Typecheck, template and modularize your Kubernetes definitions with Dhall
Go to file
Gabriel Gonzalez 2ed2ffd073
Improve user experience for generating files (#165)
This improves the documentation and scripts responsible for
generating files.  The main changes from the user's point of view
are:

* The `generate` script now supports four subcommands:

  * `generate kubernetes` - Generate all Kubernetes bindings
  * `generate examples` - Generate examples for the preferred Kubernetes version
  * `generate readme` - Generate the README from the Dhall template
  * `generate preferred` - Generate top-level files that re-export the preferred
    Kubernetes version

  The `generate` script without arguments generates everything

* Generation should be faster and more incremental now

  For example, regenerating the `README` by itself should go more quickly since
  it no longer requires regenerating any examples or Kubernetes bindings if
  they haven't changed

This required overhauling the internal Nix code for generating these files
to support more granular builds.

Another consequence of this change is that the preferred Kubernetes
version no longer packages `Prelude.dhall`, `README.dhall` or the
`examples`.  Those are now only stored within the top-level directory.
2021-03-23 11:32:34 -07:00
1.12 Update dhall-openapi and regenerate (#162) 2021-01-17 12:21:24 +00:00
1.13 Update dhall-openapi and regenerate (#162) 2021-01-17 12:21:24 +00:00
1.14 Update dhall-openapi and regenerate (#162) 2021-01-17 12:21:24 +00:00
1.15 Update dhall-openapi and regenerate (#162) 2021-01-17 12:21:24 +00:00
1.16 Update dhall-openapi and regenerate (#162) 2021-01-17 12:21:24 +00:00
1.17 Improve user experience for generating files (#165) 2021-03-23 11:32:34 -07:00
1.18 Update dhall-openapi and regenerate (#162) 2021-01-17 12:21:24 +00:00
1.19 Update dhall-openapi and regenerate (#162) 2021-01-17 12:21:24 +00:00
docs Improve user experience for generating files (#165) 2021-03-23 11:32:34 -07:00
examples Update top-level README and examples (#164) 2021-03-18 11:25:57 -07:00
logo Add logo (#58) 2019-03-31 15:24:48 +02:00
nix Improve user experience for generating files (#165) 2021-03-23 11:32:34 -07:00
scripts Improve user experience for generating files (#165) 2021-03-23 11:32:34 -07:00
.gitignore Implement generator from CustomResourceDefinition (#98) 2020-01-10 06:22:41 -08:00
defaults.dhall Generate Dhall schemas for multiple Kubernetes versions (#104) 2020-01-12 15:00:56 -08:00
LICENSE Create LICENSE 2018-05-27 19:55:07 +02:00
package.dhall Generate Dhall schemas for multiple Kubernetes versions (#104) 2020-01-12 15:00:56 -08:00
Prelude.dhall Prelude: update safety-guarantees documentation link (#149) 2020-10-01 09:17:20 -07:00
README.md Improve user experience for generating files (#165) 2021-03-23 11:32:34 -07:00
release.nix Improve user experience for generating files (#165) 2021-03-23 11:32:34 -07:00
schemas.dhall Generate Dhall schemas for multiple Kubernetes versions (#104) 2020-01-12 15:00:56 -08:00
types.dhall Generate Dhall schemas for multiple Kubernetes versions (#104) 2020-01-12 15:00:56 -08:00
typesUnion.dhall Generate Dhall schemas for multiple Kubernetes versions (#104) 2020-01-12 15:00:56 -08:00

dhall-kubernetes

dhall-kubernetes logo

dhall-kubernetes contains Dhall bindings to Kubernetes, so you can generate Kubernetes objects definitions from Dhall expressions. This will let you easily typecheck, template and modularize your Kubernetes definitions.

Why do I need this

Once you build a slightly non-trivial Kubernetes setup, with many objects floating around, you'll encounter several issues:

  1. Writing the definitions in YAML is really verbose, and the actually important things don't stand out that much
  2. Ok I have a bunch of objects that'll need to be configured together, how do I share data?
  3. I'd like to reuse an object for different environments, but I cannot make it parametric..
  4. In general, I'd really love to reuse parts of some definitions in other definitions
  5. Oh no, I typoed a key and I had to wait until I pushed to the cluster to get an error back :(

The natural tendency is to reach for a templating language + a programming language to orchestrate that + some more configuration for it... But this is just really messy (been there), and we can do better.

Dhall solves all of this, being a programming language with builtin templating, all while being non-Turing complete, strongly typed and strongly normalizing (i.e.: reduces everything to a normal form, no matter how much abstraction you build), so saving you from the "oh-noes-I-made-my-config-in-code-and-now-its-too-abstract" nightmare.

For a Dhall Tutorial, see the website, or the readme of the project, or the full tutorial.

Prerequisites

NOTE: dhall-kubernetes requires at least version 1.27.0 of the interpreter (version 11.0.0 of the language).

Quickstart - a simple Deployment

Let's say we'd like to configure a Deployment exposing an nginx webserver.

In the following example, we:

  1. Import the Kubernetes definitions as a Dhall package (the package.dhall file) from the local repo. In your case you will want to replace the local path with a remote one, e.g. https://raw.githubusercontent.com/dhall-lang/dhall-kubernetes/master/package.dhall Note: the sha256:.. is applied to some imports so that:
    1. the import is cached locally after the first evaluation, with great time savings (and avoiding network calls)
    2. prevent execution if the content of the file changes. This is a security feature, and you can read more in Dhall's "Security Guarantees" document Note: instead of using the package.dhall from the master branch, you may want to use a tagged release, as the contents of the master branch are liable to change without warning.
  2. Define the Deployment using the schema pattern and hardcoding the deployment details:
-- examples/deploymentSimple.dhall

let kubernetes =
      https://raw.githubusercontent.com/dhall-lang/dhall-kubernetes/master/package.dhall sha256:532e110f424ea8a9f960a13b2ca54779ddcac5d5aa531f86d82f41f8f18d7ef1

let deployment =
      kubernetes.Deployment::{
      , metadata = kubernetes.ObjectMeta::{ name = Some "nginx" }
      , spec = Some kubernetes.DeploymentSpec::{
        , selector = kubernetes.LabelSelector::{
          , matchLabels = Some (toMap { name = "nginx" })
          }
        , replicas = Some +2
        , template = kubernetes.PodTemplateSpec::{
          , metadata = Some kubernetes.ObjectMeta::{ name = Some "nginx" }
          , spec = Some kubernetes.PodSpec::{
            , containers =
              [ kubernetes.Container::{
                , name = "nginx"
                , image = Some "nginx:1.15.3"
                , ports = Some
                  [ kubernetes.ContainerPort::{ containerPort = +80 } ]
                }
              ]
            }
          }
        }
      }

in  deployment

We then run this through dhall-to-yaml to generate our Kubernetes definition:

dhall-to-yaml <<< ./examples/deploymentSimple.dhall

And we get:

## examples/out/deploymentSimple.yaml

apiVersion: apps/v1
kind: Deployment
metadata:
  name: nginx
spec:
  replicas: 2
  selector:
    matchLabels:
      name: nginx
  template:
    metadata:
      name: nginx
    spec:
      containers:
        - image: nginx:1.15.3
          name: nginx
          ports:
            - containerPort: 80

More modular: defining an Ingress

The above is cool, but hardcoding data is not that cool.

So in a more realistic deployment you'll probably want to define:

  • some MyService type that contains the config settings relevant to your deployments
  • some functions parametrized by this type, so that you can produce objects to send to k8s by just applying these functions to MyService objects

This is useful because then you can define your Services separately from the Kubernetes logic, and reuse those objects for configuring other things (e.g. configuring the services themselves, templating documentation, configuring Terraform deployments, you name it).

As an example of that, next we'll define an Ingress (an Nginx Ingress in this case), containing stuff like TLS certs and routes for every service - see the schema.

Things to note in the following example:

  • we define the Service type inline in the file, but in your case you'll want to have a separate ./Service.dhall file (so you can share around the project)
  • we define functions to create the TLS definitions and the routes, so that we can map them over the list of services.
  • we also defined the list of services inline, but you should instead return the mkIngress function instead of applying it, so you can do something like dhall-to-yaml <<< "./mkIngress.dhall ./myServices.dhall"
-- examples/ingress.dhall

let Prelude =
      ../Prelude.dhall sha256:10db3c919c25e9046833df897a8ffe2701dc390fa0893d958c3430524be5a43e

let map = Prelude.List.map

let kubernetes =
      https://raw.githubusercontent.com/dhall-lang/dhall-kubernetes/master/package.dhall sha256:532e110f424ea8a9f960a13b2ca54779ddcac5d5aa531f86d82f41f8f18d7ef1

let Service = { name : Text, host : Text, version : Text }

let services = [ { name = "foo", host = "foo.example.com", version = "2.3" } ]

let makeTLS
    : Service → kubernetes.IngressTLS.Type
    = λ(service : Service) →
        { hosts = Some [ service.host ]
        , secretName = Some "${service.name}-certificate"
        }

let makeRule
    : Service → kubernetes.IngressRule.Type
    = λ(service : Service) →
        { host = Some service.host
        , http = Some
          { paths =
            [ { backend =
                { serviceName = service.name
                , servicePort = kubernetes.IntOrString.Int +80
                }
              , path = None Text
              }
            ]
          }
        }

let mkIngress
    : List Service → kubernetes.Ingress.Type
    = λ(inputServices : List Service) →
        let annotations =
              toMap
                { `kubernetes.io/ingress.class` = "nginx"
                , `kubernetes.io/ingress.allow-http` = "false"
                }

        let defaultService =
              { name = "default"
              , host = "default.example.com"
              , version = " 1.0"
              }

        let ingressServices = inputServices # [ defaultService ]

        let spec =
              kubernetes.IngressSpec::{
              , tls = Some
                  ( map
                      Service
                      kubernetes.IngressTLS.Type
                      makeTLS
                      ingressServices
                  )
              , rules = Some
                  ( map
                      Service
                      kubernetes.IngressRule.Type
                      makeRule
                      ingressServices
                  )
              }

        in  kubernetes.Ingress::{
            , metadata = kubernetes.ObjectMeta::{
              , name = Some "nginx"
              , annotations = Some annotations
              }
            , spec = Some spec
            }

in  mkIngress services

As before we get the yaml out by running:

dhall-to-yaml <<< ./examples/ingress.dhall

Result:

## examples/out/ingress.yaml

apiVersion: networking.k8s.io/v1beta1
kind: Ingress
metadata:
  annotations:
    kubernetes.io/ingress.allow-http: 'false'
    kubernetes.io/ingress.class: nginx
  name: nginx
spec:
  rules:
    - host: foo.example.com
      http:
        paths:
          - backend:
              serviceName: foo
              servicePort: 80
    - host: default.example.com
      http:
        paths:
          - backend:
              serviceName: default
              servicePort: 80
  tls:
    - hosts:
        - foo.example.com
      secretName: foo-certificate
    - hosts:
        - default.example.com
      secretName: default-certificate

FAQ

Can I generate a YAML file with many objects in it?

It is usual for k8s YAML files to include multiple objects separated by --- ("documents" in YAML lingo), so you might want to do it too.

If the objects have the same type, this is very easy: you return a Dhall list containing the objects, and use the --documents flag, e.g.:

dhall-to-yaml --documents <<< "let a = ./examples/deploymentSimple.dhall in [a, a]"

If the objects are of different type, it's not possible to have separate documents in the same YAML file.
However, since k8s has a builtin List type for these cases, it's possible to use it together with the union type of all k8s types that we generate.

So if we want to deploy e.g. a Deployment and a Service together, we can do:

let k8s = ./typesUnion.dhall

in 
{ apiVersion = "v1"
, kind = "List"
, items = 
  [ k8s.Deployment ./my-deployment.dhall
  , k8s.Service ./my-service.dhall
  ]
}

Projects Using dhall-kubernetes

Development

You will need to install Nix in order to run the file-generation scripts provided by this repository. You can obtain Nix by following the instructions here:

Updating the README.md

The top-level README.md is generated from ./docs/README.md.dhall so that the examples within the ./examples directory stay in sync with the README.md. That means that in order to update the README.md you need to first edit ./docs/README.md.dhall and then run:

$ ./scripts/generate readme

Updating the examples

If you want to author new examples, add them to the ./examples directory and run:

$ ./scripts/generate examples

… which will freeze and type-check each example and generate the matching YAML output.

The ./examples directory is only built against one version of the Kubernetes API (the "preferred" version). To change the preferred version, run:

$ echo "${VERSION}" > ./nix/preferred.txt

… and then re-run the example generation script:

$ ./scripts/generate examples

Adding a new Kubernetes releases

To add a new supported Kubernetes release, run:

./scripts/add-kubernetes-release "${VERSION}"

Changing how the Kubernetes bindings are generated

The logic for generating the Dhall code doesn't reside within this repository but actually resides within the dhall-openapi subproject of the dhall-haskell repository. That means that if you want to change the generated code you will need to do so in two steps:

  • Make a pull request against the upstream dhall-haskell repository to change the code generated by dhall-openapi

  • Make a pull request against this repository to pick up a newer reference to the dhall-haskell repository incorporating the change to dhall-openapi

If you try to create a pull request to amend the generated Dhall files directly then CI will reject the pull request since it verifies that the Dhall code stored in version control matches what dhall-openapi would generate from the Kubernetes OpenAPI specification.

Once you update the dhall-openapi dependency you can regenerate the Kubernetes bindings by running:

$ ./scripts/generate kubernetes

Updating the dhall-openapi dependency

The dhall-openapi dependency is a subproject of the dhall-haskell repository, so in order to upgrade dhall-openapi you need to update the reference to the dhall-haskell repository.

If you're not prepared to make a pull request to change the dhall-haskell project then you can generate code for this project using a local checkout of the dhall-haskell repository by editing the Nix code like this:

--- a/nix/nixpkgs.nix
+++ b/nix/nixpkgs.nix
                    json =
                      builtins.fromJSON (builtins.readFile ./dhall-haskell.json);
 
-                   dhall-haskell = pkgsNew.fetchFromGitHub {
-                     owner = "dhall-lang";
-                     repo = "dhall-haskell";
-                     inherit (json) rev sha256 fetchSubmodules;
-                   };
+                   dhall-haskell = ~/path/to/dhall-haskell;
 
                  in
                    (import "${dhall-haskell}/default.nix").dhall-openapi;

Once you do change the upstream dhall-openapi project, then you can pick up the change here by runing:

$ nix-prefetch-git --fetch-submodules https://github.com/dhall-lang/dhall-haskell.git > ./nix/dhall-haskell.json

Generating everything

If you're not sure what files you need to regenerate then you can generate everything by running the generate script with no arguments:

$ ./scripts/generate

Upgrading Nixpkgs

If you want to upgrade to a newer revision of Nixpkgs, then run:

$ nix-prefetch-git https://github.com/NixOS/nixpkgs.git "${REVISION}" > ./nix/nixpkgs.json

Tests

All tests are defined in release.nix. We run these tests in CI in a Hydra project.

You can run the tests locally with the following command:

nix build --file ./release.nix