Compute
Module infographic
22 / 69

Revision poster: the four EKS compute models.

Infographic: managed node groups, Karpenter, EKS Auto Mode and Fargate compared by what you manage, what AWS manages and what each is best for
Click the poster to enlarge it

Read each column from top to bottom: your job, AWS's job, and the best-fit situation.

  1. 1Managed node groups: you choose instance types, scaling and upgrades; AWS runs the Auto Scaling group and node lifecycle.
  2. 2Karpenter: you run it and write NodePools; it launches right-sized EC2 just in time and consolidates.
  3. 3Auto Mode: AWS runs nodes, scaling, patching and core add-ons; you run workloads.
  4. 4Fargate: one isolated microVM per Pod; no DaemonSets or hostNetwork.
  5. 5Harden any EC2 node: IMDSv2 required, minimal OS such as Bottlerocket.

Self-check: cover the poster and answer

  • Why can Karpenter pick instances that a node group cannot?
  • What must a Pod do to survive a Spot interruption?

Match the compute model to how much node operations you want to own.

BackNext