (layer 3), docker (layer 4), app/service (layer 5). A Docker Swarm cluster (for me) would be Debian hosts (layer 1), docker (layer 2), app/service (layer 3).
Granted a docker swarm cluster would negate the possibility of VMs without having to install something else on the hosts but asking since I’m trying to keep my services in containers.
The setup I'm describing would just be incus running on the Debian host. You'd then spin up a VM with incus and run Docker Swarm inside the VM. Yes, there's a little bit of overhead with running the VM, but it's pretty minimal and makes it a lot easier to backup your whole server (since it's a VM) verses trying to backup a physical server.
You can run all of your containers in this incus VM. Yes, you could just run containers on the host but this setup makes it easier to keep everything self-contained IMO. Also, maybe you'll have a need to do something in the future that can't run in a Linux container (e.g. a FreeBSD server), so having the ability to spin up VMs and not only containers is useful
I've never used Incus, but it's not clear to me why you would choose it over docker, you said that it would be preferable if performance was better, I can already tell you it's not, best case scenario is equivalent performance (since docker runs natively), but I doubt any VM can match that.
Well that’s kinda why I came here to the greater community as I wasn’t really sure if there would be any performance gains or other upsides I’m not aware of. Based on general feedback, it appears that there’s no clear upside to incus.