Deployah – deploy to Kubernetes from a short spec, no Helm, nothing in-cluster

Post by atkrad » Wed, Aug 05, 2026, 7:21 PM UTC

verdverm wrote: post #6

gopherly wrote: post #4

How do you deploy apps to Kubernetes without writing Helm charts? I mean the middle ground: not hand-writing charts/YAML, and not installing an in-cluster PaaS either. What do people actually use for that, especially CLI-only?

I'm using CUE (config) + Helm (reconsiliation) This post is also using helm, putting a DSL in front because yaml sucks and templated yaml is insane. The problem here is the DSL, devs don't want a DSL for k8s ime, it becomes yet another abstraction of complexity. I render full k8s yaml for helm and avoid the templates

Post by atkrad » Wed, Aug 05, 2026, 7:23 PM UTC

verdverm wrote: post #6

gopherly wrote: post #4

How do you deploy apps to Kubernetes without writing Helm charts? I mean the middle ground: not hand-writing charts/YAML, and not installing an in-cluster PaaS either. What do people actually use for that, especially CLI-only?

I'm using CUE (config) + Helm (reconsiliation) This post is also using helm, putting a DSL in front because yaml sucks and templated yaml is insane. The problem here is the DSL, devs don't want a DSL for k8s ime, it becomes yet another abstraction of complexity. I render full k8s yaml for helm and avoid the templates

Post by atkrad » Wed, Aug 05, 2026, 8:01 PM UTC

verdverm wrote: post #6

gopherly wrote: post #4

How do you deploy apps to Kubernetes without writing Helm charts? I mean the middle ground: not hand-writing charts/YAML, and not installing an in-cluster PaaS either. What do people actually use for that, especially CLI-only?

I'm using CUE (config) + Helm (reconsiliation) This post is also using helm, putting a DSL in front because yaml sucks and templated yaml is insane. The problem here is the DSL, devs don't want a DSL for k8s ime, it becomes yet another abstraction of complexity. I render full k8s yaml for helm and avoid the templates

Post by atkrad » Thu, Aug 06, 2026, 9:44 AM UTC

znpy wrote: post #17

atkrad wrote: post #10

Earlier quoted context omitted.

Fair points. A few clarifications though. On cluster access: Deployah does not ask you to hand developers cluster-admin on prod. It is a client-side CLI, same class as helm/kubectl. Whoever runs it needs whatever RBAC you already give that identity. The intended split is deployah.yaml for the app, and deployah.platform.yaml for contexts, domains, TLS, profiles. Platform/ops can own that second file, and prod deploys…

The use case you describe (small team) can just use a shared helm chart, customised through values.yaml to run whatever image is needed. This (deployah) add a very large layer of complexity. It’s a no for me.

Post by verdverm » Thu, Aug 06, 2026, 7:20 PM UTC

mring33621 wrote: post #19

verdverm wrote: post #9

Earlier quoted context omitted.

likewise! I cannot understand the field of devops willingness not only to endure this, but find double templated yaml (Argo+helm) appealing. So many of my peers prefer to do what everyone else is doing and slog through the pain.

I was recently scolded for telling my coworkers that they are 'nose blind' to the mess that is our CI/CD 'pipeline'. Maven, docker buildx, jenkins, github actions, helm, ArgoCD, Tekton, Terraform It's an often-broken Franken-Goldberg machine