Hacker News Reader

Comment by eropple | Hacker News Reader

I’ve done “devops“ at both large and small organizations. I don’t write YAML. I use Pulumi and TypeScript for system provisioning; it’s all syntactically and type-checked as I write it, and reusable besides. I refuse to deal with the maintainability disaster that is Ansible (I used to use Chef but now pretty much everything can be handled via Fargate) and I don’t have a need for Kubernetes (except in my home lab where I use it with Pulumi’s providers).

The one place I could write YAML is AWS CodeBuild buildspecs. I write them in JSON, when they’re not being assembled on the fly from Pulumi.

You can do it too—just pick better tools for it.

View on HN

Seeing it mentioned here twice, I went and checked out their website/github. What I see is lots of 20-line examples setting up a docker-container/very basic vm/aws-lambda. While typescript might have better "testability" I'm not sure, where this goes if you just copy/execute bash-scripts or inject javascript code into aws-lambda. To me it seems like it just reinvents the "classic" sys-admin but instead of writing/copying/executing bash-scripts, dev(op)s are now churning out bash-scripts wrapped in tested wrapper code - well...

In contrast ansible has a lot of declarative building blocks (iirc all are unit-tested!), which either fail or do the specified job. And yeah, the specified job might be copying and templating a config-file for some oldschool service (and you might be shocked: people still use these!), which is inherently not really testable (except by integration tests. or maybe you write a test to get the file and to check whether you got to type the name right twice) - I'm not sure how pulumi can help you with that?

And yes, I would love a concise, descriptive DSL (compare for example spack, which does this nicely) over/as an alternative to the loops-crafted-on-YAML-mess of ansible but I take the latter any day over some "we-can-call-cloud-providers-apis-and-kubernetes-but-why-would-you-copy-files?"-stuff like pulumi.

I wouldn't be "surprised" that somebody wants better instance CM. I also don't care and I can't particularly fault the Pulumi folks for putting it on the back burner too; as each of the major instance CM providers are in the process of demonstrating, it doesn't make money and is being eaten by a container-centric and pervasively disposable approach.t I've stopped caring about machines enough to write even a systemd unit file and it's a better world to live in.

You can write instance CM yourself, I'd you want. Pulumi isn't a "cloud thing", it's a lifecycle management tool. But it's typed, and that absolves it of many, many sins given the incredibly positive surface it presents to a developer.

Interesting never heard of Pulumi, thanks

You might be interested in https://www.jetbrains.com/teamcity/

It has a Kotlin strongly typed DSL for pipelines, i absolutely despite Jenkins and Groovy, so many bugs causing huge issues due to weak typing and lack of testing :(

I’ve used Teamcity. It’s totally fine. I don’t want to host anything, though, and I want them inside my VPC. We’ll use AWS until we can’t stand it anymore and them we’ll fire up a Jenkins (just because there’s more internal expertise).

Reply on news.ycombinator.com