I think earthly demand the usage of docker. For a self hosted, fully customized CI flow I’d like to have the freedom of not using docker.
derek-zhou
sasajuric
Author of Elixir In Action
Yeah, this library will definitely support both. After all, it’s just a set of smaller abstractions which you can combine however you want 
1
dimitarvp
…oh by the way, do seriously consider supporting Hashicorp’s Nomad as well. I keep hearing that some orgs prefer it to k8s and I already heard a good amount of operations people praising it.
Have you used it?
0
hauleth
Dhall. It already have support for GitLab CI.
I was thinking about creating CI that would utilise systemd to run jobs. It could be fun.
I had, and I really liked it. It is so much simpler than k8s that it is weird that it is not used more. However I still think that in 90% of the deployments plain systemd would suffice.
1
sasajuric
Author of Elixir In Action
dimitarvp
Isn’t that the story of like 90% of IT?.. 
I would agree but then again cloud vendor lock-in pulls really hard in the other direction and the customers’ executives are buying it (figuratively and literally).
1
sasajuric
Author of Elixir In Action
It’s not just yaml, though I think that yaml is particularly horrible
IMO, yaml, xml, json, ini & co are meant to state facts. CI flow is not a collection of facts, but an imperative flow (run foo, then maybe bar, then baz&qux, …). Admittedly, in larger pipelines you might want turn it as a collection of high-level facts, so you can e.g. tell the forrest from the trees (or for any other purposes). But I think that in such case it’s probably better to roll your own DSL tailored to your particular needs. The available yamls & co are trying to solve everyone’s problems, which makes them overly generic and at the same time possibly insufficient.
1
hauleth
I was thinking about Elixir project that would work like CoreOS’ fleetctl, so it would be only management layer on top of other system that would handle the running the processes. And as systemd offers D-Bus API for that it would be a good choice, as it provides support for running process in isolated world and even support running containers via systemd-nspawn. So you would get a lot of great stuff without much work, just tackle the management on top of all of that.
1
dimitarvp
I am thinking that it’s trying to look like a declarative collection of steps and some instructions how to execute them but yeah, it’s very leaky and it doesn’t serve that function well.
I haven’t checked out Dhall just yet but I have found myself agreeing with former colleagues who used their own LISP DSLs to describe their deployment and testing pipelines. It almost did read like English and was very enjoyable to use.
But most orgs don’t do that, they want their people to hit the ground running in the first 3 days. Which is a sensible business goal but humans being humans they are of course overdoing it.
2
talentdeficit
Rebar3 Core Team
dhall is fantastic. it’s exactly the right layer of abstraction for all these ‘declarative but really wish i could use some loops’ tasks like configuration yaml and terraform modules
my concern with ci/cd in a language like elixir is that there’s way too many traps you can inadvertently fall into and end up with non deterministic builds (more so than the level of nondeterminism we already put up with)
0