Hey everyone, I'm currently using Karpenter for managing our clusters, but our team is small, and we're finding it tough to keep up with the overhead involved in upgrading multiple clusters across teams. I'm considering switching to EKS Autopilot and wanted to know if anyone has experience with both. How has Autopilot worked for you? Any pros and cons I should be aware of?
5 Answers
I recently looked into this too! EKS Autopilot does come with some limitations since it abstracts away a lot of control, which might not suit everyone. But for smaller teams, it definitely minimizes the maintenance hassle. Just keep in mind you lose some flexibility in configuration. So, if you are okay with that trade-off for easier management, it could work.
I've played around with both, and honestly, the pricing is pretty significant—especially for larger setups. Autopilot adds a 12% markup on node prices, which doubles to about 24% for spot instances. If you have small clusters, it might be worth it for speedy deployment, but established teams might not find it justifies the cost unless they need that hassle-free service. You just have to evaluate what your team really needs.
That feels steep! Are there specific features that make it worth it, or could I just stick with Karpenter?
Just a heads up, EKS Autopilot actually uses Karpenter behind the scenes! So, you essentially get a managed version of Karpenter where you don't need to handle NodeGroups or configure Node Pools yourself. However, if you rely on custom instance templates, you might hit a wall because you can't define those anymore in Autopilot. It's way less work than managing everything yourself, but it may not cover specific customization needs you have. For smaller teams struggling with upgrades, it can be a lifesaver, especially if you have a lot of addons to manage. But don't expect it to be a magic fix for all current issues!
Sounds promising! But does it handle logs or monitoring well?
Yeah, you can't see the Karpenter logs directly, but you can track NodeClaim events. It's not perfect, but it's a start for observability.
We considered moving to EKS Autopilot, but found it too pricey for what we needed. We had already set up our own solutions that covered the same issues. If we were just starting out, it might be easier to justify the cost. But it seems like if you're already running a mature setup, Karpenter on its own might still be the way to go.
We recently transitioned from vanilla Karpenter to the managed version in EKS Autopilot and honestly, it has been smooth sailing. It's one less thing to maintain, and we manage around 40 clusters across 20 accounts! Sure, I noticed a cost increase, but I'd say it was worth it for the reduction in complexity.
Was the cost increase around 10%?
What specific control are you giving up, though? I thought they still support custom node pools?