quick-coat-37875
11/13/2022, 7:36 PMkubectl get nodes
returns the master and slave correctly
Now, I just finished installing rancher-desktop into an OpenSUSE VM
On first run, I get the welcome screen, but I am unsure what exactly the differences are between the options "containerd - use with nerdctl" and "dockerd (moby) - use with Docker CLI and k3d"
and have been unable to find information a newbie like me can readily understand. can someone tell me which to choose, and a short description of why?bulky-sunset-52084
11/13/2022, 8:10 PMquick-coat-37875
11/13/2022, 8:12 PMwitty-jelly-95845
11/13/2022, 9:07 PMquick-coat-37875
11/13/2022, 9:26 PMbulky-sunset-52084
11/13/2022, 9:40 PMTo run K3s in this mode, you must have an odd number of server nodes. We recommend starting with three nodes
This is because etcd does a process called
leader-election
so the servers need a quorum
. You can see the HA matrix and more info from the etcd docs:
https://etcd.io/docs/v3.5/faq/#why-an-odd-number-of-cluster-members
This was a large area of confusion for me when I started out so hopefully this will save you some trouble.quick-coat-37875
11/13/2022, 9:55 PMbulky-sunset-52084
11/13/2022, 10:11 PMCould this be done using rancher in vm on my laptop and then the 2 Pis? Or I should get a 3rd PiThe issue with doing it this way is your laptop have a very different CPU then a raspberry pi and Kubernetes doesn't support this layout very well. it would be much better (and save you a big headache) to just get another pi. But also, a word of note is that you can totally do a non-HA install of k3s on a single raspberry pi that you can install Rancher and use the other pi as another 'downstream cluster' for rancher to manage (but only do this if you are just playing learning, if it's production critical make sure your set up is Highly Available.
quick-coat-37875
11/13/2022, 10:24 PMbulky-sunset-52084
11/13/2022, 10:30 PMSo for non-HA, I don't have to work about the even/odd database thing for nowcorrect! 1 is an odd number 🙂
but when I expand the system, I still have to add 2 each time to keep it odd from then on?yes but at a certain point its not worth adding more master modes, you shouldn't need more than 3 master nodes for most setups. cons outweigh the pros at 5 or 7 because the additional network traffic from all the syncing can really bring things too a crawl on a typical network. I want to not that this odd node requirement is ONLY for the master nodes since etcd typically only runs on master nodes. if you add worker nodes - you can have any number of those as you want.
I still follow the Helm Method?I would highly recommend it.
RMC/Rancher Management cluster
- the cluster that runs Rancher as an app inside. it's purpose is to manage all other kubernetes clusters that run your actual application workloads. We call your application specific clusters:
Downstream clusters
- your application clusters. These are connected to and managed by rancher but you will run your applications on them.
Kubernetes Terms:
master/control-plane
- the server(s) that run core kubernetes functions specifically the API server which is the glue that holds it all together and is what responds to your kubectl commands
etcd
- a database used to store kubernetes data like your application deployment manifests. This database typically is run on the master nodes.
Worker Nodes
- Nodes to run none kubernetes workloads such as your applications and even Rancher itself will run on worker nodes.
When you do a single node install you are giving that one pi ALL of the kubernetes roles such as master, etcd and worker.