What I learned reading the 2020 Datadog Container Report

The 2020 Container Report by Datadog is full of interesting insights about how their customers use container technologies!

As you read it, keep in mind the selection bias: data comes from Datadog customers only. While there is a free offering available, it is rather limited. Paying customers are committed enough to pay $15 or $23 per host/node. The data set is for some insights divided into "smaller" and "larger" deployments, where 500 nodes is the threshold. So be mindful that this data set probably relates mostly to enterprise customers.

Still, the data set is interesting. Some highlights that stood out to me:

While 90% of of containerized workloads are orchestrated, only half are done so by by Kubernetes.

First of all, my heart goes out to the admins of the remaining 10%. I'm here if you need a shoulder to cry on. Second, this shows something about the data set, because the 40% non-Kubernetes workloads indicates customers jumping on this whole containerization ride before Kubernetes was a thing, and then sticking to whichever orchestrator was cool at that time. Mesos, perhaps? Does Docker Swarm count?

In Kubernetes workloads, 40-something% of containers use less than 30% of the requested CPU and RAM.

Capacity management is hard. You need to be a good sport about making CPU requests so the scheduler can squeeze your container in, but you also don't want to underprovision so that your performance suffers. Key lesson: the Vertical Pod Autoscaler can probably help save a lot of money!

Public cloud customers really want managed Kubernetes services

Just look at this graph from the report:

Looking at this, it tells me that AWS customers started running Kubernetes way before EKS was a thing, and that they have then just kept going. I do not read it as EKS being so bad that AWS customers rather self-manage than use it, although the marketing department at the other cloud providers could probably spin it that way.

The most popular Kubernetes version is 1.15, which is 17 months old now and therefore lacks patches since June 2020

Come on, people! Kubernetes has all kinds of security issues (91 in this database), and you really need to keep up. Patches only come out for about a year nowadays (up from 9 months previously), and you are most definitely at-risk if you run something ancient. Be kind to yourself and upgrade now. It's not getting easier as you fall more versions behind.

Looking at the previous insight, it might be worth pointing out that "Managed Kubernetes" may sound enticing, but "managed" does not mean set-it-and-forget-it. It still needs to be kept up to date.

Concluding remarks

If you as a cloud provider want to cater to the Datadog customer crowd, you need to offer a managed Kubernetes service. You should keep it up to date for them. And you should help them right-size their deployments, so that they can save truckloads of money.

Now for my bias. If you want help with any of this stuff, either as a cloud provider or as a customer to one, get in touch with us at Elastisys if you want us to manage your Kubernetes platform for you.