13 ms·

Anecdata: Series B startup. I've found GKE to be almost completely painless, and I've been using it in production for more than 4 years now. I don't think the a

Anecdata: Series B startup. I've found GKE to be almost completely painless, and I've been using it in production for more than 4 years now. I don't think the article gave a fair representation on this count; sharing a link to a single GKE incident that (according to the writeup) spanned multiple days and only affected a small segment of users doesn't (for me) substantiate the claim that "it isn’t uncommon for them to have multi-hour outages". In my experience, multi-hour control-plane outages are very rare, and I've only had a single data-plane outage (in this case IIRC it was a GKE networking issue). Worst case I see is once or twice a year I'll get a node-level issue, of the level of severity where a single node will become unhealthy and the healthchecks don't catch it; most common side-effect is this can block pods from starting. These issues never translate into actual downtime, because your workloads are (if you're doing it right) spread across multiple nodes. I wouldn't be surprised if EKS is bad, they are three years behind GKE (GA released in 2018 vs 2015). EKS is a young service in the grand scheme of things, GKE is approaching "boring technology" at 5 yrs old.

whatsmyusername 6y ago
EKS is a feature parity product. The pricing makes that painfully obvious.
- theptip 6y ago
  Perhaps, in the box-ticking sense. I'd be surprised if they have the same operational maturity, which translates to uptime, which was what I was talking about. A quick search turns up these SLAs: <bad link> vs https://cloud.google.com/blog/products/containers-kubernetes/demand-more-from-your-cloud-provider-use-gke-advanced https://cloud.google.com/blog/products/containers-kubernetes... EKS: 3 nines GKE: 3.5 nines So Google is committing to 1/2 the downtime. (Note the GKE SLA is for regional clusters, which is what you should be doing if you care about uptime. The zonal cluster SLA is 2.5 nines. I couldn't find a difference in EKS, maybe there's an equivalent better SLA for regional clusters I couldn't find.) Edit - that EKS SLA link formatted weirdly, and so I did a little more digging and found a more recent SLA which matches GKE: https://aws.amazon.com/about-aws/whats-new/2020/03/amazon-eks-updates-level-agreement-to-99-95/ https://aws.amazon.com/about-aws/whats-new/2020/03/amazon-ek... So, per my original comment, I am surprised. (Having never used EKS directly I have no idea what their actual uptime is; in my experience GKE has been way higher than 3.5 nines, but obviously I don't have enough data to make statistically significant observations on this.)
  - AaronM 6y ago
    In the region we are running it, its not had an outage that has affected us since they launched it. We haven't had any issues with k8's upgrades either.
- AaronM 6y ago
  We've been using EKS since it went GA. We haven't had a single control plane outage that I am aware of.
  - whatsmyusername 6y ago
    I'm sure it works fine if you can get it running (documentation didn't work when I played with it). I'm referring more to the $200/month it was per control plane. That to me is a product they offer because someone else is offering it, but they don't want you to actually use it.
    - amerine 6y ago
      I take it the isolated $200/control plane is too expensive. What do you think the right price point is for that isolation?
      - whatsmyusername 6y ago
        ECS control planes cost nothing. If you're actually looking to build isolation in AWS then you're going to need EC2 dedicated for your EKS member hosts. So you're not getting isolation for $200/month (I'd have to spec it out but dedicated hosts are pricey to the point that it'd be competing with physical hardware in a colo). That $200/month for not-really-isolated is also per plane. So if you want a separate staging environment it's another $200/month. Client API sandbox? Same thing. It's wack.
- momokoko 6y ago
  AWS wants k8s to fail because it works against the significant lock-in AWS has tricked teams into building themselves into. They do not want people already in the AWS ecosystem to move to EKS, it is instead there to not lose potential customers. And that is why pricing and features sit right at “good enough” and not great.

whatsmyusername 6y ago

EKS is a feature parity product. The pricing makes that painfully obvious.

theptip 6y ago
Perhaps, in the box-ticking sense. I'd be surprised if they have the same operational maturity, which translates to uptime, which was what I was talking about. A quick search turns up these SLAs: <bad link> vs https://cloud.google.com/blog/products/containers-kubernetes/demand-more-from-your-cloud-provider-use-gke-advanced https://cloud.google.com/blog/products/containers-kubernetes... EKS: 3 nines GKE: 3.5 nines So Google is committing to 1/2 the downtime. (Note the GKE SLA is for regional clusters, which is what you should be doing if you care about uptime. The zonal cluster SLA is 2.5 nines. I couldn't find a difference in EKS, maybe there's an equivalent better SLA for regional clusters I couldn't find.) Edit - that EKS SLA link formatted weirdly, and so I did a little more digging and found a more recent SLA which matches GKE: https://aws.amazon.com/about-aws/whats-new/2020/03/amazon-eks-updates-level-agreement-to-99-95/ https://aws.amazon.com/about-aws/whats-new/2020/03/amazon-ek... So, per my original comment, I am surprised. (Having never used EKS directly I have no idea what their actual uptime is; in my experience GKE has been way higher than 3.5 nines, but obviously I don't have enough data to make statistically significant observations on this.)
- AaronM 6y ago
  In the region we are running it, its not had an outage that has affected us since they launched it. We haven't had any issues with k8's upgrades either.
AaronM 6y ago
We've been using EKS since it went GA. We haven't had a single control plane outage that I am aware of.
- whatsmyusername 6y ago
  I'm sure it works fine if you can get it running (documentation didn't work when I played with it). I'm referring more to the $200/month it was per control plane. That to me is a product they offer because someone else is offering it, but they don't want you to actually use it.
  - amerine 6y ago
    I take it the isolated $200/control plane is too expensive. What do you think the right price point is for that isolation?
    - whatsmyusername 6y ago
      ECS control planes cost nothing. If you're actually looking to build isolation in AWS then you're going to need EC2 dedicated for your EKS member hosts. So you're not getting isolation for $200/month (I'd have to spec it out but dedicated hosts are pricey to the point that it'd be competing with physical hardware in a colo). That $200/month for not-really-isolated is also per plane. So if you want a separate staging environment it's another $200/month. Client API sandbox? Same thing. It's wack.
momokoko 6y ago
AWS wants k8s to fail because it works against the significant lock-in AWS has tricked teams into building themselves into. They do not want people already in the AWS ecosystem to move to EKS, it is instead there to not lose potential customers. And that is why pricing and features sit right at “good enough” and not great.