Deploying Machines with MaaS and Packer - Metal as a Service + Hashicorp Packer Tutorial
ฝัง
- เผยแพร่เมื่อ 2 มิ.ย. 2024
- MaaS or Metal as a service from Canonical is a great way to provision bare metal machines as well as virtual machines. MaaS allows you to deploy Windows, Linux, ESXi, and many other operating systems to your systems helping you to build a bare metal cloud. You can even use Packer from Hashicorp to configure custom images too! We'll cover all of this and more in this tutorial on how to install and configure MaaS from start to finish with Packer!
Video Notes: technotim.live/posts/metal-as...
Support me on Patreon: / technotim
Sponsor me on GitHub: github.com/sponsors/timothyst...
Subscribe on Twitch: / technotim
Become a TH-cam member: / @technotim
Merch Shop 🛍️: l.technotim.live/shop
Gear Recommendations: l.technotim.live/gear
Get Help in Our Discord Community: l.technotim.live/discord
2nd channel: / @technotimtalks
A HUGE THANKS to Micro Center for sponsoring today's video.
New Customer Exclusive - $25 Off ALL Processors: micro.center/3si
Check out Micro Center's Custom PC Builder: micro.center/wcx
Submit your build to Micro Center's Build Showcase: micro.center/dcm
Visit Micro Center's Community Page: micro.center/2vr
(Affiliate links may be included in this description. I may receive a small commission at no cost to you.)
#baremetal #ubuntu #canonical
00:00 - What is MaaS (Metal as a Service) from Canonical?
02:00 - Micro Center / $25 Off CPUs! (Sponsor)
03:00 - Installing MaaS
06:56 - Initial MaaS Configuration
09:41 - Importing your SSH Key
10:23 - Networking Configuration & Discovery
14:05 - PXE & Network Boot with DHCP
15:33 - Commissioning a Machine (Initial Discovery)
18:45 - Power Types & Wake on LAN (WOL)
20:50 - Commissioning a Machine Part 2 (For real this time)
24:00 - Deploying Ubuntu
26:15 - SSH in to machine
26:54 - Creating Custom Images with Hashicorp Packer
33:40 - Uploading a Custom Image to MaaS
38:05 - What do I think of MaaS from Canonical?
39:57 - Stream Highlight - "100 + 50 subs dropped 🫳🎤"
Thank you for watching! - วิทยาศาสตร์และเทคโนโลยี
![ทททททททท #27 [หลุยส์ เค โตมายังไง ณ สระบุรี]](http://i.ytimg.com/vi/CaKoYw_Xmtg/mqdefault.jpg)
How do you provision new machines?
Do you see yourself actually using maas for new devices? I rarely add new hardware, so my maas / ironic experience is fairly limited to work. Really love the video, but curious how maas fits into your homelab plans.
I have been using MaaS for a while now. It is easy to set up and does what it claims to do. Nevertheless I am interested in exploring other options. I wonder if you would consider doing a video on Tinkerbell?
For our metal, we have our own dhcp/tftp server with pxelinux menu to install different systems over preseed/kickstart/cloud-init.. and a simple webserver checking dhcp logs for new machines to show on a php webpage. Once the machines install they run an inventory script to add into our snipe-it inventory. I guess mass has a fancy gui though..
You didn't say what base OS you used, Ubuntu Server, Debian?
For bare metal like you showed, I use Red Hat Satellite.
Hi! MAAS team member here, great video! You mentioned you hit an issue with the deb package, we'd be very curious to hear about it if you're willing to tell us more, and feel free to open a bug report (I'd post the link here to do so, but seems TH-cam doesn't allow external links here).
same here, apt packages give error: unrecognized arguments: region+rack --database-uri . Please add at least the arg --version to check wtf is installed.
To provision PXE boot settings in pfsense: Go to Services, DHCP Server. Click on your respective VLAN/network on the tabs that show, make sure 'ignore bootp' is off. scroll down, expand "network booting' by clicking the blue button then check the 'enable' box. enter in the 'next server' as your MAAS IP and default BIOS file name as the pxelinux.0 nothing else needs to be filled in there.
Thank you!!!
The hero we needed, not the hero we deserve.
I literally stopped the video and checked and found in my PF 2 seconds before he called it out.
On my OPNsense 22.7.11 I can't find this 'ignore bootp'. Did you mean in 'Additional options'?
Is this the same for bios and uefi? I remeber the last time using pxe boot there was a seperat entry/filename to handle uefi
We provision thousands of boxes on both on-prem and in the cloud. We're looking at MaaS to takeover regular bare metal deployments. We're already using Sidero Metal and Talos for k8s bare metal clusters. Awesome video.
I Just wanted to say that youre a life saver!
easy to follow guide...
Keep going!!
Love bringing enterprise tech to the homelab. I'm running OpenShift on bare metal and it has a built-in bare metal provisioner (powered by Ironic) and it's amazing. Especially with the assisted installer
Hey mate, is OpenShift free to download ?
The fuck you have where you can run openShift
@@notafbihoneypot8487 you can run OpenShift on a single node. I'm currently using three ASRock 4x4 with Ryzen Embedded R1605B (or maybe V1605B?). Works great for me
@@stephenreaves3205 oh based. I'll have to check it out. I've always wanted to run it, but haven't due to my misunderstanding that I thought it would need a ton of power
@@RajaseelanGaneswaran you can always use OKD the upstream for OpenShift.
IPMI makes some of this machine onboarding (commissioning, power cycling, running tests, etc) a lot easier, but cool that you tested with a nuc.
This is an incredible work. Thank you for the work you do.
Love this video, i always knew MAAS was awesome, thank you for showing us.
I'm glad that Johnny Depp started releasing such cool videos.
Lol
Thanks Tim for sharing this video 🙏. Was looking for some thing like this for long time 🙂
Glad it was helpful!
Yo this is an awesome run down of maas. I knew the majority of it already but your explaination was awesome! Subbing!
Thank you!
Holy cow! You would make this video 2 months after I spent hours and hours setting this up. 🤣 I love that I can count on you to tinker with all the most awesome stuff though.
I successfully commission, but when trying to deploy 22.04 it fails at the end with “Failed to start Execute cloud user/final scripts.” at the end. I end up with a login page on the monitor, but there is no local login and since the user final scripts failed, I also cannot SSH into it. Anyone know why this is failing? I can't see much on the screen when it fails, but it said something about python.
Great video, seems like something really useful for my 3 raspberries and my 2 laptops!
On a side note though, I might (and I hope to) be wrong about this but during the video it seemed to me that you weren't very "happy" while recording this (tired perhaps?), I don't want to speculate I just hope you're doing okay, your content is superb and you are great and I just wanted to remind you that❤
Thank you and thank you for noticing! I was / am a little sick and was very tired but the show must go on!
LOVE THE VIDEO!!! Would love to see how to do it with proxmox as well as you doing it on RPIs!
It's my first time hearing about MaaS, My first thought of using it would be for malware analysis. Some malware will detect it is within a VM. This may be a way to setup a bare metal server do your analysis and wipe the machine.
Hi Tim this is a very nice tool, just tried deploying some images, it indeed works very well. The situation of creating images seems pretty cumbersome, would be nice if they created a trustworthy depot with standard images, also I miss a lot of images for various OSses. Anyway thanks for showing us, great content.
Cool, def want to evaluate this.
In pfSense I believe the setting it is pretty simple - under Services - DHCP Server - scroll down to Other Options - Network Booting.
Hello Tim, thank you very much for your great videos. it would be nice if you could record a video doing automatic node scaling for Kubernetes based on the load.
Where ever you find this software @Techno Tim I'm always excited to expand my HomeLab. This had me thinking about an Idea I had for a project. Thanks for another amazing video.
as i don`t have more bare metal machines than one proxmox server, i use packer for proxmox templates and terraform for actual vm deployment. Also made terraform module to encapsulate vm deploy as well as dns vm record creation in powerdns. Works like a good clock, no problems with this setup. Next step is to build updated images every so often, so they will be updated, when i need new vm, probably will do this with gitlab ci with scheduled pipelines
I have experience with Fog server to capture and deploy images. I'd be interested to see if MaaS has the same flexibility or if it requires more architectural design - meaning that it should be the foundation of the infrastructure rather than something like Fog that can be integrated at any point.
Great video!
this video very helpful, thanks you, for you doing
I assume MaaS and Packer can also distrbute Rancher's harvester hypervisor? Great explanation of this software-combo, clear and mixed with some humor and fun! 🙂👍 Thank you! (I too prefer to stay away from snapd btw)
I have the same question. MAAS is a great tool and limitation is image building.
These days with kubernetes all you need to good Kubernetes distribution. We landed on Rancher and pretty happy.
If we can figure out Harvester, it would make a pretty decent combination.
Harvester cluster provisioned through MAAS for VM workloads and Rancher Cluster for container workloads. May be Tim can make a video for us on Harvester with Packer.
I have started using Talos Linux to build Kubernetes clusters, this looks really nice for non-Kubernetes use cases though!
Packer...the unsung hero
Thank you for the tutorial. Just one tip, it's hard to see many of your commands with your camera view in the corner. Some people have this but minimize it when showing the commands.
Hi Tim nice video - why not put the commands in the description? Often your commands are overlapped by your video feed
I checked your channel 2/3h ago
x)
Great video! One question though: what is the external dhcp you’re using? I couldn’t recognise it from the GUI you showed.
Hi @Techno Tim, you are getting handier with new-open source infra that's really awesome stuff, Looking for a video of your home lab security and vulnerabilities/endpoints and how you are dealing with a part from pi-hole DNS
Thank you! I already have one on security, check it out th-cam.com/video/Cs8yOmTJNYQ/w-d-xo.html
When I look online, I saw that they recommend setting up a little webserver which can received a webhook request for starting the machine with WOL.
27:45 well, that became a contentious point now ;-)
THANKS A LOT
Thanks for the tutorial, does it work with pfsense?
My setup is: Proxmox that runs pfsense and a few vms, one of those vms is MaaS, but at machines tab, it haven't detected any machine
Can you post a link to your 3 NUC rack mount kit?
Tim, any info on this?
Any interest or experience with Openstack Ironic or Equinix's Metal?
Tim, could you do a video on zero trust for the home user/lab?
Would be nice to define a template for a VLAN so I can deploy a lot of machines fast in a setup lan
Ayo used MAAS when it was new, around 2014-15 ish to provision about 30 dell optiplexs for a beowolf cluster at the time
38:15 I think I need this NUC shelf in my life. Where'd you find this?
I'm a fan of MaaS and have used it for a while. It doesn't play well with Pis at the moment. My favorite way to use it is with Terraform and KVM. I can then setup bare metal infrastructure as code.
Where does KVM come into play?
@@DrDingus KVM - kernel based virtual machine. I use Terraform to provision machines as KVM hosts that can then run virtual machines. Terraform then provisions those virtual machines as required. I also have deployed Juju via Terraform using cloud-init to virtual machines. Then, I controlled Juju via Terraform to deploy Charmed Kubernetes that I then manage via Gitops.
@@jamis2341 Awesome! Sounds complex. Is this for your job?
It's for a business I run. I do some things like this for work but those aré in GCP and AWS.
@@jamis2341 That's pretty cool. I have 10 headless NUCs. I successfully commission all of them, but when trying to deploy 22.04 it fails at the end with “Failed to start Execute cloud user/final scripts.” at the end. I end up with a login page on the monitor, but there is no local login and since the user final scripts failed, I also cannot SSH into it. Any idea why this is failing? I can't see much on the screen when it fails, but it said something about python.
Is there reason you don't tell the machine to boot pxe boot by default in the bios instead of having to smash f12? Will that break things?
No reason other than after I image the device, I want it to boot from the first hard disk :) I think the best of both worlds is to tell the machine to boot to nextwork if I receives a WOL packet that I send. That way a physical power on is always disk and a WOL packet is always network!
Ex MAAS PM here. You don’t need to worry about the order - MAAS knows it deployed the machine so even if it net boots, it directs the machine to boot from disk. Try it!
Hey Tim great video! Is Mass the equivalent of OpenStack if I am wrong?, am I?
Openstack is not equivalent to maas. Openstack as a broader project is a full cloud provider. There are projects that provide baremetal provisioning in Openstack, such as Ironic. There is a lot more that could be said, but hopefully that helps.
For those who might not like the Canonical ecosystem, check out Foreman. Ships with a Puppet server for automatic configuration of new hosts.
How complicated is the setup compared to MAAS?
@@DrDingus That's a good question. The job where I was using it already had it up and running.
My biggest issue with MaaS: at the last power outage it tried to provision an unknown (to it) server - WTF. Using stacki until now and were testing maas as a replacement, but I don't want provisioning software to touch machines if they are not configured with it ...
We looked into MaaS but found it was too much overhead for a pretty much one time server setup process. Dumping Proxmox onto the server blades over iKVM was a maybe 1-2 times in the blades lifetime setup, then everything is controlled by the hypervisor from there on out.
Where did you get that intel nuc rack mount?
Pfsense makes this super easy... everything is in dhcp settings... have all network boot devices pointed on that specific network point to next hop or server ip... you can also restrict to Mac or !Mac (set aliases to help with KNOWN DEVICES)
Still resorted to semi seg full seg-air gapping deployment servers from the production line... all good with home environment but a crisis waiting to happen in HA environment
how will you manage the partition?
Does it support being a pxe server for machines with no hidden/add
Is there any way to automate a Home Assistant install with this? I am looing for a way to quickly rebuild if my install becomes corrupted, the more time I spend on Home Assistant the more nervous I am about having to reinstall it.
Sure if you can build it with packer, you can deploy it with MaaS
I run maas for provisioning a home cluster of Dell Optiplex Micros. These machines are quiet and take little power to run, so they are perfect for a homelab. However the power controller on these machines are controlled with Intel vPro or AMT (Active Management Technology). It took me quite a while to get maas to play nice with these machines. DHCP assigns a random IP but the power controllers are static. This is supposed to be handled by assigning a hostname but for some reason maas doesn't recognize it. I solved it with static DHCP entries to assign the same ip to the mac of the machine. Weird. If you run into this problem LMK.
Maas is a great system. Tikerbell does not yet support vPro AMT so I can't use it in my lab.
I successfully commission, but when trying to deploy 22.04 it fails at the end with “Failed to start Execute cloud user/final scripts.” at the end. I end up with a login page on the monitor, but there is no local login and since the user final scripts failed, I also cannot SSH into it. Anyone know why this is failing? I can't see much on the screen when it fails, but it said something about python.
Will MAAS run on a Pi? Also could you make your NUC netbook by default in the bios? You know so you don't need to hit F12?
I have a 2 node Proxmox cluster right now, node 3 should be here in a week, each with 20c/40t, 128GB ram, 250GB boot drive and 1.2TB CEPH. So VMs inbound for sure...
Apprently PXE for pfSense can be enabled using the tftpd package. I won't try it myself but you could :D
tftp server needs to run from maas, he meant the dhcp settings..
@@Pytte yeah, but the tftpd package for pfSense enables the "TFTP Server" and "Network Boot" options in the pfSense DHCP settings. Not sure whether that can be used to forward the requests to the mass server, though.
@@Darkk6969 gotcha. I though the pfSense DHCP service didn't have "Network Boot" all by itself. Just checked and it's already there, tftpd package not needed. Sorry for the confusion :)
Is your subnet managed or unmanagedon MAAS?
If you are using a virtualized MaaS, make sure you have nested virtualization enabled on hypervisor or qemu will not allow kvm to work. You can quickly check your guest by running the following command: egrep -c '(vmx|svm)' /proc/cpuinfo, the number should be greater than zero if virtualization from host cpu is being passed through to guest ;)
Does the Maas controller utilize kvm?
@@DrDingus In my case I’m running MaaS controller as a VM on Proxmox so it’s under kvm in my situation. I believe in the video tutorial Tim is also using Proxmox but never explicitly mentioned that you need to allow host cpu pass through for the Packer extension to function correctly. If you’re not using Packer then you are fine.
@@jschoenberger Interesting! Thanks for that info.
I don't have DHCP network boot in my router :(
I remember to make it work somehow PXE on my network but can't remember HOW I did it :D
Kinda frustrated, otherwise your video is really well like always, I could follow all the steps until reach this critical step.
How do you deal with mashing a function key when you have hundreds (or thousands) of hosts? How can this be scaled beyond the home lab?
Depends on the machine, but if it’s legit servers you’d use their out of band management, eg IPMI, iDRAC to do it, which MaaS supports. It refers to it as BMC
Get a machine with vPro and you can use intel AMT to automate this from MAAS
Some NUCs have vPro/amt
Which is better..
Truenas or Proxmox.. I have a bare metal server and dont know what to do.. confused with the os.. need guidance
The saying goes that if you want a NAS that can do some virtualization, you go with TrueNAS Scale. If you want a hypervisor that could be made to take on the responsibilities of a NAS, you go with Proxmox.
Both use KVM under the hood.
@@execration_texts Thanks, for the guidance... I need to think now what to use !!!.. because both are needed
I've been leaning towards Tinkerbell over MaaS. Seems easier and more flexible.
I've read about Tinkerbell but seemed a little more complicated than I was willing to take on in a week. LMK what you think!
@@TechnoTim I concur. Tinkerbell looked very complicated when I was assessing it (unless I missed something). I opted for Matchbox + dnsmasq because of the simplicity and ignition support for Flatcar or CoreOS.
@@Gunzy83 hand you successfully used matchbox?
Good tutorial, although some things don't work as they are. For example, to make the custom image, you must run the make custom-clouding.tar.gz as root, because no matter how much sudo, it always fails. The other detail, several times you unintentionally cover the commands with the box of your image. Thanks for the video
Thanks and sorry about that! All of the commands can be copied and pasted from my web site!
I reinstalled MAAS and was able to get it to finalize the scripts when deploying, now it gets stuck at 'rebooting'. When I look at the monitor for the servers, they are stuck at a screen that says 'pxelinux ... booting local disk...' and then I'm stuck at a prompt that just says "boot:"
I figured out the issue. It was trying to boot from uefi bootdisk because it was enabled in bios. I unchecked it and it booted into the systems fine.
I have 10 headless NUCs. I successfully commission all of them, but when trying to deploy 22.04 it fails at the end with “Failed to start Execute cloud user/final scripts.” at the end. I end up with a login page on the monitor, but there is no local login and since the user final scripts failed, I also cannot SSH into it. Any idea why this is failing? I can't see much on the screen when it fails, but it said something about python.
I ended up fixing this. I had some dependencies get out of whack.. I think it was the way I updated it. I built a fresh MAAS install on a fresh VM and no longer had the problem.
What is those dots on beside of proxmox vm
Tags!
@@TechnoTim I literally started using tags last night even though ive been using Proxmox for years 😂
sudo to run packer? Not necessary. Everything it does can be done in user space unless you're doing something really odd locally... but even then, the init should be fine. When it "installs" it is just downloading the plugins to the current (pwd) directory area.
EDIT: Packer is a good tool to learn, so I would say it is preferable to the built-in paid part of Canonical MaaS.
Hi Tim, how can i get in touch with you regarding business opportunities? Thanks.
You have GOT to fix that "bed head" of hair, my friend. LOL
I'm using the Starlink router... I remember this was not configurable, anyone think otherwise?
Interesting.
The "DAY" HAS COME, Bro look at JUJU from canonical.
shame there's no proxmox in packer. would be great to build a cluster
This SSH Key is kicking my butt, I tried putty, SSH on Ubuntu, and Windows CMD/Powershell. I am able to get it working but after a reboot I get Permission denied (publickey). I have been doing this over and over for like 3 weeks using 4 computers and it's killing me! I am trying my best to not say Maas is a Peace of S++t. Please help me get this working and thanks!!!
35:15 it's Storinator sir ?
I own one, yes!
@@TechnoTim can't wait you build HCI Proxmox with Storinator backed by Ceph also 🤩
gave the vid a 👍🏼, but I'll never use it myself
You are mucho good for revitalizing my passion for all this automation and new tech...
But muuucho bad for my wallet.. Dont need to find excuses to buy new infra hardware to play with and build on :(
So this is basically Harvester, just.... different?
It's more akin to Foreman
@@funpunx never heard of that, thanks!
@@cheebadigga4092 have you heard of Red Hat Satellite? Satellite 6 is based on Foreman. There is also SUSE Manger based on Uyuni that is pretty similar from my understanding (I've never actually used it).
@@FlexibleToast no I didn't know about any of them. I'll check it out, thanks!
@@cheebadigga4092 just a heads up about Foreman. From what I've seen companies usually buy RH Satellite and use it as a glorified repository. It's completely overkill fro that as it has all these extra functions. I'm not a RH Satellite specialist, but there is a ton it can do and tends to get real complex real fast. I've hear that SUSE Manager can actually be easier to use. It's based on using Salt and again, I've just personally never used it.
If what you're doing is just virtual machines, it might just be a lot simpler to use Terrafrom/Ansible together. That's a very common approach in the enterprise world.
My feeling on this is that it's either 20 years too late, or a solution looking for a problem. The reason why I say this is this would have been awesome 20 years ago, when imaging new machines were super expensive, and virtualization wasn't as heavily used as it is right now. The other reason why I say it's a solution looking for a problem, is that it's super cheap to get images, even custom images, on machines during the actual building of these servers. Hell, I believe Supermicro will work with your engineers and will literally fly out on a big enough order to make sure that all the software is working correctly, BEFORE imaging the machines, and then send them to you. Let's say you just want to use this for regular pcs, instead of servers. Even then, this doesn't do enough to be worth the support. There is no remote managing solution, other than ssh and whatever else you push into your image.
My real guess is that some businesses must just have spare machines available for development/deployment, and rather than having to constantly buy or sell machines, they just keep them around, turned off, and ready to turn back on in a relatively quick time. Otherwise, k8s, openshift, and others just make more sense with an ipmi call via vmware esxi or hyperv just makes more sense.
I talk about this at the end when I discuss how this fits in, I think bare metal k8s clusters is the way the world is moving.
I think you miss the point. Installing an OS on a machine or using the one from the manufacturer and maintaining it is looking at infrastructure like pets. Being able to reprovision quickly on demand is the game changer of cloud and bringing this capability to bare metal is awesome.
@@Gunzy83 No, the point is I can reprovision with vmware esxi just as easily as MAAS. I can even install remotely VXE, just as MAAS does, with esxi. Whether it's bare metal or not is the real issue. What advantage does bare metal install vs bare metal hypervisor install get you?
@@stefsmurf It gets me a cluster of machines a fresh OS without esxi on it because I don't want/need ESXI for my needs.
MAAS, monitoring as a service? Its metal?
MAAS is easy to set up & does so many things right, but there a few nuances that are very frustrating.
I successfully commission, but when trying to deploy 22.04 it fails at the end with “Failed to start Execute cloud user/final scripts.” at the end. I end up with a login page on the monitor, but there is no local login and since the user final scripts failed, I also cannot SSH into it. Anyone know why this is failing? I can't see much on the screen when it fails, but it said something about python. Any ideas?
You ok buddy?
Aaah crap... Snap... I'm just not going to do it. I do not want to fall in the trap of getting locked in again.
Docs have apt commands too!
@TechnoTim OMG!!! Ok, back on the train again!, do you have a link? I can only see older versions for apt installation.
Yeah, I think that since 3.x there is only a Snap version. Sooooo... Sorry Canonical. No-Go
@@Caliber632 just bite the bullet. it's the way of the future.
Oof, 4 minutes in and resorting to snap. That's a big no from me. I would love to see an updated video with apt.
The docs also have the apt commands!
@@TechnoTim I don't see the instructions for v3+ for apt. For version 2 I see apt instructions and it seems to be limited to Ubuntu which is also a nonstarter for me. I'll never understand Ubuntu Server when Debian exists.
Tim, when's the last time you got a hair cut?
It’s been a while! Letting it go wild for a bit
@@TechnoTim That's good news. I thought your head had exploded from figuring all this stuff out!
бро у тебя точно все нормально ?
5:57 lolz port 5420, he's either dyslexic or a stoner.
My feeling on MaaS:
Zoomers discovered renting hardware and made a dumb name for it.
Honestly, Canonical is so disgustingly in contempt of the open source community that I'll never touch one of their products again.
6.2k views and only 439 likes? Come on people, you can do better ...
Snap man... I can't like a video or product telling me to use snap.
snap - no thanks.
Please tell me you have changed that public key already.
Why, it’s a public key? If you use it, that means you trust me. They are designed to be shared.
@@TechnoTim I trust you, Tim.
Bad/lack of documentation should be a NO-GO to every serious contender.. such a waste of propably good product. Same with juju, literally nobody uses it.
Packer for free.. probably not true anymore
BrB ima work on my new business model : BaaS (Business as a Service) we do everything for you and you make money without doing anything.
I need to start calling my rental property HaaS (Housing as a Service).
@FlexibleToast Well, "HaaS" is just "house" but with a strong accent
@@Jimmy_Jones even more reason to call it HaaS.