Skip to main content

Deploy to Kubernetes with Dagger

This tutorial illustrates how to use Dagger to build, push and deploy Docker images to Kubernetes.

Prerequisites

For this tutorial, you will need a Kubernetes cluster.

Kind is a tool for running local Kubernetes clusters using Docker.

1. Install kind

Follow these instructions to install Kind.

Alternatively, on macOS using homebrew:

brew install kind

2. Start a local registry

docker run -d -p 5000:5000 --name registry registry:2

3. Create a cluster with the local registry enabled in containerd

cat <<EOF | kind create cluster --config=-
kind: Cluster
apiVersion: kind.x-k8s.io/v1alpha4
containerdConfigPatches:
- |-
[plugins."io.containerd.grpc.v1.cri".registry.mirrors."localhost:5000"]
endpoint = ["http://registry:5000"]
EOF

4. Connect the registry to the cluster network

docker network connect kind registry

Initialize a Dagger Project and Environment

(optional) Setup example app

You will need the local copy of the Dagger examples repository used in previous guides

git clone https://github.com/dagger/examples

Make sure that all commands are run from the todoapp directory:

cd examples/todoapp

Organize your package

Let's create a new directory for our Cue package:

mkdir kube

Deploy using Kubectl

Kubernetes objects are located inside the k8s folder:

ls -l k8s
# k8s
# ├── deployment.yaml
# └── service.yaml

# 0 directories, 2 files

As a starting point, let's deploy them manually with kubectl:

kubectl apply -f k8s/
# deployment.apps/todoapp created
# service/todoapp-service created

Verify that the deployment worked:

kubectl get deployments
# NAME READY UP-TO-DATE AVAILABLE AGE
# todoapp 1/1 1 1 10m

kubectl get service
# NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE
# todoapp-service NodePort 10.96.225.114 <none> 80:32658/TCP 11m

The next step is to transpose it in Cue. Before continuing, clean everything:

kubectl delete -f k8s/
# deployment.apps "todoapp" deleted
# service "todoapp-service" deleted

Create a basic plan

Create a file named todoapp.cue and add the following configuration to it.

todoapp/kube/todoapp.cue
package main

import (
"alpha.dagger.io/dagger"
"alpha.dagger.io/kubernetes"
)

// input: kubernetes objects directory to deploy to
// set with `dagger input dir manifest ./k8s -e kube`
manifest: dagger.#Artifact & dagger.#Input

// Deploy the manifest to a kubernetes cluster
todoApp: kubernetes.#Resources & {
"kubeconfig": kubeconfig
source: manifest
}

This defines a todoApp variable containing the Kubernetes objects used to create a todoapp deployment. It also references a kubeconfig value defined below:

The following config.cue defines:

  • kubeconfig a generic value created to embed this string kubeconfig value
todoapp/kube/config.cue
package main

import (
"alpha.dagger.io/dagger"
)

// set with `dagger input text kubeconfig -f "$HOME"/.kube/config -e kube`
kubeconfig: string & dagger.#Input

Setup the environment

Create a new environment

Let's create a project:

dagger init

Let's create an environment to run it:

dagger new 'kube' -p kube

Configure the environment

Before we can bring up the deployment, we need to provide the kubeconfig input declared in the configuration. Otherwise, Dagger will complain about a missing input:

dagger up -e kube
# 5:05PM ERR system | required input is missing input=kubeconfig
# 5:05PM ERR system | required input is missing input=manifest
# 5:05PM FTL system | some required inputs are not set, please re-run with `--force` if you think it's a mistake missing=0s

You can inspect the list of inputs (both required and optional) using dagger input list:

dagger input list -e kube
# Input Value Set by user Description
# kubeconfig string false set with `dagger input text kubeconfig -f "$HOME"/.kube/config -e kube`
# manifest dagger.#Artifact false input: source code repository, must contain a Dockerfile set with `dagger input dir manifest ./k8s -e kube`
# todoApp.namespace *"default" | string false Kubernetes Namespace to deploy to
# todoApp.version *"v1.19.9" | string false Version of kubectl client

Let's provide the missing inputs:

# we'll use the "$HOME"/.kube/config created by `kind`
dagger input text kubeconfig -f "$HOME"/.kube/config -e kube

# Add as an artifact the k8s folder
dagger input dir manifest ./k8s -e kube

Deploying

Now is time to deploy to Kubernetes.

dagger up -e kube
# deploy | computing
# deploy | #26 0.700 deployment.apps/todoapp created
# deploy | #27 0.705 service/todoapp-service created
# deploy | completed duration=1.405s

Let's verify if the deployment worked:

kubectl get deployments
# NAME READY UP-TO-DATE AVAILABLE AGE
# todoapp 1/1 1 1 1m

Before continuing, cleanup deployment:

kubectl delete -f k8s/
# deployment.apps "todoapp" deleted
# service "todoapp-service" deleted

Building, pushing, and deploying Docker images

Rather than deploying an existing (todoapp) image, we're going to build a Docker image from the source, push it to a registry, and update the Kubernetes configuration.

Update the plan

Let's see how to deploy an image locally and push it to the local cluster

kube/todoapp.cue faces these changes:

  • repository, source code of the app to build. It needs to have a Dockerfile
  • registry, URI of the registry to push to
  • image, build of the image
  • remoteImage, push an image to the registry
  • kustomization, apply kustomization to image
todoapp/kube/todoapp.cue
package main

import (
"encoding/yaml"

"alpha.dagger.io/dagger"
"alpha.dagger.io/docker"
"alpha.dagger.io/kubernetes"
"alpha.dagger.io/kubernetes/kustomize"
)

// input: source code repository, must contain a Dockerfile
// set with `dagger input dir repository . -e kube`
repository: dagger.#Artifact & dagger.#Input

// Registry to push images to
registry: string & dagger.#Input
tag: "test-kind"

// input: kubernetes objects directory to deploy to
// set with `dagger input dir manifest ./k8s -e kube`
manifest: dagger.#Artifact & dagger.#Input

// Todoapp deployment pipeline
todoApp: {
// Build the image from repositoru artifact
image: docker.#Build & {
source: repository
}

// Push image to registry
remoteImage: docker.#Push & {
target: "\(registry):\(tag)"
source: image
}

// Update the image from manifest to use the deployed one
kustomization: kustomize.#Kustomize & {
source: manifest

// Convert CUE to YAML.
kustomization: yaml.Marshal({
resources: ["deployment.yaml", "service.yaml"]

images: [{
name: "public.ecr.aws/j7f8d3t2/todoapp"
newName: remoteImage.ref
}]
})
}

// Deploy the customized manifest to a kubernetes cluster
kubeSrc: kubernetes.#Resources & {
"kubeconfig": kubeconfig
source: kustomization
}
}

Connect the Inputs

Next, we'll provide the two new inputs, repository and registry.

# A name after `localhost:5000/` is required to avoid error on push to the local registry
dagger input text registry "localhost:5000/kind" -e kube

# Add todoapp (current folder) to repository value
dagger input dir repository . -e kube

Bring up the changes

dagger up -e kube
# 4:09AM INF manifest | computing
# 4:09AM INF repository | computing
# ...
# 4:09AM INF todoApp.kubeSrc | #37 0.858 service/todoapp-service created
# 4:09AM INF todoApp.kubeSrc | #37 0.879 deployment.apps/todoapp created
# Output Value Description
# todoApp.remoteImage.ref "localhost:5000/kind:test-kind@sha256:cb8d92518b876a3fe15a23f7c071290dfbad50283ad976f3f5b93e9f20cefee6" Image ref
# todoApp.remoteImage.digest "sha256:cb8d92518b876a3fe15a23f7c071290dfbad50283ad976f3f5b93e9f20cefee6" Image digest

Let's verify if the deployment worked:

kubectl get deployments
# NAME READY UP-TO-DATE AVAILABLE AGE
# todoapp 1/1 1 1 50s

Before continuing, cleanup deployment:

kubectl delete -f k8s/
# deployment.apps "todoapp" deleted
# service "todoapp-service" deleted

CUE Kubernetes manifest

This section will convert Kubernetes YAML manifest from k8s directory to CUE to take advantage of the language features.

For a more advanced example, see the official CUE Kubernetes tutorial

Convert Kubernetes objects to CUE

First, let's create re-usable definitions for the deployment and the service to remove a lot of boilerplate and repetition.

Let's define a re-usable #Deployment definition in kube/deployment.cue.

todoapp/kube/deployment.cue
package main

// Deployment template containing all the common boilerplate shared by
// deployments of this application.
#Deployment: {
// Name of the deployment. This will be used to label resources automatically
// and generate selectors.
name: string

// Container image.
image: string

// 80 is the default port.
port: *80 | int

// 1 is the default, but we allow any number.
replicas: *1 | int

// Deployment manifest. Uses the name, image, port and replicas above to
// generate the resource manifest.
manifest: {
apiVersion: "apps/v1"
kind: "Deployment"
metadata: {
"name": name
labels: app: name
}
spec: {
"replicas": replicas
selector: matchLabels: app: name
template: {
metadata: labels: app: name
spec: containers: [{
"name": name
"image": image
ports: [{
containerPort: port
}]
}]
}
}
}
}

Indeed, let's also define a re-usable #Service definition in kube/service.cue.

todoapp/kube/service.cue
package main

// Service template containing all the common boilerplate shared by
// services of this application.
#Service: {
// Name of the service. This will be used to label resources automatically
// and generate selector.
name: string

// NodePort is the default service type.
type: *"NodePort" | "LoadBalancer" | "ClusterIP" | "ExternalName"

// Ports where the service should listen
ports: [string]: number

// Service manifest. Uses the name, type and ports above to
// generate the resource manifest.
manifest: {
apiVersion: "v1"
kind: "Service"
metadata: {
"name": "\(name)-service"
labels: app: name
}
spec: {
"type": type
"ports": [
for k, v in ports {
name: k
port: v
},
]
selector: app: name
}
}
}

Generate Kubernetes manifest

Now that you have generic definitions for your Kubernetes objects. You can use them to get back your YAML definition without having boilerplate nor repetition.

Create a new definition named #AppManifest that will generate the YAML in kube/manifest.cue.

todoapp/kube/manifest.cue
package main

import (
"encoding/yaml"
)

// Define and generate kubernetes deployment to deploy to kubernetes cluster
#AppManifest: {
// Name of the application
name: string

// Image to deploy to
image: string

// Define a kubernetes deployment object
deployment: #Deployment & {
"name": name
"image": image
}

// Define a kubernetes service object
service: #Service & {
"name": name
ports: http: deployment.port
}

// Merge definitions and convert them back from CUE to YAML
manifest: yaml.MarshalStream([deployment.manifest, service.manifest])
}

Update manifest

You can now remove the manifest input in kube/todoapp.cue and instead use the manifest created by #AppManifest.

kube/todoapp.cue configuration has following changes:

  • removal of unused imported encoding/yaml and kustomize packages.
  • removal of manifest input that is doesn't need anymore.
  • removal of kustomization to replace it with #AppManifest definition.
  • Update kubeSrc to use manifest field instead of source because we don't send Kubernetes manifest of dagger.#Artifact type anymore.
todoapp/kube/todoapp.cue
package main

import (
"alpha.dagger.io/dagger"
"alpha.dagger.io/docker"
"alpha.dagger.io/kubernetes"
)

// input: source code repository, must contain a Dockerfile
// set with `dagger input dir repository . -e kube`
repository: dagger.#Artifact & dagger.#Input

// Registry to push images to
registry: string & dagger.#Input
tag: "test-kind"

// Todoapp deployment pipeline
todoApp: {
// Build the image from repositoru artifact
image: docker.#Build & {
source: repository
}

// Push image to registry
remoteImage: docker.#Push & {
target: "\(registry):\(tag)"
source: image
}

// Generate deployment manifest
deployment: #AppManifest & {
name: "todoapp"
image: remoteImage.ref
}

// Deploy the customized manifest to a kubernetes cluster
kubeSrc: kubernetes.#Resources & {
"kubeconfig": kubeconfig
manifest: deployment.manifest
}
}

Remove unused input

Now that we manage our Kubernetes manifest in CUE, we don't need manifest anymore.

# Remove `manifest` input
dagger input unset manifest -e kube

Deployment

dagger up -e kube
# 4:09AM INF manifest | computing
# 4:09AM INF repository | computing
# ...
# 4:09AM INF todoApp.kubeSrc | #37 0.858 service/todoapp-service created
# 4:09AM INF todoApp.kubeSrc | #37 0.879 deployment.apps/todoapp created
# Output Value Description
# todoApp.remoteImage.ref "localhost:5000/kind:test-kind@sha256:cb8d91518b076a3fe15a33f7c171290dfbad50283ad976f3f5b93e9f33cefag7" Image ref
# todoApp.remoteImage.digest "sha256:cb8d91518b076a3fe15a33f7c171290dfbad50283ad976f3f5b93e9f33cefag7" Image digest

Let's verify that the deployment worked:

kubectl get deployments
# NAME READY UP-TO-DATE AVAILABLE AGE
# todoapp 1/1 1 1 37s

Next Steps

Integrate Helm with Dagger: