Cost cutting at KubeCon
Cost optimization in kubernetes is harder than in the general cloud - this has long been an issue in finops. There was a panel at Kubecon Europe that discussed modern ways of cutting costs that was pretty interesting, in particular the continuous profiling tools that are exposed to the development teams.
It’s not enough to say “make sure your app is performant” - there has to be a feedback mechanism to tell the engineering teams how they are doing. Feedback loops are the foundation of all things finops.
That one time when AWS managed services was actually a better solution…
It’s actually pretty frequent, especially when you don’t have the in-house knowledge/bandwidth to run your own k8s cluster.
Infrastructure from code
Klotho looks interesting. Whereas Pulumi uses Terraform under the hood Klotho has their own underlying provider tooling, but the most interesting part is that instead of taking a CDK and building a DSL out of it, this appears to be more CDK-ish than CDK.
Dev and Ops is Breaking Up
First off, let’s level set on some definitions of devops: there are a ton of them and they vary depending on who you ask. I’ll use the common and rather boring definition of
The relationship between development and operational groups that enables faster and more reliable development using a variety of tools and practices but especially communication.
Devops has the foundation of being a relationship paradigm, much like finops.
Over time (through the efforts of recruiters, C-suites, newbies, and charlatans) devops has taken on new definitions including being equated to titles instead of practices (SRE, platform teams, CICD engineers, sysadmins, the person who got stuck writing Cloudformation this sprint, etc). This notion is largely ignored within most engineering circles by now, which is a relief, but still exists nonetheless.
This relationship paradigm has seen some mixed results: some teams have fully leaned into it while others have maintained their silos. Some call themselves one thing but actually do another. Thus devops has turned into a 4-letter word in the past few years. It also doesn’t help that now everything has an -ops.
But now it’s breaking up
The YAML hate train
It’s easy to get the sense that developers hate YAML, and by extension any domain specific language that isn’t a proper “programming language” - even things like Terraform. That’s not to say that those ecosystems are panaceas, but there are things that can be done to ease developer frustrations: having the right tools in place for your team given the size of your team.
Why tooling matters
There’s some hyperbole here, but you get the idea.
Your three person team cannot manage kubernetes and build your product and your engineers will quit
Simple scripts will ruin your 6-product monorepo and your engineers will quit
Terraform gets unwieldy in giant monorepos if you don’t do things correctly up front - take time to think about how to structure your project for the future
Cloudformation is fine for small projects, and small projects only
AWS CDK will require at least one dedicated person to build and manage, quite unlike other CDK-types like Pulumi and CDK-Terraform, because at the end of the day it’s still cloudformation
Distancing devs from operations is still a bad idea so let them pick the tools
The FOCUS spec
AWS revenues are slowing down
Some follow up to one of the observations made in the last post - AWS revenues are starting to slow down. Much of the reason seems to be centered around the idea that AWS is lowering their margins. This is an interesting thought but I don’t think it’s correct - it doesn’t seem to line up directly with their stock price, which for years has been mostly influenced by AWS and not the rest of Amazon.
However there are some notable companies (at least in the tech world) that are moving off of AWS and back to the data center. Amazon’s own Prime Video service went back to old “boring” style architectures instead of that new-fangled cloud stuff. Then there’s an underlying “reconsideration” of many companies as a whole. Readers of HackerNews will note that there’s an underlying distaste of AWS as a whole if you read the comments. Either way AWS doesn’t have an heir apparent so it’s not going away any time soon.