Discuss: Deploying Production K8s Cluster to Baremetal/VMs guide/video

Hi all -

@wendell @lacion Hoping we can move the discussion here?

Goal

  • write up or step by step for k8s
  • video by wendell

next week, I will have my build finish with a similar setup

TR 2990wx 128gb of ram. 3TB NVME and 2TB SSD with 8TB spins

I will start writing on a guide to deploy a K8S cluster with that.

will probably modernize my old home lab setup https://github.com/lacion/k8s_homelab

@derekv will probably also be very interested in this.

Next stepsā€¦

  • Create master + slave ā€˜hostsā€™ in KVM/virsh in linux.

  • Ansible based playbooks to deploy production ready k8s cluster

  • Deploy demo microservices app + expose public access to static IP for ingress traffic.

  • git push remote master > k8s vm to check/link/composer install/whatever > new k8s VM pool and/or rsync changes to other pods or w/e.

Hereā€™s an example ā€˜appā€™ that we could test with potentially,

@lacion @wendell I would also be happy to run/test this deployment on my local setup and provide feedback.

there are already a gazillion tutorials videos, and documentation describing what your proposing there. i dont see any value in that whatsoever. and would be under utilizing the hardware @wendell was mentioning.

i think it will be a lot more interesting deploying kube with industry standard tools. and not writing any custom ansible for the sake of it.

configuring kube to properly work without a cloud provider and still leverage all its functionality.

a deploy a production worthy app that will load the cluster fully. i dig the addition of tesla cards he mention so maybe deploy some ML jobs that will leverage that.

1 Like

OK sounds good!

Agree. I did a 2 Day Docker/Kubernetes Workshop last week. We ran 4 VMā€™s and around 70 Containers on a Thinkpad T540 with 16G of RAM.
But iā€™m narrow minded. I couldnā€™t think of any workflow that would require that hardware.
For a customer we made Database Test with a mongodb cluster and a ā€œFind a location webfrontendā€ to proove we could sustain 4.5k Requests per Minute. That took 32 cores and no RAM to speak off. ML certainly would be interesting.

None the less, iā€™d be highly interested in a Video concerning Kubernetes as iā€™m just learning all of that and will have to do some work with it sooner rather than later.

1 Like

an ML workload would be cool using v100s, also some kind of image processing or video streaming could use that kind of hardware.

possible workloads
Computer Vision, Machine Learning.
Video Processing, Streaming.
Photogrammetry.

are the workloads I can think of that would make a very interesting test.

1 Like