Use code christianlempa at the link below to get an exclusive 60% off an annual Incogni plan: incogni.com/christianlempa Thanks to Incogni for sponsoring this video.
Hey Christian. Just so you know, you don't need a dedicated "nic" for proxmox clustering. Just a dedicated network, a vlan works more than fine. Using a vlan also means you can bond your NICs giving you overall better performance and redundancy. I recommend in your case, bonding the SFP ports, then bonding the bond with the inbuilt ethernet for fail over. At that point you can vlan out to your hearts content :). This is also how I handle my ceph network. (My team and I manage a proxmox cluster at work, I only watched this video because I really enjoy your content)
I run proxmox at work, with a big cluster, multiple shared storage devices. It works great, and, I'm happy to see you added a qdev to your cluster, because it really is needed to maintain cluster integrity. As far as shared storage goes, any cheap NAS will work. I'd recommend something that has NFS shares. For homelab, gigabit is mostly fine unless you have a ton of storage. The price jump networking wise to 10gbps isn't bad, but, the storage price jump that can actually utilize that network speed can be quite expensive.
@@jonathandavis4711 depends how you build things. Just saying you have something doesn’t imply it works or is bad. How was it setup and engineered? It’s people with raspberri pies as nas. You can use the synology os on x86.
Christian, you can do a poor man's HA with local disk by scheduling replication of your VM's. That way a copy of your VM is standing by on the failover node. Also, I didn't have luck with SSD RAID for hosting VM's. The cost was exorbitant, the capacity was too low for my purposes, and the i/o wasn't much better. I went with more HDD's and an SSD ZFS read/write cache. P.S. You're doing great work. Thank you for sharing and for your joyful presentation style.
Exactly! I used to have a 3rd server running TrueNAS Scale with NVME shared storage. Eventually I retired it, since I got a little tired of the extra high-pitched 1U noise. Now I just have NVMEs in the two ProxMox servers for VM storage and have set up replication for every 5 minutes. Works fine for my needs.
I vote for a CEPH cluster! I would love to see you set up a CEPH Object Store and File System on your 3 ProxMox servers instead of relying on a TrueNAS server or other NAS. I understand that RU-vid content creators have a need for a lot of storage and high speed networking but not the average person running a home lab. Having a “local” file system for your work load that is mirrored to other servers would be very appealing.
FYI, check your VM CPU type is not "Host" as both PVE hosts will definitely need the same CPU if "Host" is choose but should be fine with most of the other CPU Types.
I am wondering: I have 2 machines, one is Ryzen 7 and another is Ryzen 9 - would in this case "host" be ok or not? I understand it is abouch architecture nor really the CPU type
Christian during a Stream: "you know energy is realy expensive here, i was thinking about downscaling" Also Chsiatian: "wanna se my new 2 node pve cluster"😂😂😂 Ps. Love your content
Christian, I have been running Proxmox for ~3 years now in my home lab and used a TrueNAS shared storage solution. But mid-year last year I decided to add a 3rd node to my setup and implemented an all SSD Ceph Storage. So far it has been flawless.
Been following your vids for a while and they just keep getting better. Finally have the confidence to press forward with my homelab thanks to you. Rock on brutha!
I run two node Proxmox cluster at home with one of VMs being a router. Nothing beats migrating router to second node, while performing upgrades on first one 😀
I have successfully running a Proxmox cluster with 2 nodes and a quorum device in my small home lab with a simple storage sharing solution without external network shares. Both nodes have a dedicated SSD with the same size just for this purpose. How to set it up? On the first node you go to disks, create ZFS, select the disk and give a name (this name should be a general name, not dedicated to a node, like 'pve-data' or something similar), select add storage, everything else on default and check the box on the disk you want to have as storage in the device list, click on create. The first node now have a new storage. On node 2 do the same, go to disk, create zfs, now important use the same name like on node 1 (in my example 'pve-data') and also it is very important to deselect the checkbox 'Add Storage', select the disk from the device list and click on create. After that you can go to Datacenter -> Storage, select the ZFS storage 'pve-cluster' and click on Edit. In the dialog you select all nodes in the dropdown menu named Nodes. After that the nodes will be come the storage listed as local storage on each node. With replication tasks for each VM or LXC you can set how often a storage of an VM or LXC will be synchronized between the nodes. In HA you can also set the list of containers or VMs you want to have migrated live between the nodes. You have to do this for each VM or LXC container, but it works very well for me. Thank you for covering this Proxmox topic. Hope to see more of this.
@christianlempa I recently created my Proxmox node and had the same problem when migrating VMs, but I fixed it by selecting on the VM configuration > Hardware > Processors > Type > x86-64-v2-AES. This works flawlessly between my Intel N100 and my i7 6700T
I was going to mention this also but you beat me to it. This is the way for different CPU's. Just find out the common instruction sets and use the highest x86-64-Vx so your cpu's can use those instruction sets.
For HA you definitely have to assess what you want vs. what you actually need. Having HA cluster with only single node shared storage, you are back to square one on being fully redundant.
Currently I am not running a cluster, but have plan for this in future. I was planing on failing services over, but I see that is going to take more than initially expected. At least I can do like you are doing now.
Nice work. Proxmox is such a nice project. I came here attrscted by the 2 nodes tittle. I imagened the voting issue and the qdevice to help out. HA is really nice. As you said a 3rd fully cspable node allows you Ceph among other things. And that is another rabbit hole. Regards Christian!
You can do HA with two nodes by setting up a replication job of the VM between the two nodes. That way there is an (older) copy of the disks on the other node so HA will be able to start up on the 2nd node.
ALL cluster nodes need to have the latest live data for it to be classed as HA. Otherwise you could run into some nasty split brain situations depending on what you are running in the VMs.
You should also be able to set the "CPU TYPE" under your VM hardware to a version supported by both processors (like x86-64-v2/3/4 processors) to improve live migration compatibility. PVE has a ton of options for each generation of chips going back to 486, so you can really tune in the instruction sets made available to the VM this way. Might be worth testing if you have a mixed cluster environment.
very fun! Thanks for the great video! have a beelink mini pc (intel) as primary, and an old 2012 MacBook Air (intel) that I recently put in a cluster. They are sharing storage for backups. Migration works great so far, even with the CPUs being different models. Looking forward to your SSD NAS storage in the next 1/2 of the year.
Yes, there are some quirks with a 2 node proxmox cluster. I run the same, but no qdevice. I just have a direct 10Gb connection between them. I did it for a similar reason as you, I wanted fast migration between the servers. Fun fact, I'm running PBS in HA with the virtual disk running on a 1Gb link to my Synology NAS, and the PBS datastore is on the Synology also. It works quite well considering it's going over 1Gb. I've backed and waiting on the Zima Cube Pro to arrive as that will become my 3rd Proxmox node running TrueNas VM configured for my HA storage with a 10Gb link and NVME storage available.
Hi @ChristianLempa one note i made was when you used qdevice when running HA you need to ensure that the q device has root access which I had to do to get my cluster working but great video I use 1 dell 5090 & lenovo in my ha setup pm cluster & have a raspberry pi 4 (which also runs 24/7/365) as my q device but great content
Thanks for the video. I have a proxmox cluster running, but I need to reinstall proxmox on the cluster. There are a few things I didn’t take care at the first time. Hopefully with your video it will be better :)
I'm running a truenas cm with 4x4TB NVMe drives just for VM migration. Works very for VM migration and very quickly. If you can, it's definitely worthwhile to have a look
I was running a 3 node cluster but have put them back to standalone nopw, after ~12months it ate through the alwayson nodes OS ssd to around 97% wearout, this was a cheap consumer SSD so would deff recommend either a enterprise ssd for the OS or a hdd and configure it as a mount point for the logs and quorum db to write to.
That's so cool! I enjoy seeing your setup grow. I'm considering building such a cluster myself. Given the date/time synchronization requirement, I wonder if it would be worth implementing PTP with something like a Time Card.
really good content - take this to its ultimate conclusion is all i would say - think about 2 node 25 or 40g cluster - no switch needed - you could use the 56g dual port cards (40gbe) for cluster and then a 10g management network - the netfs is a good one to dig into - ceph/nfs/zfs/gluster? i think you are going to have to experiment - maybe try some nvme arrays? thanks for the content!
Hi Christian, thanks a lot - another great video! 👍 - Would you please also share which NIC you installed and if you are satisfied with it (for Proxmox usage)? Thank you.
i’ve been running proxmox 3 node cluster with ceph ha for over 6 months now and it’s been stable so far. i even upgraded to proxmox 8.1 from pve 7 without any issues. curious why you decided not to go this route. maybe an opportunity to make this topic a 2 part series? 😅
Probably due to cost in electricity and hardware. He might pick up another AMD motherboard to run as third node in the future. He's also thinking of what home lab users would do. For us home labs we tend to reuse what we can find. I use mostly old Enterprise hardware from work as it was going to be e-wasted anyway. Figure I give it a second lease on life. Plus I source old gear from e-bay.
You don't actually NEED a super fast storage for the CTs/VMs for migration. It can help when/if you are booting up/shutting them down regularly, but if they're running pretty much 24/7, once the CT/VM is booted up, it doesn't really need much (unless you are doing I/O intensive tasks). I have three N95 Mini PCs and run Ceph between them over GbE, and it works just fine. If the CT/VM disks are actually on the shared storage, I have live-migrated my CTs/VMs between the nodes in as little as 8 seconds (because it doesn't need to move the CT/VM disks around since it is already on shared storage).
Great video as always. To address your shared disk with 2 nodes issue you could use Stormagic SvSAN which takes local drives from each node and presents it as a iSCSI shared disk to the cluster. They also directly support Proxmox. Give it a try.
@@danilfun I currently run CEPH on proxmox and my VM's are slow, I am using all ssd disk and think it maybe the 1gig network and how CEPH replicates the data. Reads are fine but writes are really slow get like 25meg writes.
I am using the cloning to make sure that my vms/containers are migrated automatically if one of the nodes are down. That way, I don’t have to deal with the network storage which brings its own problems. Also you can change the votes in corosync config. That’s how I ran my cluster for a while where my main node had 2 votes and slave had 1 votes.
I am running a similair setup in my homelab. 2 nucs, 1 pi zero with ethernet hat as qdevice. Also configured HA. Ive setup ZFS on the nucs with replication. Therefore not requiring a shared Storage. The nucs are connected to 1gbps connection but thats fast enough for my lightweight vms and containers.
so I run a two node cluster, one machine is a custom build that is used for things like media streaming / home lab and NAS (TrueNAS Scale VM with SSD's passed directly into it) and is only powered on when I need it (WOL), the second is on a Zimaboard that runs a Docker LXC for my 24-7 services and PiHole DNS LXC which is all working great. As I didn't need (or want!) HA services on Proxmox (I disabled them as they seem to wear SSD's out super fast for some reason?) and one of the nodes would be switched off most of the time I looked into getting a qdev but found out you don't need one if you set "two_node: 1" & "wait_for_all: 0" in the "quorum" section of "/etc/pve/corosync.conf". If one of the nodes is down the VMs will all boot like normal and can edit the VM configs, backups run, etc.
My home lab is simple 3x N100 mini PCs in Proxmox HA config. Synology DS423+ (NFS) used as shared storage. This HA thing save me so many times when one node goes down. Even with my local 1Gbps network everything feels snappy.
Even I will be rebuilding my Proxmox with the hardware I recently got from work , Nutanix nodes G6 3.8TB RAM 60 Cores 120TB Storage Will be epic homelab
You should definitely set the shutdown_policy to "migrate". If you try to shutdown one Node, it will automagicly copy all VMs to to remaining Nodes based on your HA-group prioritys. I've never used the VM replication in my prod environment because it's running ceph, but maybe it would solve your Problem with the missing disk after shutdown.
I use replication with HA and have different sync schedules depending on the VM importance. All of my hosts have storage pools with the same name and I’ve been able to fail over pretty effortlessly. The main issue I have with HA is when I do a switch update, I have do disable HA otherwise all of my HA hosts will shutdown when a total loss of connectivity is detected bring down my whole lab even tho I have fail over connections on a separate switch. Still trying to figure that out. lol.
8:17 It is still possible to change the IP address of the server. You just have to be very careful because you have to change many different things, not just one file or one adapter!
One more thing that will bite you is that your storage has to be the same on both machines. If a VM is on local-zfs on one node, you have to have a local-zfs on the second node in order to migrate.
@@gilfreund That applies to any storage type. The name of the storage that was saved in the VM config is what the hypervisor will be looking for to load the disk image from. That applies to local, shared or replicated storage.
Hey Chris, take a look at CEPH and maybe you can build a network over USB4 thunderbolt? ;) at least this is what I run on a 3 node cluster with NUCs 13ths gen. But even then: it takes around 2-3 min after a Node dies to run the HA migration. But not because of slow network 😂 getting around 20-30 gbit over the TB4 ports
Hi @Christian Lempa, would you mind sharing the model of Be Quite power supply you used in this build and confirm if you used 120mm fan on top of the CPU heating? Would greatly appreciate 😊
You can make a replication job to node B, the migration is faster then, because only the icrement to the last replica have to transfered (+RAM Data). If node A dies, over the HA function, node B boots up the latest replica of the vm.
@@kevinneufeld3195 if it is for services that dont't need top performance (IOPS), setting ceph in VM's is more than enough. You can also have almost bare-metal performance is you passthrough the sdd's or nvme's instead of virtual disks.
Two nodes plus a q device (pi 5) seems perfect for me. They can each backup to the other node. I tried ceph but my ssds were too slow. So i just have redundant services on each node. So a pihole lxc and a tailscale lxc on each one.
@@rogertan1130 you want IOPS, but also consumer SSDs typically slow down as the drive fills up which is not ideal for ceph (initially a nand cell operates in SLC mode, when the drive fills up it divides that cell in to smaller chunks) and due to the way write caching works write performance falls off a cliff as well. However, a lot of people misconfigure ceph as well. You can have a ceph cluster with HDDs that performs well enough as long as you have enough OSDs, but you also actually need the network speeds and latency to support this. 10gbit is often recommended, but I see it more of a minimum to get any decent sort of performance with ceph. Getting ceph on it's own network should help a lot with this as well.
ssd or network interface? a typical SSD has a speed of 3500MB/s 1 gigabit network has 1gbps = 1000 mbps = 125 mb/s 10 gigabit network has 10gbps = 10 000 Mbps = 1250 mb/s with is ~1/3 of the typical SSD speed...
So I've been wanting to give this a try, but after reading some of the comments below regarding hardware requirements, I'm thinking it may not be worth the effort. For the average home-lab guy who has a windows SFF pc (NTFS file sharing / WSL2 for linux docker stuff), an HP micro-mini pc (Proxmox) and a raspberry pi4 (docker) all making up the environment, do you think this is something that is workable or possible by adding an additional Proxmox node running on a capable Dell laptop?
FYI I tried to do a qdevice inside of a docker container & it was a mess. I end up messing up my entire cluster. I got it added to the cluster but it was non-voting. While messing around with it i borked my entire cluster so i probably won't be doing that again.
Quick question: I have 2 hosts as well, I dont really want to get a 3rd either for HA to work better. Can a Raspberry Pi 5 per your video be used also as a Voting member and it would work a bit better spinning up the VMs on the other host with the Pi being up? The one I was gonna get is the 8GB ram and 128GB storage micro sd card. Thanks, just looking for general feedback on this idea.
If you use a 3 node with ceph/gluster or a network filesystem, the migration is happena almost instantly cause only ram e config file need to migrate from nodes.
I tried two nodes proxmox cluster too a couple of months ago, but I ran into some weird unstable situation when one of the nodes goes off, I think the second node frees until the first goes online or something like that, so I noticed that is not a good idea.
to get live migration work more reliably this can help: pve8 has cpu types which wheren't available before. sth like x64_x86_v3 is for example intel skylake and newer and amd epyc (i assume zen1). just klick help in cpu menu of a VM and as others pointed out, with replication HA is possible. but loose data up to migration interval. otherwise do a bulk migration before shutdown?
My first foray with proxmox has been less than fun, I migrated from esxi, first few days were great then my truenas vm started doing some weird stuff, like randomly rebooting, seriously considering fgoing back to esxi
You dont need shared storage or ceph for HA... It can be done like you have it now, but you need enabe VM replication for example to replicate vm every minute... Im not sure if you can do it only on ZFS for this because I have done it only on ZFS... Maybe also check HA groups because if you plan turning off one node periodicly you can make 2 ha groups and set in which host you preffer run VM for example grop run_on_pve1 and run_on_pve2 then you set in ha which vm use which group and select for pve1 group that it use pve1 host and pve2 group use pve2 host. Then when you want turn off or maintenance host pve2 you edit ha group run_on_pve2 and instead of selected pve2 host select pve1 and save. System will online migrate all VMs which run on pve2 host to pve1 host and you can turn off pve2 host when it finish... I think that this can be done also with api calls(edit ha group), check if vms are migrated and then turn off host on schedule(cronjob) on your zuma board... After host is turned off you can edit run_on_pve2 ha group and select thqt group use pve2 host and when host will be turned back VMs will be migrated back to pve 2 host...
Yes, I just did that, but it's very little space, so I'm not sure how much it will make a difference. I have also added 2x 80mm fans at the front of the case and I believe that helps much more than the cpu fan...
You don’t need shared storage, you can also setup replication to run ever 15 mins…or less if you want then you don’t have to worry about HA failing when you pull the plug
I use application high availability btw I mean, I have three pfSense virtual firewalls in HA (not the recommended scenario, they go with 2 but, meh, works) I have three VMs running mariadb galera in a cluster, HA database I'm learning kubernetes just for this reason (will have 3 master nodes and 6 worker nodes, 3 VMs for each of my proxmox hosts) I don't need a shared storage, LOL I also containerize everything so, kube it is
You dont “need” a separate device for shared storage. You “could” use CephFS or GlusterFS to create a shared storage pool from the disks/partitions in you servers. Im using quotes because its possible, but a seperate device is probably still beter.
Just add dedicated HDDs to both Proxmox nodes just for VMs and then format those using ZFS. Then you can mount those drives as shared and have ZFS doing replication every XX minutes from node to the other. No need for external storage. Craft computing YT channel has a nice tutorial
Yep, that is currently what I do for home and at work. Yes I could have used CEPH for work but I don't have time to troubleshoot cluster issues. ZFS is easier to deal with and fix.
I've been thinking about doing something similar but have been struggling with shared storage since I don't want to use the slow HDD on my NAS and don't want to build a SAN. I didn't catch what you decided to use for shared storage. Do you mind clarifying?
@@xConundrumx iSCSI seems a bit dated at this point and there is very little used gear available on ebay that isn't ancient. I'm trying to avoid using my NAS because I'm using Unraid and it's not really fast enough without the right RAID configuration for a target.
recommend 3 nodes at minimum, though you can get away with 2 and a qdevice.... Even for a local cluster you need 3 nodes, or a QDevice running outside the nodes. 2-node clusters are only for home-labs and experiments... If one node in this cluster goes down, the other will be in R/O mode or will fence itself. Cluster must have quorum, in a two node cluster - quorum is 2. In 3 node its also 2, so either one node + vote survives, or both prod nodes (if vote is down).
Good question, I've done it because in the future I might use a switch or another network instead of a direct link, but yea... it doesn't have to be that big :D
7:00 not sure which would be better... direct connection between the two nodes? Or bound interface from the two 10 gigabit ports. through switch. Quick question I can think of: If you decide to add a third node? What do you do with direct links? btw you should look at the new settings for v8 for vlans in the datacenter/network/
I don’t understand why you would want to go for a 2U case instead of a 3U case especially if you are planning to use an ATX PSU. That’s still gonna occupy 3U in your rack because you can’t put another server directly above as the PSU needs to suck in air from above. You could put a full height PCIe card in the 3U case without additional riser cables and you get proper front to back airflow…
I wanted to have a little more space in the rack, and since Silverstone doesn't have a 3U case, I went with 2U instead of 4. But yeah it has some concerns about airflow that I didn't thought about in the first place.
One Tip do not run it on a SD Card because it writes alot of data for the Q DB. I blew up an expensive SD Card within 2 wks. You can also use a cheap OrangePi as long as it can run a linux derivate preferable Debian because Proxmox is Debian based.1 GB LAN Port works fine as well.
Errrm... venting for a power supply on THE TOP OF A RACKMOUNTED case, where the unit above will obviously obstruct the fan?? Hmmmm, I'm not sure this is well thought out.... not at all. Looks cool, but this should be a serious argument against a full size ATX power supply in there...
@@christianlempa No issues, it's always easier to play football from the viewing area than the field :D:D:D You can always count on the comment section to point out things :D Much love.
