this post was submitted on 10 Feb 2024
2 points (100.0% liked)

DevOps

1738 readers
1 users here now

DevOps integrates and automates the work of software development (Dev) and IT operations (Ops) as a means for improving and shortening the systems development life cycle.

Rules:

Icon base by Lorc under CC BY 3.0 with modifications to add a gradient

founded 2 years ago
MODERATORS
top 15 comments
sorted by: hot top controversial new old
[–] [email protected] 0 points 1 year ago

I'm not devops but this seemed reasonable from what I've seen.

Though having infrastructure weigh in with "go should be your language of choice" seems weird.

[–] [email protected] 0 points 1 year ago

This is all incredibly valuable information.

[–] [email protected] 0 points 1 year ago (1 children)

This is excellent. I may copy the rough format for tracking things internally at my company!

Btw, I agree with most of your decisions in here with just a few exceptions.

  • kustomize > helm
  • Argo > flux

My last thought is less clear though. There are good observability solutions besides datadog. Grafana Cloud is great. Honeycomb has a similar offering. But all are pretty expensive though.

If you aren’t using OpenTelemetry, you’re probably doing observability wrong!

[–] [email protected] 0 points 1 year ago (1 children)

I've only used helm and hadn't considered kustomize as an equivalent, what about kustomize makes it bette in your opinion?

[–] [email protected] 0 points 1 year ago

First is complexity. A simple helm chart works great but more elaborate charts can turn into a maintenance problem. This is especially when managing a large number of apps and need to establish and maintain standards across them. E.g. you want to add a new label to every helm chart you use. You now get to making 60 PRs for 60 charts. Or you can tie them all together with chart dependencies. This can be done well but almost never is. It's just too easy to build a bad helm chart. Kustomize allows you to do this from a "top-down" perspective

Second is modifications. Consider as an example that you want to run filebeat as a sidecar container on some pod to capture its logs. But the helm chart you're using doesn't include this feature. You have two choices: modify the pod when it's created with a mutatingwebhook or similar (super complicated solution) or you can copy/fork the chart, add the functionality, and maintain it going forward. Kustomize just doesn't have this problem. You can just modify a base manifest with overlays.

Last is the nature of Go templates which helm charts are based on. Everything outside of {{ }} is just plaintext. This leads to a ton of limitations. Got a whitespace issue? You'll probably find out at runtime. Want your IDE to identify syntax issues, provide, intellisense, etc. on the final manifest? Good luck! You need to render that chart first. With Kustomize, every manifest is structured text (yaml). So you get the benefits of all standard tooling for yaml data in your IDEs and CI/CD pipelines.

Honestly, I could keep going (helm releases ugghhhh!). But helm definitely wins on one point and it's a big one; Helm is the standard for distributing k8s manifests. So every meaningful project supplies helm charts. Kustomize doesn't even come close on this one. That said, I think Kustomize manifests are just simpler to build. So having an official base manifest for every project just doesn't matter too much.

[–] [email protected] 0 points 1 year ago

A good read, and interesting to see what services to consider.

[–] [email protected] 0 points 1 year ago (1 children)

I can't believe I haven't seen external secrets before. Sealed secrets are cool, but such a pain as you described. Gonna be setting up external secrets next week sounds like. Thanks for the great post

[–] [email protected] 0 points 1 year ago (1 children)

What do you think about storing your encrypted secrets in your repos using Sops?

[–] [email protected] 0 points 1 year ago (1 children)

I prefer Sealed Secrets over sops since it has the namespace scoping element and can also be stored in repo (once encrypted). I also generally prefer having a controller deployed rather than forcing devs to learn kustomize (which we don't widely use yet) so I guess less of a support burden for me.

[–] [email protected] 0 points 1 year ago (1 children)

I understand your point. Anyway, if your devs are using Helm they can still use Sops with the helm-secrets plugin. Just create a separated values file (can be named as secrets.yaml) contaning all sensitive values and encrypt it with Sops.

[–] [email protected] 0 points 1 year ago (1 children)

Thanks for sharing! I definitely hadn't seen that plugin. We definitely use helm, even though I hate it lol. I will take a look when I get around to looking at external secrets since I still haven't had a chance to (you know how it goes... priorities made up by some random PM or whatever)

[–] [email protected] 0 points 1 year ago

If you still want more you can use Helmfile. Take care of your PMs 😁

[–] [email protected] 0 points 1 year ago

Even for someone who’s only dabbled in infrastructure, this was a fun read.

[–] [email protected] 0 points 1 year ago

If I could be bothered to sit and write down a distilled version of my last decade at work it would be something very similar. Any junior SRE can benefit from this.

[–] [email protected] 0 points 1 year ago

Super useful write up. I've taken some notes I'll use moving forward on a personal project. Thanks for sharing