Skip to content

Latest commit

 

History

History
236 lines (138 loc) · 4.83 KB

2020-06-26-kainos-keep-your-k8s-healthy-patched-secure.md

File metadata and controls

236 lines (138 loc) · 4.83 KB
Raw
title theme revealOptions
How to keep your k8s workers healthy, patched and secure without any impact on hosted applications.
solarized
slideNumber
true

How to keep your k8s workers healthy, patched and secure without any impact on hosted applications

#agenda

  • glossary
  • worker lifecycle
  • be smart about your worker OS
  • tools to make your life easier
  • Q&A

#whoAmI

Michal Schott

  • WebOps Engineer
  • Street Manager
  • k8s or die!

https://schottlabs.eu/resume/

#glossary

  • k8s - kubernetes

  • k8s masterplane - set of VMs to control state of k8s cluster

  • k8s workers - these are running your containers, regular VM with OS attached to k8s masterplane

  • kubectl - ssh replacement

Note:

  • k8s workers - regular VM with OS attached to k8s masterplane

img

  • pod - collection of containers

  • endpoint - IP address of single pod

  • service - load balanced collection of endpoints

  • PDB - pod disruption budged

Note:

  • service - (round robin) dedicated single IP to access your containers
  • PDB - how many pods needs to stay in healthy state during maintenance

  • node is healthy - no kernel deadlocks, kubelet is ready...

  • patched - OS is up-to-date with all patches available

  • secure - attack vector is as small as possible

  • impact on hosted application - end user do not notice downtime

  • node unscheduleable - k8s scheduler can NOT assign pods to node // kubectl cordon

  • node scheduleable - k8s scheduler can assign pods to node // kubectl uncordon

  • eviction - pod termination event, usually happens when you want to relocate workload to different worker // kubectl drain

#workerLifecycle

  • Worker VMs runs kubelet

  • kubelet contacts k8s-apiserver, presents identity and request for client certificate

  • k8s-apiserver validates identity and returns certificate to the client

  • kubelet starts any further communication using client certificate // mTLS

  • hearthbeat, node lease objects in kube-node-lease

  • kubectl get/describe node

Note:

kubectl -n kube-node-lease get lease

Each Node has an associated Lease object in the kube-node-lease namespace. An abstraction used by Kubernetes to support multiple virtual clusters on the same physical cluster.

Lease is a lightweight resource, which improves the performance of the node heartbeats as the cluster scales.

Lets shutdown a node in a dirty way! // demo

Note: terminate aws instance

show node is still there in k8s store

it will disappear after a while

  • containers had no time for graceful shutdown
  • possible service disruption - how to prevent this?

Note: Any stateful workload could be interrupted, which might lead to data inconsistency

You cant fully prevent such situations but you can do it way better

Lets shutdown a node in a clean way! // demo

Note: kubectl cordon

kubectl drain

reboot

kubectl uncordon

  • containers were gracefully terminated
  • k8s-scheduler was able to schedule new replicas on different nodes
  • 100% service uptime

#workerOS

  • k8 supports both Linux and Windows
  • ubuntu? rhel?
  • what else have we got?

meet Container Optimized Linux

Cloud Providers Solutions

  • AWS - Amazon Linux
  • GCP - Container-Optimized OS
  • Azure - ?

Universal / made by geeks

  • RHEL / CentOS / Fedora Atomic Host
  • Ubuntu have most recent and updated kernel (first distro with 5.X)
  • Core CoreOS
  • RHEL/Fedora CoreOS
  • Kinvolks FlatCar Linux

CORE

They've build:

rkt // etcd // quay

clair // dex // flannel

ignition // tectonic

AND A LOT MORE

Recently they were acquired by Red Hat (and by IBM)

Kinvolks Flatcar Linux continues CoreOS legacy

  • reduce attack surface
  • immutable file system
  • driven by principle of least privilege
  • automatic updates
  • secure container runtime

img

#makeYourLifeEasier

  • pick container-optimised linux flavour
  • kured
  • coreos/container-linux-update-operator
  • kinvolk/flatcar-linux-update-operator
  • node-problem-detector
  • node-local-dns-cache
  • k8s-spot-termination-handler
  • cluster-autoscaler!
  • overprovision nodes with cluster-proportional-autoscaler!

Q&A