This repository shows how to:
- create an ephemeral EKS cluster using Pulumi & install Traefik as Ingress Controller the CRD way using
IngressRoute
objects - configure a Route53 domain record dynamically to provide sub domain based routing through Traefik for all services (base services & application services)
- prepare ArgoCD for application deployment in the GitOps style
- install & configure Tekton on EKS and run a Cloud Native Buildpacks powered Pipeline
- integrate Tekton with GitLab (application https://gitlab.com/jonashackt/microservice-api-spring-boot) using direct trigger via
tkn
CLI (via aws-kubectl-tkn Docker image) or Tekton Triggers incl. GitLab Webhooks & reporting Tekton pipeline status back to GitLab using gitlab-set-status task - application deployment using ArgoCD based on the application configuration repo https://gitlab.com/jonashackt/microservice-api-spring-boot-config
It is structured according to all tools used:
.
├── argocd
│  └── ArgoCD related configuration
├── eks-deployment
│  └── Pulumi configuration (TypeScript style)
├── tekton
│  ├── install
│  │  └── Tekton CRDs, Pipelines, Dashboard, Triggers etc. installation
│  ├── misc
│  │  └── ServiceAccounts, PVCs, Secrets
│  ├── pipelines
│  │  └── Tekton Pipelines
│  ├── tasks
│  │  └── Tekton Tasks
│  └── triggers
│  └── Tekton Triggers Event Listener configuration
├── traefik
│  └── Traefik IngressRoute configurations
Let's simply roll out a AWS EKS cluster with Pulumi:
https://www.pulumi.com/docs/guides/crosswalk/aws/eks/
Our eks-deployment/index.ts looks like this:
import * as eks from "@pulumi/eks";
// Create an EKS cluster with the default configuration.
const cluster = new eks.Cluster("eks-for-tekton");
// Export the cluster's kubeconfig.
export const kubeconfig = cluster.kubeconfig;
export const eksUrl = cluster.eksCluster.endpoint;
To execute our Pulumi program, be sure to be logged into the correct Account at https://app.pulumi.com/your-account-here via pulumi login
using your Pulumi account's token (do a pulumi logout
before, if you're already logged into another Pulumi account).
Now select the correct stack and fire up Pulumi with:
pulumi stack select dev
pulumi up
After your EKS cluster has been setup correctly, use the kubeconfig
const exported inside our Pulumi program to create the kubeconfig.yml
:
pulumi stack output kubeconfig > kubeconfig.yml
To access the cluster be sure to have kubectl
installed. Try accessing it with:
kubectl --kubeconfig kubeconfig.yml get nodes
For merging the new kubeconfig into your systems profile see https://kubernetes.io/docs/concepts/configuration/organize-cluster-access-kubeconfig/
For example you could do:
pulumi stack output kubeconfig > ~/.kube/config-eks-for-tekton
export KUBECONFIG=~/.kube/config:~/.kube/config-eks-for-tekton
Now access via kubectx
is also possible.
First we need to create GitHub repository secrets containing our AWS API key id & access key (AWS_ACCESS_KEY_ID
& AWS_SECRET_ACCESS_KEY
) and our Pulumi access token (PULUMI_ACCESS_TOKEN
):
Our provision.yml workflow uses Pulumi like we did locally:
name: provision
on: [push]
jobs:
provision-aws-eks:
runs-on: ubuntu-latest
env:
AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_ID }}
AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
PULUMI_ACCESS_TOKEN: ${{ secrets.PULUMI_ACCESS_TOKEN }}
AWS_DEFAULT_REGION: 'eu-central-1'
# Create an GitHub environment referencing our EKS cluster endpoint
environment:
name: tekton-flux-eks-pulumi-dev
url: ${{ steps.pulumi-up.outputs.eks_url }}
steps:
- name: Checkout
uses: actions/checkout@master
- name: Setup node env
uses: actions/[email protected]
with:
node-version: '14'
- name: Cache node_modules
uses: actions/cache@v2
with:
path: ~/.npm
key: ${{ runner.os }}-node-${{ hashFiles('**/package-lock.json') }}
restore-keys: |
${{ runner.os }}-node-
- name: Install Pulumi dependencies before npm run generate to prevent it from breaking the build
run: npm install
working-directory: ./eks-deployment
- name: Install Pulumi CLI
uses: pulumi/[email protected]
- name: Provision AWS EKS cluster with Pulumi
id: pulumi-up
run: |
pulumi stack select dev
pulumi preview
echo "lets use --suppress-outputs here in order to prevent Pulumi from logging the kubeconfig into public GitHub Action logs"
pulumi up --yes --suppress-outputs
pulumi stack output kubeconfig > kubeconfig.yml
echo "::set-output name=eks_url::$(pulumi stack output eksUrl)/api/hello"
working-directory: ./eks-deployment
- name: Try to connect to our EKS cluster using kubectl
run: kubectl --kubeconfig kubeconfig.yml get nodes
working-directory: ./eks-deployment
Mind to use --suppress-outputs
flag for our pulumi up
to prevent the kubeconfig
from getting logged unmasked.
We also export our eks endpoint url
as an GitHub Environment (as described here).
I got this error (see log):
...
<botocore.awsrequest.AWSRequest object at 0x7f067c580670>
Unable to connect to the server: getting credentials: exec: executable aws failed with exit code 255
Error: Process completed with exit code 1.
Luckily this answer brought me into the right direction: https://stackoverflow.com/a/59184490/4964553
I needed to define the AWS_DEFAULT_REGION: 'eu-central-1'
also solely for kubectl
in GitHub Actions. With this the error was gone, since the other two variables for aws-cli
(which is already installed in the GitHub Actions virtual environment) were defined properly.
Per default Pulumi creates an EKS cluster with only 2 worker nodes - doing all this fancy Tekton & Argo stuff we should upgrade to at least 3 or 4 worker nodes.
Therefore we can tell Pulumi to use more using the desiredCapacity
configuration:
// Create an EKS cluster with the default configuration.
const cluster = new eks.Cluster("eks-for-tekton", {
desiredCapacity: 3,
minSize: 3,
maxSize: 4,
});
In order to prevent errors like error waiting for CloudFormation Stack update: failed to update CloudFormation stack (UPDATE_ROLLBACK_COMPLETE): ["Desired capacity:3 must be between the specified min size:1 and max size:2
we should also update the minSize
, maxSize
parameters.
See also the examples - e.g. https://github.com/pulumi/pulumi-eks/blob/master/examples/cluster/index.ts
I've started to have the Tekton Dashboard, the ArgoCD server/dashboard & Tekton Trigger EventListener exposed as their own separate Services of type LoadBalancer
, which leads to the creation of multiple classic Elastic Load Balancers in AWS:
In the section LoadBalancer for every http service
of https://blog.pipetail.io/posts/2020-05-04-most-common-mistakes-k8s/ the problem is described as:
resources might get expensive (external static ipv4 address, compute, per-second pricing ,...)
To prevent that one could use the concept of an api gateway or Ingress in K8s terms. One of the best solutions out there is Traefik https://traefik.io/
As of the docs there are 3 ways on how to use Traefik in Kubernetes:
- IngressRoute Custom Resource Definition (CRD) for Kubernetes: https://doc.traefik.io/traefik/providers/kubernetes-crd/
- "old familiar" Ingress Controller as the Kubernetes Ingress provider: https://doc.traefik.io/traefik/providers/kubernetes-ingress/
- Experimental Kubernetes Gateway API: https://doc.traefik.io/traefik/providers/kubernetes-gateway/
CRDs seem to be the current defacto standard way to extend Kubernetes (by extending the Kubernetes API): https://kubernetes.io/docs/concepts/extend-kubernetes/api-extension/custom-resources/
So let's go with the IngressRoute CRD. There's also a full guide including Let's Encrypt in the Traefik docs https://doc.traefik.io/traefik/user-guides/crd-acme/
Install Traefik via Helm: https://doc.traefik.io/traefik/getting-started/install-traefik/#use-the-helm-chart from it's chart at https://github.com/traefik/traefik-helm-chart:
This chart bootstraps Traefik version 2 as a Kubernetes ingress controller, using Custom Resources IngressRoute: https://docs.traefik.io/providers/kubernetes-crd/
We do all this right inside our GitHub Actions workflow provision.yml:
- name: Install Traefik via Helm
run: |
echo "--- Install Traefik via Helm (which is already installed in GitHub Actions environment https://github.com/actions/virtual-environments)
helm repo add traefik https://helm.traefik.io/traefik
helm repo update
helm upgrade -i traefik traefik/traefik
But instead of helm install traefik traefik/traefik
we use helm upgrade -i traefik traefik/traefik
to prevent the error Error: INSTALLATION FAILED: cannot re-use a name that is still in use
(see https://stackoverflow.com/a/70465191/4964553).
Now Traefik is already deployed and we can see it's Service (aka the Traefik Ingress Controller) in k9s for example:
You may temporarily expose the dashboard with a local kubectl port-forward
like this (but we will create a nice domain later also):
kubectl port-forward $(kubectl get pods --selector "app.kubernetes.io/name=traefik" --output=name) 9000:9000
And access it at http://127.0.0.1:9000/dashboard/
Right now our Traefik installation uses no pinned version and every new GitHub Actions workflow run simply uses the newest version of Traefik.
So how can we use a pinned version with Helm? Simply using --version
isn't enough for us, since Renovate needs a dependency file to look at: https://docs.renovatebot.com/modules/manager/helm-values/
But there's another way to use a simple Chart.yaml to pin our version and have a manageble file for Renovate:
apiVersion: v2
type: application
name: traefik
version: 0.0.0 # unused
appVersion: 0.0.0 # unused
dependencies:
- name: traefik
repository: https://helm.traefik.io/traefik
version: 10.19.4
Now with the commands:
helm dependency update traefik/install
helm upgrade -i traefik traefik/install
We can now install Traefik in a Renovate-ready way.
Now let's configure the IngressRoute
objects to get our Services accessible through Traefik
https://doc.traefik.io/traefik/user-guides/crd-acme/#traefik-routers
https://doc.traefik.io/traefik/routing/routers/#rule
So start by creating our first IngressRoute
definition - right now only statically to see it working inside traefik-application-ingress-routes.yml:
apiVersion: traefik.containo.us/v1alpha1
kind: IngressRoute
metadata:
name: microservice-api-spring-boot-ingressroute
namespace: default
spec:
entryPoints:
- web
routes:
- match: Host(`microservice-api-spring-boot-main`)
kind: Rule
services:
- name: microservice-api-spring-boot-main
port: 80
Apply it with kubectl apply -f traefik/traefik-application-ingress-routes.yml
Finally use a REST client like Postman to access our Service:
You need to provide the Host:microservice-api-spring-boot-main
header in your request in order to make the call work.
First create a Domain with AWS Route53 - this will take a while & you should finally receive a mail, that your domain was registered successfully (this took around 20mins for me).
When we have our domain ready - for me this is tekton-argocd.de - we can configure the Route53 hosted zone with the correct records.
Let's test it by enhancing our IngressRoute
inside traefik-application-ingress-routes.yml:
apiVersion: traefik.containo.us/v1alpha1
kind: IngressRoute
metadata:
name: microservice-api-spring-boot-ingressroute
namespace: default
spec:
entryPoints:
- web
routes:
- match: Host(`microservice-api-spring-boot-main.tekton-argocd.de`)
kind: Rule
services:
- name: microservice-api-spring-boot-main
port: 80
And apply it with kubectl apply -f traefik/traefik-application-ingress-routes.yml
Now creating the Route53 record manually isn't what we should aim for. Instead let's use AWS CLI to do that for us.
Here's a starting point https://aws.amazon.com/premiumsupport/knowledge-center/alias-resource-record-set-route53-cli/
But we don't want to do this using a static file like with the proposed --change-batch file://sample.json
- instead we want to have it more dynamic so we can use a command inside our GitHub Actions workflow.
The idea is derived from this so answer https://stackoverflow.com/a/49228748/4964553, where we can simply use the json snippet inline without an extra file.
We also want to have an idempotent solution which we can run 1 or many times in our GitHub Actions CI. Therefore we use the "Action" : "UPSERT"
(see https://aws.amazon.com/premiumsupport/knowledge-center/simple-resource-record-route53-cli/).
echo "--- Creating or updating ('UPSERT') Route53 hosted zone A record to point to ELB Traefik (see https://aws.amazon.com/premiumsupport/knowledge-center/simple-resource-record-route53-cli/)"
echo "--- Creating Route53 hosted zone record (mind to wrap the variables in double quotes in order to get them evaluated, see https://stackoverflow.com/a/49228748/4964553)"
aws route53 change-resource-record-sets \
--hosted-zone-id $ROUTE53_DOMAIN_HOSTED_ZONE_ID \
--change-batch '
{
"Comment": "Create or update Route53 hosted zone A record to point to ELB Traefik is configured to"
,"Changes": [{
"Action" : "UPSERT"
,"ResourceRecordSet" : {
"Name" : "*.'"$ROUTE53_DOMAIN_NAME"'"
,"Type" : "A"
,"AliasTarget": {
"HostedZoneId": "'"$ELB_HOSTED_ZONE_ID"'",
"DNSName": "dualstack.'"$ELB_URL"'",
"EvaluateTargetHealth": true
}
}
}]
}
'
Using variables inside the json provided to the
--change-batch
parameter, we need to use single quotes and open them up immediately after (also see https://stackoverflow.com/a/49228748/4964553)
As you can see, we need to configure 4 variables to make this command run:
$ROUTE53_DOMAIN_HOSTED_ZONE_ID
: This is the hosted zone id of your Route53 domain you need to register before (the registration itself is a manual step)$ROUTE53_DOMAIN_NAME
: Your Route53 registered domain name. As we want all routing to be done by Traefik, we can configure a wildcard record here using*.$ROUTE53_DOMAIN_NAME
$ELB_HOSTED_ZONE_ID
: A different hosted zone id than your domain!. This is the hosted zone id of the Elastic Load Balancer, which gets provisioned through the Traefik Service deployment (via Helm).$ELB_URL
: The ELB url of the Traefik Service. We need to preface it withdualstack.
in order to make it work (see https://docs.aws.amazon.com/Route53/latest/APIReference/API_AliasTarget.html)
Obtaining all those variables isn't trivial. We can start with the Route53 domain name, we need to configure as a static GitHub Actions environment varialbe at the top of our provision.yml:
name: provision
on: [push]
env:
...
ROUTE53_DOMAIN_NAME: tekton-argocd.de
...
- name: Create or update Route53 hosted zone A record to point to ELB Traefik is configured to
run: |
echo "--- Obtaining the Route53 domain's hosted zone id"
ROUTE53_DOMAIN_HOSTED_ZONE_ID="$(aws route53 list-hosted-zones-by-name | jq --arg name "$ROUTE53_DOMAIN_NAME." -r '.HostedZones | .[] | select(.Name=="\($name)") | .Id')"
echo "--- Obtaining the ELB hosted zone id"
echo "Therefore cutting the ELB url from the traefik k8s Service using cut (see https://stackoverflow.com/a/29903172/4964553)"
ELB_NAME="$(kubectl get service traefik -n default --output=jsonpath='{.status.loadBalancer.ingress[0].hostname}' | cut -d "-" -f 1)"
echo "Extracting the hosted zone it using aws cli and jq (see https://stackoverflow.com/a/53230627/4964553)"
ELB_HOSTED_ZONE_ID="$(aws elb describe-load-balancers | jq --arg name "$ELB_NAME" -r '.LoadBalancerDescriptions | .[] | select(.LoadBalancerName=="\($name)") | .CanonicalHostedZoneNameID')"
echo "--- Obtaining the Elastic Load Balancer url as the A records AliasTarget"
ELB_URL="$(kubectl get service traefik -n default --output=jsonpath='{.status.loadBalancer.ingress[0].hostname}')"
https://doc.traefik.io/traefik/operations/dashboard/
As we now have our Route53 record configuration in place to access our apps, we can also create a nice access to our Traefik dashboard to avoid the need of a manually started local port-forward
:
https://doc.traefik.io/traefik/getting-started/install-traefik/#exposing-the-traefik-dashboard
Therefore let't create a IngressRoute
for the Traefik dashboard at traefik/traefik-dashboard.yml:
apiVersion: traefik.containo.us/v1alpha1
kind: IngressRoute
metadata:
name: dashboard
spec:
entryPoints:
- web
routes:
- match: Host(`traefik.tekton-argocd.de`)
kind: Rule
services:
- name: api@internal
kind: TraefikService
Now install it with:
kubectl apply -f traefik/traefik-dashboard.yml
We also directly expose our nice Traefik url traefik.tekton-argocd.de as GitHub Actions Environment:
environment:
name: traefik-eks-url
url: ${{ steps.traefik-expose.outputs.traefik_url }}
...
- name: Expose Traefik url as GitHub environment
id: traefik-expose
run: |
echo "--- Apply Traefik-ception IngressRule"
kubectl apply -f traefik/traefik-dashboard.yml
echo "--- Wait until Loadbalancer url is present (see https://stackoverflow.com/a/70108500/4964553)"
until kubectl get service/traefik -n default --output=jsonpath='{.status.loadBalancer}' | grep "ingress"; do : ; done
TRAEFIK_URL="http://traefik.$ROUTE53_DOMAIN_NAME"
echo "All Services should be accessible through Traefik Ingress at $TRAEFIK_URL - creating GitHub Environment"
echo "::set-output name=traefik_url::$TRAEFIK_URL"
Now Traefik should be accessible at http://traefik.tekton-argocd.de also through our pipeline.
https://tekton.dev/docs/getting-started/
https://tekton.dev/docs/dashboard/
Install it with:
kubectl apply --filename https://github.com/tektoncd/dashboard/releases/latest/download/tekton-dashboard-release.yaml
Now as we already ran some Tasks let's have a look into the Tekton dashboard:
kubectl proxy --port=8080
Then open your Browser at http://localhost:8080/api/v1/namespaces/tekton-pipelines/services/tekton-dashboard:http/proxy/
So let's use Ingress with our Traefik and the nice Route53 domain & wildcard record to route from tekton.tekton-argocd.de. Simply create an Traefik IngressRoute
as described in traefik/tekton-dashboard.yml:
apiVersion: traefik.containo.us/v1alpha1
kind: IngressRoute
metadata:
name: tekton-dashboard
namespace: tekton-pipelines
spec:
entryPoints:
- web
routes:
- kind: Rule
match: Host(`tekton.tekton-argocd.de`)
services:
- name: tekton-dashboard
port: 9097
Apply it with kubectl apply -f traefik/tekton-dashboard.yml
and you should be able to access the Tekton dashboard already at http://tekton.tekton-argocd.de:
Now that our AWS ELB is provisioned we can finally grad it's URL with:
kubectl get service tekton-dashboard-external-svc-manual -n tekton-pipelines --output=jsonpath='{.status.loadBalancer.ingress[0].hostname}'
Then it's easy to use the output and create a GitHub step variable, which is used in the GitHub Environment definition at the top of the job:
echo "--- Create GitHub environment var"
DASHBOARD_HOST=$(kubectl get service tekton-dashboard-external-svc-manual -n tekton-pipelines --output=jsonpath='{.status.loadBalancer.ingress[0].hostname}')
echo "The Tekton dashboard is accessible at $DASHBOARD_HOST - creating GitHub Environment"
echo "::set-output name=dashboard_host::http://$DASHBOARD_HOST"
https://tekton.dev/docs/getting-started/#installation
So let's add the installation and wait for Tekton to become available:
...
- name: Install Tekton Pipelines
run: kubectl apply --filename https://storage.googleapis.com/tekton-releases/pipeline/latest/release.yaml
- name: Wait for Tekton to become ready & show running Tekton pods
run: |
kubectl wait --for=condition=ready pod -l app=tekton-pipelines-controller --namespace tekton-pipelines
kubectl get pods --namespace tekton-pipelines
https://tekton.dev/docs/getting-started/#persistent-volumes
Let's check if our EKS cluster already has a StorageClass
defined with kubectl get storageclasses
:
$ kubectl get storageclasses
NAME PROVISIONER RECLAIMPOLICY VOLUMEBINDINGMODE ALLOWVOLUMEEXPANSION AGE
gp2 (default) kubernetes.io/aws-ebs Delete WaitForFirstConsumer false 179m
Before creating it let's check if there is already a ConfigMap
defined:
kubectl describe configmap config-artifact-pvc -n tekton-pipelines
From the docs:
Your Kubernetes cluster, such as one from Google Kubernetes Engine, may have persistent volumes set up at the time of creation, thus no extra step is required
If there's no Persistens Volume defined, we need to create a ConfigMap
which defines the Persistent Volume Tekton will request:
kubectl create configmap config-artifact-pvc \
--from-literal=size=10Gi \
--from-literal=storageClassName=gp2 \
-o yaml -n tekton-pipelines \
--dry-run=client | kubectl replace -f -
Install the Tekton CLI e.g. via homebrew:
brew tap tektoncd/tools
brew install tektoncd/tools/tektoncd-cli
See the task-hello-world.yaml:
apiVersion: tekton.dev/v1beta1
kind: Task
metadata:
name: hello
spec:
steps:
- name: hello
image: ubuntu
command:
- echo
args:
- "Hello World!"
Let's apply it to our cluster:
kubectl apply -f tekton/tasks/task-hello-world.yaml
Let's show our newly created task:
$ tkn task list
NAMESPACE NAME DESCRIPTION AGE
default hello 24 seconds ago
Now this is only a Tekton Task definition. We need another Tekton object - the TaskRun
- in order to run our Task. Create it with:
tkn task start hello
Follow the logs of the TaskRun with:
tkn taskrun logs --last -f
https://buildpacks.io/docs/tools/tekton/
Install git clone and buildpacks Task:
kubectl apply -f https://raw.githubusercontent.com/tektoncd/catalog/master/task/git-clone/0.4/git-clone.yaml
kubectl apply -f https://raw.githubusercontent.com/tektoncd/catalog/master/task/buildpacks/0.3/buildpacks.yaml
To access the GitLab Container Registry we need to first create a PAT or deploy token (see https://docs.gitlab.com/ee/user/packages/container_registry/#authenticate-with-the-container-registry)
Go to Settings/Repository
inside your GitLab repo - for me this is https://gitlab.com/jonashackt/microservice-api-spring-boot/-/settings/repository
There create a token TektonBuildpacksToken
under Deploy tokens
with a username gitlab-token
and read_registry
& write_registry
access.
Now create GitHub Repository Secrets called GITLAB_CR_USER
and GITLAB_CR_PASSWORD
accordingly with the Tokens username and token.
Finally we can create our Secret inside our GitHub Actions pipeline:
https://buildpacks.io/docs/tools/tekton/#42-authorization
kubectl create secret docker-registry docker-user-pass \
--docker-server=registry.gitlab.com \
--docker-username=${{ secrets.GITLAB_CR_USER }} \
--docker-password=${{ secrets.GITLAB_CR_PASSWORD }} \
--namespace default
After the first successful secret creation, we sadly get the error error: failed to create secret secrets "docker-user-pass" already exists
- which is correct, since the secret already exists.
But there's help (see https://stackoverflow.com/a/45881259/4964553): We add --save-config --dry-run=client -o yaml | kubectl apply -f -
to our command like this:
kubectl create secret docker-registry gitlab-container-registry \
--docker-server=registry.gitlab.com \
--docker-username=${{ secrets.GITLAB_CR_USER }} \
--docker-password=${{ secrets.GITLAB_CR_PASSWORD }} \
--namespace default \
--save-config --dry-run=client -o yaml | kubectl apply -f -
Now we made an apply
out of our create
kubectl command, which we can use repetitively :)
We also need to create a ServiceAccount
that uses this secret as buildpacks-service-account-gitlab.yml
apiVersion: v1
kind: ServiceAccount
metadata:
name: buildpacks-service-account-gitlab
secrets:
- name: gitlab-container-registry
https://buildpacks.io/docs/tools/tekton/#41-pvcs
Create new buildpacks-source-pvc.yml:
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
name: buildpacks-source-pvc
spec:
accessModes:
- ReadWriteOnce
resources:
requests:
storage: 500Mi
Create pipeline.yml:
apiVersion: tekton.dev/v1beta1
kind: Pipeline
metadata:
name: buildpacks-test-pipeline
spec:
params:
- name: image
type: string
description: image URL to push
workspaces:
- name: source-workspace # Directory where application source is located. (REQUIRED)
- name: cache-workspace # Directory where cache is stored (OPTIONAL)
tasks:
- name: fetch-repository # This task fetches a repository from github, using the `git-clone` task you installed
taskRef:
name: git-clone
workspaces:
- name: output
workspace: source-workspace
params:
- name: url
value: https://github.com/buildpacks/samples
- name: subdirectory
value: ""
- name: deleteExisting
value: "true"
- name: buildpacks # This task uses the `buildpacks` task to build the application
taskRef:
name: buildpacks
runAfter:
- fetch-repository
workspaces:
- name: source
workspace: source-workspace
- name: cache
workspace: cache-workspace
params:
- name: APP_IMAGE
value: "$(params.image)"
- name: SOURCE_SUBPATH
value: "apps/java-maven" # This is the path within the samples repo you want to build (OPTIONAL, default: "")
- name: BUILDER_IMAGE
value: paketobuildpacks/builder:base # This is the builder we want the task to use (REQUIRED)
- name: display-results
runAfter:
- buildpacks
taskSpec:
steps:
- name: print
image: docker.io/library/bash:5.1.4@sha256:b208215a4655538be652b2769d82e576bc4d0a2bb132144c060efc5be8c3f5d6
script: |
#!/usr/bin/env bash
set -e
echo "Digest of created app image: $(params.DIGEST)"
params:
- name: DIGEST
params:
- name: DIGEST
value: $(tasks.buildpacks.results.APP_IMAGE_DIGEST)
And now apply all three configs with:
kubectl apply -f tekton/misc/buildpacks-source-pvc.yml
kubectl apply -f tekton/misc/buildpacks-service-account-gitlab.yml
kubectl apply -f tekton/pipelines/pipeline.yml
Create pipeline-run.yml:
apiVersion: tekton.dev/v1beta1
kind: PipelineRun
metadata:
generateName: buildpacks-test-pipeline-run-
spec:
serviceAccountName: buildpacks-service-account-gitlab # Only needed if you set up authorization
pipelineRef:
name: buildpacks-test-pipeline
workspaces:
- name: source-workspace
subPath: source
persistentVolumeClaim:
claimName: buildpacks-source-pvc
- name: cache-workspace
subPath: cache
persistentVolumeClaim:
claimName: buildpacks-source-pvc
params:
- name: image
value: registry.gitlab.com/jonashackt/microservice-api-spring-boot # This defines the name of output image
A crucial point here is to change the metadata: name: buildpacks-test-pipeline-run
into metadata: generateName: buildpacks-test-pipeline-run-
. Why? Because if we use the name
parameter every kubectl apply
tries to create the PipelineRun
object with the same name which results in errors like this:
Error from server (AlreadyExists): error when creating "tekton/pipelines/pipeline-run.yml": pipelineruns.tekton.dev "buildpacks-test-pipeline-run" already exists
Using the generateName
field fixes our problem (see https://stackoverflow.com/questions/69880096/how-to-restart-tekton-pipelinerun-having-a-pipeline-run-yml-defined-in-git-e-g/69880097#69880097), although we should implement a kind of garbage collection for our PipelineRun objects...
Also mind the params: name: image
and insert an image name containing the correct namespace of your Container Registry you created a Secret for!
Also apply with
kubectl apply -f tekton/pipelines/pipeline-run.yml
Looking into the Tekton dashboard we should now finally see a successful Pipeline run:
It's extremey easy to leverage the Buildpacks cache image feature, that will create a separate cache image for building and will store it alongside the resulting app image in our Container Registry.
Therefore simply add the CACHE_IMAGE
parameter using our $(params.IMAGE)
definition and appending :paketo-build-cache
like this inside our Pipeline:
params:
- name: APP_IMAGE
value: "$(params.IMAGE)"
- name: CACHE_IMAGE
value: "$(params.IMAGE):paketo-build-cache"
- name: BUILDER_IMAGE
value: paketobuildpacks/builder:base # This is the builder we want the task to use (REQUIRED)
Now the next build should produce a separate image inside our Container Registry:
What about extending our Tekton Pipeline with a Maven Task, that initiates a test run before we build our app container using Buildpacks.
Therefore we can simply use the https://hub.tekton.dev/tekton/task/maven Task from Tekton Hub.
First we need to install the Task in our GitHub Actions provision.yml:
kubectl apply -f https://raw.githubusercontent.com/tektoncd/catalog/main/task/maven/0.2/maven.yaml
Now we need to enhance our pipeline.yml with a new workspace maven-settings
and the Task definition maven-test
, which should also be defined as runAfter
target in the buildpacks
Task:
workspaces:
- name: maven-settings # Maven settings, see https://hub.tekton.dev/tekton/task/maven
...
tasks:
...
- name: maven-test
taskRef:
name: maven
runAfter:
- fetch-repository
params:
- name: GOALS
value:
- verify
workspaces:
- name: maven-settings
workspace: maven-settings
- name: source
workspace: source-workspace
- name: buildpacks # This task uses the `buildpacks` task to build the application
taskRef:
name: buildpacks
runAfter:
- maven-test
We also should enhance our pipeline-run.yml and gitlab-push-listener.yml to define the additional workspace:
pipelineRef:
name: buildpacks-test-pipeline
workspaces:
- name: maven-settings
emptyDir: {}
Finally we should also add the new workspace with --workspace name=maven-settings,emptyDir=
to our tkn pipeline start
command in the project's branch https://gitlab.com/jonashackt/microservice-api-spring-boot/-/tree/trigger-tekton-via-gitlabci inside the .gitlab-ci.yml
:
tkn pipeline start buildpacks-test-pipeline \
--serviceaccount buildpacks-service-account-gitlab \
--workspace name=maven-settings,emptyDir= \
--workspace name=source-workspace,subPath=source,claimName=buildpacks-source-pvc \
--workspace name=cache-workspace,subPath=cache,claimName=buildpacks-source-pvc \
--param IMAGE=registry.gitlab.com/jonashackt/microservice-api-spring-boot \
--param SOURCE_URL=https://gitlab.com/jonashackt/microservice-api-spring-boot \
--param REPO_PATH_ONLY=jonashackt/microservice-api-spring-boot \
--param SOURCE_REVISION=main \
--param GITLAB_HOST=gitlab.com \
--param TEKTON_DASHBOARD_HOST="http://abd1c6f235c9642bf9d4cdf632962298-1232135946.eu-central-1.elb.amazonaws.com" \
--timeout 240s \
--showlog
It seems that the Tekton Hub's Maven Task doesn't implement caching for us. Our builds tend to download all Maven dependencies over and over again:
As stated in https://developers.redhat.com/blog/2020/02/26/speed-up-maven-builds-in-tekton-pipelines#maven_task_with_a_workspace we need to define our own Task for that.
As we don't need the settings.xml
configuration (e.g. for Proxy settings), which is the main point of the Tekton Hub's Maven Task, we can simply create our own - see task-maven-with-cache.yml:
apiVersion: tekton.dev/v1alpha1
kind: Task
metadata:
name: maven-with-cache
spec:
workspaces:
- name: source
description: The workspace consisting of maven project.
- name: maven-repo-cache
description: The workspace holding the Maven repository for caching.
params:
- name: GOALS
description: The Maven goals to run
type: array
default:
- "package"
steps:
- name: mvn
image: gcr.io/cloud-builders/mvn
workingDir: $(workspaces.source.path)
command: ["/usr/bin/mvn"]
args:
- -Dmaven.repo.local=$(workspaces.maven-repo-cache.path)
- "$(params.GOALS)"
We need to kubectl apply
our Task maven-with-cache
with:
kubectl apply -f tasks/task-maven-with-cache.yml
We also need to use our new Maven Task inside our pipeline.yml:
workspaces:
- name: maven-repo-cache # Maven repository cahce, see https://developers.redhat.com/blog/2020/02/26/speed-up-maven-builds-in-tekton-pipelines#run_a_maven_pipeline
...
- name: maven-test
taskRef:
name: maven-with-cache
runAfter:
- fetch-repository
params:
- name: GOALS
value:
- verify
workspaces:
- name: source
workspace: source-workspace
- name: maven-repo-cache
workspace: maven-repo-cache
Now we could try to also create a new PersistentVolumeClaim
as stated in https://developers.redhat.com/blog/2020/02/26/speed-up-maven-builds-in-tekton-pipelines#run_a_maven_pipeline
- name: maven-repo-cache
persistentVolumeClaim:
claimName: maven-repo-cache-pvc
But we would run into errors like:
completionTime: '2021-12-06T09:13:49Z'
conditions: - lastTransitionTime: '2021-12-06T09:13:49Z'
message: more than one PersistentVolumeClaim is bound
reason: TaskRunValidationFailed
status: 'False'
type:
It seems that Tekton doesn't like to have multiple PVC inside one Task: tektoncd/pipeline#3480 and tektoncd/pipeline#3085
In general, try to only use a single PVC for each task.
So we need to use a separate subPath
inside our already existing PVC buildpacks-source-pvc
(which doesn't have a matching name any more it seems :) ).
As stated in https://buildpacks.io/docs/tools/tekton/#43-pipeline a Tekton workspace could be simply seen as a shared directory (see https://tekton.dev/docs/pipelines/workspaces/, where I didn't get this first).
So we finally simply provide a new workspace using the existing PVC but a different subPath
to our pipeline-run.yml & gitlab-push-listener.yml:
- name: maven-repo-cache
subPath: maven-repo-cache
persistentVolumeClaim:
claimName: buildpacks-source-pvc
And we can even optimize our solution by simply using the maven-settings
workspace definition of the standard Tekton Hub's Maven Task inside our Pipeline
workspaces:
- name: maven-repo-cache # Maven repository cahce, see https://developers.redhat.com/blog/2020/02/26/speed-up-maven-builds-in-tekton-pipelines#run_a_maven_pipeline
...
- name: maven-test
taskRef:
name: maven
runAfter:
- fetch-repository
params:
- name: GOALS
value:
- -Dmaven.repo.local=$(workspaces.maven-settings.path)
- verify
workspaces:
- name: source
workspace: source-workspace
- name: maven-settings
workspace: maven-repo-cache
Now our Pipeline should run faster then before, since the Maven cache is used:
How to trigger Tekton PipelineRun
from GitLab?
The simplest possible solution is to leverage GitLab CI and trigger Tekton from there.
See https://stackoverflow.com/a/69991508/4964553
Have a look at this example gitlab.com project https://gitlab.com/jonashackt/microservice-api-spring-boot/-/tree/trigger-tekton-via-gitlabci
Full getting-started guide: https://github.com/tektoncd/triggers/tree/v0.17.0/docs/getting-started
BUT FIRST Examples are a great inspiration - for GitLab this is especially:
https://github.com/tektoncd/triggers/tree/main/examples/v1beta1/gitlab
https://tekton.dev/docs/triggers/install/
kubectl apply --filename https://storage.googleapis.com/tekton-releases/triggers/latest/release.yaml
kubectl apply --filename https://storage.googleapis.com/tekton-releases/triggers/latest/interceptors.yaml
See https://github.com/tektoncd/triggers/blob/v0.17.0/examples/rbac.yaml
So we also create serviceaccount-rb-crb.yml:
apiVersion: v1
kind: ServiceAccount
metadata:
name: tekton-triggers-example-sa
---
apiVersion: rbac.authorization.k8s.io/v1
kind: RoleBinding
metadata:
name: triggers-example-eventlistener-binding
subjects:
- kind: ServiceAccount
name: tekton-triggers-example-sa
roleRef:
apiGroup: rbac.authorization.k8s.io
kind: ClusterRole
name: tekton-triggers-eventlistener-roles
---
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRoleBinding
metadata:
name: triggers-example-eventlistener-clusterbinding
subjects:
- kind: ServiceAccount
name: tekton-triggers-example-sa
namespace: default
roleRef:
apiGroup: rbac.authorization.k8s.io
kind: ClusterRole
name: tekton-triggers-eventlistener-clusterroles
kubectl apply -f tekton/triggers/serviceaccount-rb-crb.yml
As our Tekton Trigger API will be setup as a public API in the end, we need to secure our Trigger API somehow.
One way is to create a secret ID the calling JSON must contain. So let's create tekton-trigger-secret.yml:
apiVersion: v1
kind: Secret
metadata:
name: gitlab-secret
type: Opaque
stringData:
secretToken: "1234567"
kubectl apply -f tekton/triggers/tekton-trigger-secret.yml
So let's start with the EventListener
. We'll adapt the EventListener
from the example (see https://github.com/tektoncd/triggers/blob/main/examples/v1beta1/gitlab/gitlab-push-listener.yaml) to use our Buildpacks Pipeline defined in pipeline.yml.
Therefore let's create a new file called gitlab-push-listener.yml:
apiVersion: triggers.tekton.dev/v1beta1
kind: EventListener
metadata:
name: gitlab-listener
spec:
serviceAccountName: tekton-triggers-example-sa
triggers:
- name: gitlab-push-events-trigger
interceptors:
- name: "verify-gitlab-payload"
ref:
name: "gitlab"
kind: ClusterInterceptor
params:
- name: secretRef
value:
secretName: "gitlab-secret"
secretKey: "secretToken"
- name: eventTypes
value:
- "Push Hook"
bindings:
- name: gitrevision
value: $(body.checkout_sha)
- name: gitrepositoryurl
value: $(body.repository.git_http_url)
template:
spec:
params:
- name: gitrevision
- name: gitrepositoryurl
- name: message
description: The message to print
default: This is the default message
- name: contenttype
description: The Content-Type of the event
resourcetemplates:
- apiVersion: tekton.dev/v1beta1
kind: PipelineRun
metadata:
generateName: buildpacks-test-pipeline-run-
#name: buildpacks-test-pipeline-run
spec:
serviceAccountName: buildpacks-service-account-gitlab # Only needed if you set up authorization
pipelineRef:
name: buildpacks-test-pipeline
workspaces:
- name: source-workspace
subPath: source
persistentVolumeClaim:
claimName: buildpacks-source-pvc
- name: cache-workspace
subPath: cache
persistentVolumeClaim:
claimName: buildpacks-source-pvc
params:
- name: IMAGE
value: registry.gitlab.com/jonashackt/microservice-api-spring-boot # This defines the name of output image
- name: SOURCE_URL
value: https://gitlab.com/jonashackt/microservice-api-spring-boot
- name: SOURCE_REVISION
value: main
Now apply it to our cluster via:
kubectl apply -f tekton/triggers/gitlab-push-listener.yml
Tekton Triggers creates a new Deployment and Service for the EventListener
As stated in https://tekton.dev/docs/triggers/eventlisteners/#understanding-the-deployment-of-an-eventlistener
Tekton Triggers uses the EventListener name prefixed with el- to name the Deployment and Service when instantiating them.
this will also deploy a K8s Service
called el-gitlab-listener
and a Deployment
also called el-gitlab-listener
:
$ k get deployment
NAME READY UP-TO-DATE AVAILABLE AGE
el-gitlab-listener 1/1 1 1 25h
$ k get service
NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE
el-gitlab-listener ClusterIP 10.100.101.207 <none> 8080/TCP,9000/TCP 12m
...
What we need here is an example test event as .json
file. For an example see https://github.com/tektoncd/triggers/blob/main/examples/v1beta1/gitlab/gitlab-push-event.json
But we can craft this file ourselves while creating the GitLab Webhook.
Therefore go to your project (in my example here this is https://gitlab.com/jonashackt/microservice-api-spring-boot) and head over to Settings/Webhooks.
Now insert our a fake URL like http://www.google.com
into the URL field (we need a semantically correct url here in order to be able to save the Webhook, but will change it later to the EventListeners Ingress URL).
Also insert our Secret Token 1234567
in the Secret Token field.
Finally choose Push events, deselect SSL verification and scroll down and hit Add webhook.
This should result in the Webhook beeing created and listed right at the bottom of this page (scroll down again) in a list called Project Hooks
.
Click on Test and choose Push events. This will fire a test event (which will produce a HTTP error). Now scroll down again an click on edit. Scroll down inside our WebHook (don't get confused with the UI :)) and you should see the Recent events list:
Choose the latest event and click on View details
. Scroll down again and you should see the Request frame with the generated request json:
Copy the whole part into a local file. In this example project this file is called gitlab-push-test-event.json:
{
"object_kind": "push",
"event_name": "push",
"before": "5bbc8580432fc7a16f50be27eb513db42aad0860",
"after": "c25a74c8f919a72e3f00928917dc4ab2944ab061",
"ref": "refs/heads/main",
"checkout_sha": "c25a74c8f919a72e3f00928917dc4ab2944ab061",
"message": null,
"user_id": 2351133,
"user_name": "Jonas Hecht",
"user_username": "jonashackt",
"user_email": "",
"user_avatar": "https://secure.gravatar.com/avatar/a5c83d481ac20557b775703761aef7dc?s=80&d=identicon",
"project_id": 30444286,
"project": {
"id": 30444286,
"name": "microservice-api-spring-boot",
"description": "Forked from https://github.com/jonashackt/microservice-api-spring-boot",
"web_url": "https://gitlab.com/jonashackt/microservice-api-spring-boot",
"avatar_url": null,
"git_ssh_url": "[email protected]:jonashackt/microservice-api-spring-boot.git",
"git_http_url": "https://gitlab.com/jonashackt/microservice-api-spring-boot.git",
"namespace": "Jonas Hecht",
"visibility_level": 20,
"path_with_namespace": "jonashackt/microservice-api-spring-boot",
"default_branch": "main",
"ci_config_path": "",
"homepage": "https://gitlab.com/jonashackt/microservice-api-spring-boot",
"url": "[email protected]:jonashackt/microservice-api-spring-boot.git",
"ssh_url": "[email protected]:jonashackt/microservice-api-spring-boot.git",
"http_url": "https://gitlab.com/jonashackt/microservice-api-spring-boot.git"
},
"commits": [
{
"id": "c25a74c8f919a72e3f00928917dc4ab2944ab061",
"message": "Fixing cache image naming\n",
"title": "Fixing cache image naming",
"timestamp": "2021-10-19T10:32:58+02:00",
"url": "https://gitlab.com/jonashackt/microservice-api-spring-boot/-/commit/c25a74c8f919a72e3f00928917dc4ab2944ab061",
"author": {
"name": "Jonas Hecht",
"email": "[email protected]"
},
"added": [
],
"modified": [
".gitlab-ci.yml",
"README.md"
],
"removed": [
]
},
{
"id": "06a7f1d2ad646acef149b1aad4600eb2b1268f0c",
"message": "Merge branch 'refactor-ci' into 'main'\n\nRefactor ci\n\nSee merge request jonashackt/microservice-api-spring-boot!2",
"title": "Merge branch 'refactor-ci' into 'main'",
"timestamp": "2021-10-19T08:30:46+00:00",
"url": "https://gitlab.com/jonashackt/microservice-api-spring-boot/-/commit/06a7f1d2ad646acef149b1aad4600eb2b1268f0c",
"author": {
"name": "Jonas Hecht",
"email": "[email protected]"
},
"added": [
],
"modified": [
".gitlab-ci.yml",
"README.md"
],
"removed": [
]
},
{
"id": "5bbc8580432fc7a16f50be27eb513db42aad0860",
"message": "Add cache image\n",
"title": "Add cache image",
"timestamp": "2021-10-19T10:05:31+02:00",
"url": "https://gitlab.com/jonashackt/microservice-api-spring-boot/-/commit/5bbc8580432fc7a16f50be27eb513db42aad0860",
"author": {
"name": "Jonas Hecht",
"email": "[email protected]"
},
"added": [
],
"modified": [
".gitlab-ci.yml",
"README.md"
],
"removed": [
]
}
],
"total_commits_count": 3,
"push_options": {
},
"repository": {
"name": "microservice-api-spring-boot",
"url": "[email protected]:jonashackt/microservice-api-spring-boot.git",
"description": "Forked from https://github.com/jonashackt/microservice-api-spring-boot",
"homepage": "https://gitlab.com/jonashackt/microservice-api-spring-boot",
"git_http_url": "https://gitlab.com/jonashackt/microservice-api-spring-boot.git",
"git_ssh_url": "[email protected]:jonashackt/microservice-api-spring-boot.git",
"visibility_level": 20
}
}
Port forward with a new Service
locally:
kubectl port-forward service/el-gitlab-listener 8080
Now test-drive the trigger via curl:
curl -v \
-H 'X-GitLab-Token: 1234567' \
-H 'X-Gitlab-Event: Push Hook' \
-H 'Content-Type: application/json' \
--data-binary "@tekton/triggers/gitlab-push-test-event.json" \
http://localhost:8080
Let's use our Ingress with Traefik and the nice Route53 domain & wildcard record to route from gitlab-listener.tekton-argocd.de. Simply create an Traefik IngressRoute
as described in traefik/gitlab-listener.yml:
apiVersion: traefik.containo.us/v1alpha1
kind: IngressRoute
metadata:
name: gitlab-listener
spec:
entryPoints:
- web
routes:
- kind: Rule
match: Host(`gitlab-listener.tekton-argocd.de`)
services:
- name: el-gitlab-listener
port: 8080
Apply it with kubectl apply -f traefik/gitlab-listener.yml
and it should be ready for access at http://gitlab-listener.tekton-argocd.de
Now let's try our curl
using the predefined gitlab-push-test-event.json:
TEKTON_TRIGGER_GITLAB_LISTENER="http://gitlab-listener.tekton-argocd.de"
curl -v \
-H 'X-GitLab-Token: 1234567' \
-H 'X-Gitlab-Event: Push Hook' \
-H 'Content-Type: application/json' \
--data-binary "@tekton/triggers/gitlab-push-test-event.json" \
$TEKTON_TRIGGER_GITLAB_LISTENER
Finally we can implement all this inside our GitHub Action workflow .github/workflows/provision.yml:
- name: Expose Tekton Triggers EventListener as Traefik IngressRoute & testdrive Trigger
run: |
echo "--- Apply Tekton EventListener Traefik IngressRoute"
kubectl apply -f traefik/gitlab-listener.yml
echo "--- Testdrive Trigger via curl"
curl -v \
-H 'X-GitLab-Token: 1234567' \
-H 'X-Gitlab-Event: Push Hook' \
-H 'Content-Type: application/json' \
--data-binary "@tekton/triggers/gitlab-push-test-event.json" \
http://gitlab-listener.$ROUTE53_DOMAIN_NAME
We now should extend our gitlab-push-listener.yml to use the values send by the GitLab Webhook via json.
Our file now looks like this:
apiVersion: triggers.tekton.dev/v1beta1
kind: EventListener
metadata:
name: gitlab-listener
spec:
serviceAccountName: tekton-triggers-example-sa
triggers:
- name: gitlab-push-events-trigger
interceptors:
- name: "verify-gitlab-payload"
ref:
name: "gitlab"
kind: ClusterInterceptor
params:
- name: secretRef
value:
secretName: "gitlab-secret"
secretKey: "secretToken"
- name: eventTypes
value:
- "Push Hook"
bindings:
- name: gitrevision
value: $(body.checkout_sha)
- name: gitrepositoryurl
value: $(body.repository.git_http_url)
- name: gitrepository_pathonly
value: $(body.project.path_with_namespace)
template:
spec:
params:
- name: gitrevision
- name: gitrepositoryurl
- name: gitrepository_pathonly
- name: message
description: The message to print
default: This is the default message
- name: contenttype
description: The Content-Type of the event
resourcetemplates:
- apiVersion: tekton.dev/v1beta1
kind: PipelineRun
metadata:
generateName: buildpacks-test-pipeline-run-
spec:
serviceAccountName: buildpacks-service-account-gitlab
pipelineRef:
name: buildpacks-test-pipeline
workspaces:
- name: source-workspace
subPath: source
persistentVolumeClaim:
claimName: buildpacks-source-pvc
- name: cache-workspace
subPath: cache
persistentVolumeClaim:
claimName: buildpacks-source-pvc
params:
- name: IMAGE
value: "registry.gitlab.com/$(tt.params.gitrepository_pathonly)" #here our GitLab's registry url must be configured
- name: SOURCE_URL
value: $(tt.params.gitrepositoryurl)
- name: SOURCE_REVISION
value: $(tt.params.gitrevision)
There's not so much that changes here. The param SOURCE_URL
will be provided the git repo url via $(tt.params.gitrepositoryurl)
- this is done using the tt.params
notation as the docs state.
The same is used for SOURCE_REVISION
. Both parameters must be defined in the template:spec:params
section also.
And they must all be defined in the TriggerBinding inside the bindings
section. Here we read the values from the parsed json request from GitLab using the $(body.key1)
notation - see the docs on TriggerBindings json payload access.
Finally we also read another value in the bindings
section: the gitrepository_pathonly
value will be obtained from $(body.project.path_with_namespace)
- which represents our GitLab repo's group and repo names. In this example this is jonashackt/microservice-api-spring-boot
.
But what do we need this value for? You'll see it inside the IMAGE
parameter of the PipelineRun
definition inside the TriggerTemplate.
With value: "registry.gitlab.com/$(tt.params.gitrepository_pathonly)"
we use the gitrepository_pathonly
to craft the correct GitLab Container Registry URL with the predefined GitLab CR domain name and the appended group and repo name.
Go to your project (in my example here this is https://gitlab.com/jonashackt/microservice-api-spring-boot) and head over to Settings/Webhooks.
Now insert our Tekton Triggers EventListener URL http://gitlab-listener.tekton-argocd.de into the already created Webhook's URL field ().
You need to scroll down to Project Hooks and edit your existing Webhook.
Finally Test-drive the Webhook again choosing Test and Push event from the drop down.
Switch to the Tekton Dashboard in your Browser and you should see the PipelineRun triggered:
The last step in our journey of integrating GitLab with Tekton is to report the status of our Tekton Pipelines back to GitLab.
There are multiple options. Let's first start simple using the Tekton Hub task https://hub.tekton.dev/tekton/task/gitlab-set-status
Inside our provision.yml workflow we need to install the gitlab-set-status Task:
kubectl apply -f https://raw.githubusercontent.com/tektoncd/catalog/main/task/gitlab-set-status/0.1/gitlab-set-status.yaml
To access the GitLab commit API (see https://docs.gitlab.com/ee/api/commits.html#post-the-build-status-to-a-commit) using the gitlab-set-status
task we need to create an access token as stated in https://docs.gitlab.com/ee/api/index.html#authentication
On self-managed GitLab instances you can create project access tokens for example.
Using gitlab.com we cannot use project access tokens, but can create personal access tokens instead: https://docs.gitlab.com/ee/user/profile/personal_access_tokens.html#create-a-personal-access-token
Therefore head over to Edit profile and choose Access Tokens on the left. Then create a token called gitlab-api-token
for example:
Now since we're using GitHub Actions to provision our EKS cluster and install Tekton, we need to enhance our provision.yml workflow.
First create a new GitHub repository secret GITLAB_API_TOKEN
containing the personal access token which we just created:
Now we can use these GitHub repo secrets to create the actual Kubernetes secret in our provision.yml workflow:
kubectl create secret generic gitlab-api-secret \
--from-literal=token=${{ secrets.GITLAB_API_TOKEN }} \
--namespace default \
--save-config --dry-run=client -o yaml | kubectl apply -f -
Using the Tekton Hub task https://hub.tekton.dev/tekton/task/gitlab-set-status we can create a new step inside our Tekton Pipeline. But first ne need to create some new parameters for our Pipeline:
params:
...
- name: REPO_PATH_ONLY
type: string
description: GitLab group & repo name only (e.g. jonashackt/microservice-api-spring-boot)
...
- name: GITLAB_HOST
type: string
description: Your GitLabs host only (e.g. gitlab.com)
- name: TEKTON_DASHBOARD_HOST
type: string
description: The Tekton dashboard host name only
Now we can implement the task:
- name: report-pipeline-end-to-gitlab
taskRef:
name: "gitlab-set-status"
runAfter:
- buildpacks
params:
- name: "STATE"
value: "success"
- name: "GITLAB_HOST_URL"
value: "$(params.GITLAB_HOST)"
- name: "REPO_FULL_NAME"
value: "$(params.REPO_PATH_ONLY)"
- name: "GITLAB_TOKEN_SECRET_NAME"
value: "gitlab-api-secret"
- name: "GITLAB_TOKEN_SECRET_KEY"
value: "token"
- name: "SHA"
value: "$(params.SOURCE_REVISION)"
- name: "TARGET_URL"
value: "$(params.TEKTON_DASHBOARD_HOST)/#/namespaces/default/pipelineruns/$(context.pipelineRun.name)"
- name: "CONTEXT"
value: "tekton-pipeline"
- name: "DESCRIPTION"
value: "Finished building your commit in Tekton"
First we should make sure this task runs only after the buildpacks
task using runAfter
.
Then we need to provide the GITLAB_HOST_URL
and REPO_FULL_NAME
.
Also the GITLAB_TOKEN_SECRET_NAME
needs to refer to the Kubernetes secret gitlab-api-secret
we created leveraging a Personal Access Token. The GITLAB_TOKEN_SECRET_KEY
must reference to the key
name inside the kubectl create secret
command, where we used --from-literal=token=${{ secrets.GITLAB_API_TOKEN }}
. So the GITLAB_TOKEN_SECRET_KEY
is token
here (which is also the default).
The TARGET_URL
is a bit more tricky and needs to be crafted with care:
- name: "TARGET_URL"
value: "$(params.TEKTON_DASHBOARD_HOST)/#/namespaces/default/pipelineruns/$(context.pipelineRun.name)"
It consists of the Tekton dashboard host $(params.TEKTON_DASHBOARD_HOST)
and the Pipeline details prefix /#/namespaces/default/pipelineruns/
. The $(context.pipelineRun.name)
finally gives us the current PipelineRun's name we need to be able to reference the actual PipelineRun from the GitLab UI.
Finally the CONTEXT
and DESCRIPTION
should contain useful information to be displayed in the GitLab UI later:
Our pipeline-run.yml (for manual triggering):
...
params:
- name: IMAGE
value: registry.gitlab.com/jonashackt/microservice-api-spring-boot # This defines the name of output image
- name: SOURCE_URL
value: https://gitlab.com/jonashackt/microservice-api-spring-boot
- name: REPO_PATH_ONLY
value: jonashackt/microservice-api-spring-boot
- name: SOURCE_REVISION
value: main
- name: GITLAB_HOST
value: gitlab.com
- name: TEKTON_DASHBOARD_HOST
value: http://abd1c6f235c9642bf9d4cdf632962298-1232135946.eu-central-1.elb.amazonaws.com
and the EventListener (for automatic triggering by our gitlab.com projects) need to pass some new parameters in order to get the gitlab-set-status
task working:
params:
- name: IMAGE
value: "registry.gitlab.com/$(tt.params.gitrepository_pathonly)" #here our GitLab's registry url must be configured
- name: SOURCE_URL
value: $(tt.params.gitrepositoryurl)
- name: REPO_PATH_ONLY
value: $(tt.params.gitrepository_pathonly)
- name: SOURCE_REVISION
value: $(tt.params.gitrevision)
- name: GITLAB_HOST
value: gitlab.com
- name: TEKTON_DASHBOARD_HOST
value: {{TEKTON_DASHBOARD_HOST}}
Adding the REPO_PATH_ONLY
is no problem, since we alread used $(tt.params.gitrepository_pathonly)
inside the IMAGE
parameter. As with the GitLab registry url we can also "hard code" the GITLAB_HOST
here for now.
The TEKTON_DASHBOARD_HOST
is the trickiest part, since we need to substitute this value from outside of the Tekton Trigger process, which doesn't know about the Tekton dashboard url.
But luckily inside our GitHub Actions provision.yml workflow we can use https://stackoverflow.com/questions/48296082/how-to-set-dynamic-values-with-kubernetes-yaml-file/70152914#70152914:
echo "--- Insert Tekton dashboard url into EventListener config and apply it (see https://stackoverflow.com/a/70152914/4964553)"
TEKTON_DASHBOARD_HOST="${{ steps.dashboard-expose.outputs.dashboard_host }}"
sed "s#{{TEKTON_DASHBOARD_HOST}}#$TEKTON_DASHBOARD_HOST#g" tekton/triggers/gitlab-push-listener.yml | kubectl apply -f -
Using sed we simply replace {{TEKTON_DASHBOARD_HOST}}
with the already defined GitHub Actions variable ${{ steps.dashboard-expose.outputs.dashboard_host }}
.
Testing our full workflow is simple pushing a change to our repo using a branch without GitLab CI: https://gitlab.com/jonashackt/microservice-api-spring-boot/-/commits/trigger-tekton-via-webhook
Now our GitLab Pipelines view gets filled with a Tekton Pipelines status (success
only for now):
Now that we generally know how to use the gitlab-set-status
Task, we could also use another Task definition to report the starting of a Tekton Pipeline run to GitLab UI.
Therefore we enhance our Tekton Pipeline with a new Task starting the whole Pipeline called report-pipeline-start-to-gitlab
:
tasks:
- name: report-pipeline-start-to-gitlab
taskRef:
name: gitlab-set-status
params:
- name: "STATE"
value: "running"
- name: "GITLAB_HOST_URL"
value: "$(params.GITLAB_HOST)"
- name: "REPO_FULL_NAME"
value: "$(params.REPO_PATH_ONLY)"
- name: "GITLAB_TOKEN_SECRET_NAME"
value: "gitlab-api-secret"
- name: "GITLAB_TOKEN_SECRET_KEY"
value: "token"
- name: "SHA"
value: "$(params.SOURCE_REVISION)"
- name: "TARGET_URL"
value: "$(params.TEKTON_DASHBOARD_HOST)/#/namespaces/default/pipelineruns/$(context.pipelineRun.name)"
- name: "CONTEXT"
value: "tekton-pipeline"
- name: "DESCRIPTION"
value: "Building your commit in Tekton"
- name: fetch-repository # This task fetches a repository from github, using the `git-clone` task you installed
taskRef:
name: git-clone
runAfter:
- report-pipeline-start-to-gitlab
And the fetch-repository
only starts after the status is reported to GitLab. With this new Task in place a push to our repository https://gitlab.com/jonashackt/microservice-api-spring-boot directly presents a running
Pipeline inside the GitLab UI:
And in the details view we can directly access our running Tekton Pipeline via a correct Tekton dashboard link:
How do we catch all status from our Tekton pipeline and then report based on that to GitLab?
In v0.14 Tekton introduced finally Tasks, which run at the end of every PipelineRun - regardless which Task failed or succeeded. As the docs state:
finally tasks are guaranteed to be executed in parallel after all PipelineTasks under tasks have completed regardless of success or error.
Finally tasks look like this:
spec:
tasks:
- name: tests
taskRef:
name: integration-test
finally:
- name: cleanup-test
taskRef:
name: cleanup
With Guard[ing] finally Task execution using when expressions we can enhance this even further.
Because using when
expressions we can run Tasks based on the overall Pipeline status (or Aggregate Pipeline status) - see https://tekton.dev/docs/pipelines/pipelines/#when-expressions-using-aggregate-execution-status-of-tasks-in-finally-tasks
finally:
- name: notify-any-failure # executed only when one or more tasks fail
when:
- input: $(tasks.status)
operator: in
values: ["Failed"]
taskRef:
name: notify-failure
The Aggregate Execution Status we can grap using $(tasks.status)
is stated to have those 4 possible status:
Succeeded
("all tasks have succeeded") or Completed
("all tasks completed successfully including one or more skipped tasks"), which could be translated into the gitlab-set-status
Tasks STATE
value success
.
And Failed
("one ore more tasks failed") or None
("no aggregate execution status available (i.e. none of the above), one or more tasks could be pending/running/cancelled/timedout"), which could both be translated into the gitlab-set-status
Tasks STATE
value failed
. For None
this is only valid, since we're in a finally task
, since pending/running
could otherwise also mean that a Pipeline is in a good state.
Luckily the when
expressions
So we're able to do
when:
- input: $(tasks.status)
operator: in
values: [ "Failed", "None" ]
and
when:
- input: $(tasks.status)
operator: in
values: [ "Succeeded", "Completed" ]
In the end this results in our Tekton Pipeline's finally
block locking like this:
...
finally:
- name: report-pipeline-failed-to-gitlab
when:
- input: $(tasks.status)
operator: in
values: [ "Failed", "None" ] # see aggregated status https://tekton.dev/docs/pipelines/pipelines/#using-aggregate-execution-status-of-all-tasks
taskRef:
name: "gitlab-set-status"
params:
- name: "STATE"
value: "failed"
- name: "GITLAB_HOST_URL"
value: "$(params.GITLAB_HOST)"
- name: "REPO_FULL_NAME"
value: "$(params.REPO_PATH_ONLY)"
- name: "GITLAB_TOKEN_SECRET_NAME"
value: "gitlab-api-secret"
- name: "GITLAB_TOKEN_SECRET_KEY"
value: "token"
- name: "SHA"
value: "$(params.SOURCE_REVISION)"
- name: "TARGET_URL"
value: "$(params.TEKTON_DASHBOARD_HOST)/#/namespaces/default/pipelineruns/$(context.pipelineRun.name)"
- name: "CONTEXT"
value: "tekton-pipeline"
- name: "DESCRIPTION"
value: "An error occurred building your commit in Tekton"
- name: report-pipeline-success-to-gitlab
when:
- input: $(tasks.status)
operator: in
values: [ "Succeeded", "Completed" ] # see aggregated status https://tekton.dev/docs/pipelines/pipelines/#using-aggregate-execution-status-of-all-tasks
taskRef:
name: "gitlab-set-status"
params:
- name: "STATE"
value: "success"
- name: "GITLAB_HOST_URL"
value: "$(params.GITLAB_HOST)"
- name: "REPO_FULL_NAME"
value: "$(params.REPO_PATH_ONLY)"
- name: "GITLAB_TOKEN_SECRET_NAME"
value: "gitlab-api-secret"
- name: "GITLAB_TOKEN_SECRET_KEY"
value: "token"
- name: "SHA"
value: "$(params.SOURCE_REVISION)"
- name: "TARGET_URL"
value: "$(params.TEKTON_DASHBOARD_HOST)/#/namespaces/default/pipelineruns/$(context.pipelineRun.name)"
- name: "CONTEXT"
value: "tekton-pipeline"
- name: "DESCRIPTION"
value: "Finished building your commit in Tekton"
Executing our Tekton Pipeline should now be reported correctly to our GitLab. Failures look like this:
and in the detail view:
The solution is based on https://stackoverflow.com/questions/70156006/report-tekton-pipeline-status-to-gitlab-regardless-if-pipeline-failed-or-succeed/70156007#70156007.
Right now we have a huge pipeline with only 2-3 relevant Tekton tasks, the other 3 Tasks are solely used to communicate the pipeline's status to GitLab.
So is there a way we could refactor these to a more generic form? I digged into PipelineResources
as I thought they could be a nice option to handle this. But sadly they didn't make it into the Tekton beta: https://tekton.dev/docs/pipelines/migrating-v1alpha1-to-v1beta1/#replacing-pipelineresources-with-tasks
That's why we need another way to accomplish our refactoring. So what about simply using Tasks that reference other Tasks? Sadly that doesn't seem to be possible. Since Tasks only specify steps
not tasks
with taskRef
.
But how about using Pipelines using other Pipelines? Is this possible? Yes, but currently only as experimental: https://tekton.dev/docs/pipelines/pipelines/#compose-using-pipelines-in-pipelines & https://github.com/tektoncd/experimental/blob/main/pipelines-in-pipelines/examples/pipelinerun-with-pipeline-in-pipeline.yaml
As stated we can create a normal Tekton Pipeline as we're already used to - and then use this Pipeline in another Pipeline simply by using taskRef
with kind: Pipeline
and apiVersion: tekton.dev/v1beta1
accompanying the referenced pipeline name:
- name: reference-other-pipeline
taskRef:
apiVersion: tekton.dev/v1beta1
kind: Pipeline
name: other-pipeline-name
In order to be able to use this feature, we need to install the Pipelines-In-Pipelines Controller
with (but be aware of tektoncd/experimental#817 which is fixed by tektoncd/experimental#818)
kubectl apply --filename https://storage.googleapis.com/tekton-releases-nightly/pipelines-in-pipelines/latest/release.yaml
Now running our pipelines would result in the following error:
Pipeline default/buildpacks-test-pipeline can't be Run; it contains Tasks that don't exist: Couldn't retrieve Task "generic-gitlab-set-status": tasks.tekton.dev "generic-gitlab-set-status" not found
That's because the Pipeline-in-Pipelines feature is an alpha feature - see this issue tektoncd/experimental#785
In order to activate Tekton alpha features we need to Customize the Pipeline Controllers behavior.
As stated in https://stackoverflow.com/a/70336211/4964553 this could be done on-the-fly with curl
and sed
piped into kubectl apply
like this:
curl https://storage.googleapis.com/tekton-releases/pipeline/latest/release.yaml | sed "s#stable#alpha#g" | kubectl apply -f -
Let's try to create a generic Tekton Pipeline for the gitlab-set-status
as generic-gitlab-set-status.yml:
apiVersion: tekton.dev/v1beta1
kind: Pipeline
metadata:
name: generic-gitlab-set-status
spec:
params:
- name: STATE
type: string
description: The gitlab-set-status Tasks state to set. Can be one of the following pending, running, success, failed, or canceled.
- name: PIPELINE_NAME
type: string
description: The calling pipelines name.
- name: REPO_PATH_ONLY
type: string
description: GitLab group & repo name only (e.g. jonashackt/microservice-api-spring-boot)
- name: SOURCE_REVISION
description: The branch, tag or SHA to checkout.
default: ""
- name: GITLAB_TOOLTIP
type: string
description: The tooltip to be shown in the GitLab Pipelines details view.
tasks:
- name: report-pipeline-start-to-gitlab
taskRef:
name: gitlab-set-status
params:
- name: "STATE"
value: "$(params.STATE)"
- name: "GITLAB_HOST_URL"
value: "gitlab.com"
- name: "REPO_FULL_NAME"
value: "$(params.REPO_PATH_ONLY)"
- name: "GITLAB_TOKEN_SECRET_NAME"
value: "gitlab-api-secret"
- name: "GITLAB_TOKEN_SECRET_KEY"
value: "token"
- name: "SHA"
value: "$(params.SOURCE_REVISION)"
- name: "TARGET_URL"
value: "{{TEKTON_DASHBOARD_HOST}}/#/namespaces/default/pipelineruns/$(params.PIPELINE_NAME)"
- name: "CONTEXT"
value: "tekton-pipeline"
- name: "DESCRIPTION"
value: "$(params.GITLAB_TOOLTIP)"
As you can see we define the TEKTON_DASHBOARD_HOST
using brackets so we can later use sed
to dynamically set the actual Tekton Dashboard URL in our GitHub Actions pipeline (using sed
for this is also described in https://stackoverflow.com/a/70152914/4964553)
TEKTON_DASHBOARD_HOST="${{ steps.dashboard-expose.outputs.dashboard_host }}"
sed "s#{{TEKTON_DASHBOARD_HOST}}#$TEKTON_DASHBOARD_HOST#g" tekton/pipelines/generic-gitlab-set-status.yml | kubectl apply -f -
In our pipeline.yml we can now reduce many lines that we don't need to pass to the generic gitlab-set-status pipeline any more. So our pipeline becomes much more readable and only the things remain that are naturally defined inside a pipeline. See the usage of our generic gitlab-set-status pipeline here using the Pipelines-in-Pipelines feature:
- name: report-pipeline-start-to-gitlab
taskRef:
apiVersion: tekton.dev/v1beta1
kind: Pipeline
name: generic-gitlab-set-status
params:
- name: "STATE"
value: "running"
- name: "REPO_PATH_ONLY"
value: "$(params.REPO_PATH_ONLY)"
- name: "SHA"
value: "$(params.SOURCE_REVISION)"
- name: "GITLAB_TOOLTIP"
value: "Building your commit in Tekton"
- name: "PIPELINE_NAME"
value: "$(context.pipelineRun.name)"
As the Pipeline-in-Pipeline feature is in alpha state, this is really cool to see it working. In a future version there might be the option to also remove the PIPELINE_NAME
parameter (but currently I see no option for that https://github.com/tektoncd/experimental/tree/main/pipelines-in-pipelines).
The code needed to invoke the gitlab-set-status task is reduced all the way:
Finally we can also remove code from our pipeline-run.yml and gitlab-push-listener.yml:
- name: GITLAB_HOST
value: gitlab.com
- name: TEKTON_DASHBOARD_HOST
value: {{TEKTON_DASHBOARD_HOST}}
because this is now centrally configured in our generic pipeline :)
There's maybe one thing that could be considered as downside: One logic pipeline now triggers 3 PipelineRuns - those are also shown in the Tekton Dashboard:
We want to implement the "Pull-based" deployment approach in GitOps using ArgoCD.
brew install argocd
You can install ArgoCD as described in the getting started guide - but that will lead you to some problems together with Traefik:
You can expose the ArgoCD API Server via Loadbalancer, Ingress or port forwarding to localhost: https://argo-cd.readthedocs.io/en/stable/getting_started/#3-access-the-argo-cd-api-server
https://argo-cd.readthedocs.io/en/stable/operator-manual/ingress/#ingress-configuration
Argo CD runs both a gRPC server (used by the CLI), as well as a HTTP/HTTPS server (used by the UI). Both protocols are exposed by the argocd-server service object on the following ports:
443 - gRPC/HTTPS & 80 - HTTP (redirects to HTTPS)
There are several ways how Ingress can be configured
So let's use Ingress with our Traefik and the nice Route53 domain & wildcard record to route from argocd.tekton-argocd.de. Simply create an Traefik IngressRoute
as described in traefik/argocd-dashboard.yml:
As of writing the exact
IngressRoute
from the docs produces an error:
$ kubectl apply -f traefik/argocd-dashboard.yml
error: error validating "traefik/argocd-dashboard.yml": error validating data: ValidationError(IngressRoute.spec.tls.options): missing required field "name" in us.containo.traefik.v1alpha1.IngressRoute.spec.tls.options; if you choose to ignore these errors, turn validation off with --validate=false
apiVersion: traefik.containo.us/v1alpha1
kind: IngressRoute
metadata:
name: argocd-server
namespace: argocd
spec:
entryPoints:
- websecure
routes:
- kind: Rule
match: Host(`argocd.tekton-argocd.de`)
priority: 10
services:
- name: argocd-server
port: 80
- kind: Rule
match: Host(`argocd.tekton-argocd.de`) && Headers(`Content-Type`, `application/grpc`)
priority: 11
services:
- name: argocd-server
port: 80
scheme: h2c
tls:
certResolver: default
With this approach we also don't need to kubectl patch svc argocd-server -n argocd -p '{"spec": {"type": "LoadBalancer"}}'
to use a LoadBalancer
, which provisions an AWS ELB (incl. additional costs).
Apply it with kubectl apply -f traefik/argocd-dashboard.yml
. Now ArgoCD should be accessible via http://argocd.tekton-argocd.de
If we installed ArgoCD as described in the getting started guide by using kubectl apply -f
we will run into HTTP 307
redirects! What's the problem here?
See argoproj/argo-cd#2953 (comment)
The problem is that by default Argo-CD handles TLS termination itself and always redirects HTTP requests to HTTPS. Combine that with an ingress controller that also handles TLS termination and always communicates with the backend service with HTTP and you get Argo-CD's server always responding with a redirects to HTTPS.
And the ArgoCD docs for Traefik Ingress configuration at https://argo-cd.readthedocs.io/en/stable/operator-manual/ingress/#traefik-v22 tell us to
The API server should be run with TLS disabled. Edit the argocd-server deployment to add the --insecure flag to the argocd-server command.
But How to configure argocd-server Deployment to run with TLS disabled (where to put --insecure flag)?
As stated in this answer there's great way to manage custom configuration for Kubernetes deployments like ArgoCD by using Kustomize!
A great way is to use a declarative approach, which should be the default Kubernetes-style. Skimming the ArgoCD docs there's a additional configuration section where the possible flags of the ConfigMap argocd-cmd-params-cm
can be found. The flags are described in argocd-cmd-params-cm.yaml. One of them is the flag server.insecure
## Server properties
# Run server without TLS
server.insecure: "false"
The argocd-server
deployment which ships with https://raw.githubusercontent.com/argoproj/argo-cd/stable/manifests/install.yaml will use this parameter, if it is defined in the argocd-cmd-params-cm
ConfigMap.
In order to declaratively configure the ArgoCD configuration, the ArgoCD docs have a great section on how to do that with Kustomize. In fact the ArgoCD team itself uses this approach to deploy their own ArgoCD instances - a live deployment is available here https://cd.apps.argoproj.io/ and the configuration used can be found on GitHub.
Adopting this to our use case, we need to switch our ArgoCD installation from simply using kubectl apply -f
to a Kustomize-based installation. The ArgoCD docs also have a section on how to do this. Here are the brief steps:
We slightly enhance the kustomization.yaml
proposed in the docs and create it inside argocd/install/kustomization.yaml:
apiVersion: kustomize.config.k8s.io/v1beta1
kind: Kustomization
resources:
- https://raw.githubusercontent.com/argoproj/argo-cd/v2.3.3/manifests/install.yaml
## changes to config maps
patchesStrategicMerge:
- argocd-cmd-params-cm-patch.yml
namespace: argocd
Since the docs state
It is recommended to include the manifest as a remote resource and apply additional customizations using Kustomize patches.
we use the patchesStrategicMerge
configuration key, which contains another new file we need to create called argocd-cmd-params-cm-patch.yml
.
This new argocd/install/argocd-cmd-params-cm-patch.yml only contains the configuration we want to change inside the ConfigMap argocd-cmd-params-cm
:
apiVersion: v1
kind: ConfigMap
metadata:
name: argocd-cmd-params-cm
data:
server.insecure: "true"
There's a separate kustomize
CLI one can install e.g. via brew install kustomize
. But as Kustomize is build into kubectl
we only have to use kubectl apply -k
and point that to our newly created argocd/installation
directory like this:
kubectl create namespace argocd --dry-run=client -o yaml | kubectl apply -f -
kubectl apply -k argocd/install
As you can see we also need to make sure the namespace argocd
is present before Kustomize can apply all the ArgoCD resources.
This will install ArgoCD and configure the argocd-server
deployment to use the --insecure
flag as needed to stop Argo from handling the TLS termination itself and giving that responsibility to Traefik.
Now accessing https://argocd.tekton-argocd.de should open the ArgoCD dashboard as expected:
Obtain ArgoCD admin account's initial password
kubectl -n argocd get secret argocd-initial-admin-secret -o jsonpath="{.data.password}" | base64 -d; echo
Login to argocd-server
hostname, which was exposed as IngressRoute through Traefik - using username admin
and the initial password obtained above:
argocd login argocd.tekton-argocd.de
This will result in a hopefully successful login like this:
argocd login yourservername.eu-central-1.elb.amazonaws.com
WARNING: server certificate had error: x509: certificate is valid for localhost, argocd-server, argocd-server.argocd, argocd-server.argocd.svc, argocd-server.argocd.svc.cluster.local, not argocd.tekton-argocd.de. Proceed insecurely (y/n)? y
Username: admin
Password:
'admin:login' logged in successfully
Context 'argocd.tekton-argocd.de' updated
Finally change the initial password via:
argocd account update-password
Using your Browser open argocd.tekton-argocd.de
and accept the certificate warnings. Then sign in using the admin
user credentials from above:
If you don't want to deploy to a different Kubernetes cluster than the current one where Argo was installed, then everything should be prepared to deploy our first application.
Let's just create a new GitHub Actions job for this purpose, which also needs the kubeconfig
from the first Pulumi task which bootstraps our EKS cluster:
install-and-run-argocd-on-eks:
runs-on: ubuntu-latest
needs: provision-eks-with-pulumi
environment:
name: argocd-dashboard
url: ${{ steps.dashboard-expose.outputs.dashboard_host }}
steps:
- name: Checkout
uses: actions/checkout@master
# We must use single quotes (!!!) here to access the kubeconfig like this:
# echo '${{ needs.provision-eks-with-pulumi.outputs.kubeconfig }}' > ~/.kube/config
# Otherwise we'll run into errors like (see https://stackoverflow.com/a/15930393/4964553):
# "error: error loading config file "/home/runner/.kube/config": yaml: did not find expected ',' or '}'"
- name: Configure kubeconfig to use with kubectl from provisioning job
run: |
mkdir ~/.kube
echo '${{ needs.provision-eks-with-pulumi.outputs.kubeconfig }}' > ~/.kube/config
echo "--- Checking connectivity to cluster"
kubectl get nodes
- name: Install ArgoCD
run: |
echo "--- Create argo namespace and install it"
kubectl create namespace argocd --dry-run=client -o yaml | kubectl apply -f -
echo "--- Install & configure ArgoCD via Kustomize - see https://stackoverflow.com/a/71692892/4964553"
kubectl apply -k argocd/installation
As you can see we must apply
the argocd
namespace here instead of create
ing it - otherwise the workflow will only run once.
We should also configure a GitHub Actions environment for our ArgoCD dashboard (as we already did with the Tekton dashboard):
- name: Expose ArgoCD Dashboard as GitHub environment
id: dashboard-expose
run: |
echo "--- Expose ArgoCD Dashboard via K8s Service"
kubectl apply -f traefik/argocd-dashboard.yml
echo "--- Create GitHub environment var"
DASHBOARD_HOST="https://argocd.$ROUTE53_DOMAIN_NAME"
echo "The ArgoCD dashboard is accessible at $DASHBOARD_HOST - creating GitHub Environment"
echo "::set-output name=dashboard_host::$DASHBOARD_HOST"
Even with ArgoCD there are two ways on how to deploy our application: push-based and pull-based.
We need a example project here - so what about https://github.com/jonashackt/restexamples and it's GitHub Container Registry image https://github.com/jonashackt/restexamples/pkgs/container/restexamples ?!
To access the GHCR we need to also create a secret inside our EKS cluster, so let's do that in our Actions workflow too:
- name: Create GitHub Container Registry Secret to be able to pull from ghcr.io
run: |
echo "--- Create Secret to access GitHub Container Registry"
kubectl create secret docker-registry github-container-registry \
--docker-server=ghcr.io \
--docker-username=${{ secrets.GHCR_USER }} \
--docker-password=${{ secrets.GHCR_PASSWORD }} \
--namespace default \
--save-config --dry-run=client -o yaml | kubectl apply -f -
Don't forget to create the needed repository secrets GHCR_USER
and GHCR_PASSWORD
containing a GitHub PAT (since we can't simply use the GITHUB_TOKEN
as we want to access another repositorie's images):
See https://argo-cd.readthedocs.io/en/stable/getting_started/#creating-apps-via-ui
In order to be able to deploy an application with ArgoCD UI, we need to fill in the source url and names etc. But we also need to provide a path
and this must contain some form of Kubernetes Deployment
and Service
configuration like this:
restexamples-deployment.yml
apiVersion: apps/v1
kind: Deployment
metadata:
name: restexamples
spec:
replicas: 1
revisionHistoryLimit: 3
selector:
matchLabels:
app: restexamples
template:
metadata:
labels:
app: restexamples
spec:
containers:
- image: ghcr.io/jonashackt/restexamples:latest
name: restexamples
ports:
- containerPort: 8090
imagePullSecrets:
- name: github-container-registry
restexamples-service.yml
apiVersion: v1
kind: Service
metadata:
name: restexamples
spec:
ports:
- port: 80
targetPort: 8090
selector:
app: restexamples
See both in the repo also https://github.com/jonashackt/restexamples/tree/argocd/deployment
If you filled out everything, click on create
and then manually on sync
. Now your app should be deployed to the EKS cluster:
Now have a look into k9s
- you should see your app beeing deployed as a Service
and as a Deployment
incl. Pods etc.
You can simply use a Port Forwarding with kubectl to access your service. For the example service this is:
kubectl port-forward svc/restexamples 8090:80
Now access the deployed example app at http://localhost:8090/restexamples/hello in your Browser.
If you want to play around a bit, you can edit the deploy
ment and set the replicas to 3 for example - now the ArgoUI shows the 3 pods now beeing spun up:
Alternatively we can create our Argo app via the CLI:
argocd app create restexamples-cli --repo https://github.com/jonashackt/restexamples.git --path deployment --dest-server https://kubernetes.default.svc --dest-namespace default --revision argocd
https://argo-cd.readthedocs.io/en/stable/user-guide/ci_automation/
Argo CD follows the GitOps model of deployment, where desired configuration changes are first pushed to Git, and the cluster state then syncs to the desired state in git.
It seems to be good practice to separate the Kubernetes manifests from your application (https://argo-cd.readthedocs.io/en/stable/user-guide/ci_automation/#update-the-local-manifests-using-your-preferred-templating-tool-and-push-the-changes-to-git)
The use of a different Git repository to hold your kubernetes manifests (separate from your application source code), is highly recommended.
Also from the best ArgoCD best practices section (https://argo-cd.readthedocs.io/en/stable/user-guide/best_practices/):
It provides a clean separation of application code vs. application config. There will be times when you wish to modify just the manifests without triggering an entire CI build. For example, you likely do not want to trigger a build if you simply wish to bump the number of replicas in a Deployment spec.
and
If you are automating your CI pipeline, pushing manifest changes to the same Git repository can trigger an infinite loop of build jobs and Git commit triggers.
Our example project https://github.com/jonashackt/restexamples right now holds both: application configuration AND source code.
Let's split that up into 2 repos. All Kubernetes configuration is now held in this new repo https://github.com/jonashackt/restexamples-k8s-config/tree/argocd
We need to tell Argo about this new configuration location by creating a new app based on this repo:
argocd app create restexamples-cli --repo https://github.com/jonashackt/restexamples-k8s-config.git --path deployment --dest-server https://kubernetes.default.svc --dest-namespace default --revision argocd --sync-policy auto
Our app should be created - the UI should show us the last commit Update restexamples to f01ffa895db8b7e25d5
as the latest sync status:
In CI/CD Pipelines you typically use the Git commit hash to tag your container image, which has just been build inside the pipeline.
So inside the restexamples-k8s-config root directory, run a kubectl patch
:
GIT_COMMIT_HASH=f01ffa895db8b7e25d5410ce4d33493fd8db9d8e8b089aaa265020be8099ff38
IMAGE_NAME=ghcr.io/jonashackt/restexamples@sha256:$GIT_COMMIT_HASH
kubectl patch --local -f deployment/restexamples-deployment.yml -p '{"spec":{"template":{"spec":{"containers":[{"name":"restexamples","image":"ghcr.io/jonashackt/restexamples:latest@sha256:"$GIT_COMMIT_HASH""}]}}}}' -o yaml > temp.yml && mv temp.yml deployment/restexamples-deployment.yml
Sadly this doesn't seem to work, since Kubernetes doesn't seem to provide the ability to substitute variables inside json with kubectl.
So what about using yq for that - see https://learnk8s.io/templating-yaml-with-code#templating-with-yq
Therefore install yq with:
brew install yq
Reading our image
tag from deployment/restexamples-deployment.yml
looks like this:
yq e '.spec.template.spec.containers[0].image' deployment/restexamples-deployment.yml
so: How to pass environment variable as value to yq?
GIT_COMMIT_HASH=f01ffa895db8b7e25d5410ce4d33493fd8db9d8e8b089aaa265020be8099ff38
IMAGE_NAME=ghcr.io/jonashackt/restexamples@sha256:$GIT_COMMIT_HASH
yq e ".spec.template.spec.containers[0].image = \"$IMAGE_NAME\"" -i deployment/restexamples-deployment.yml
Now finally add this to our configuration repository:
git add .
git commit -m "Update restexamples to $GIT_COMMIT_HASH"
git push
Now we should grab a coffee (if it's done in under 3 minutes, since Argo polles for changes every 3 minutes
) and have a look at the UI:
There we see the new version of our app beeing deployed, while the old pods are gradually beeing undeployed.
We could use the APP_IMAGE_DIGEST
result variable from our buildpacks Tekton Task https://hub.tekton.dev/tekton/task/buildpacks
But this value is different to the original Git Commit SHA in GitLab, it seems to be generated somehow.
So we should provide "$(params.IMAGE):$(params.SOURCE_REVISION)"
as the APP_IMAGE
parameter to our buildpacks
task:
params:
- name: APP_IMAGE
value: "$(params.IMAGE):$(params.SOURCE_REVISION)"
The resulting image will for example have the full name and tag like this
registry.gitlab.com/jonashackt/microservice-api-spring-boot:3c4131f8566ef157244881bacc474543ef96755d
As already used to clone the application repository, we simply use the git-clone Tekton task to fetch our application configuration repository:
- name: fetch-config-repository
taskRef:
name: git-clone
runAfter:
- buildpacks
workspaces:
- name: output
workspace: config-workspace
params:
- name: url
value: "$(params.CONFIG_URL)"
- name: revision
value: "$(params.CONFIG_REVISION)"
It uses CONFIG_URL
and CONFIG_REVISION
instead, which we both need to provide inside our pipeline-run.yml.
In order to have insights what files are fetched by the git-clone task, we can implement our own custom Task to show us these files. Let's imagine a dump-directory.yml:
apiVersion: tekton.dev/v1beta1
kind: Task
metadata:
name: dump-directory
spec:
workspaces:
- name: source
description: A workspace that contains the file which need to be dumped.
steps:
- name: dump-directory
image: alpine
workingDir: $(workspaces.source.path)
command:
- /bin/sh
args:
- '-c'
- |
set -ex
find /workspace
cd /workspace/source/deployment
ls -la
resources: {}
Use the custom task after you applied it with kubectl apply -f tekton/tasks/dump-directory.yml
inside the Tekton pipeline like this:
- name: dump-contents
taskRef:
name: dump-directory
runAfter:
- fetch-config-repository
workspaces:
- name: source
workspace: config-workspace
This will show us all the files in the workspace and also prints a detailled output of what is inside the deployment
directory:
We now need to somehow substitute the image
tag's name and tag to match our buildpack build application image. This was defined as "$(params.IMAGE):$(params.SOURCE_REVISION)"
.
There's a yq Tekton task we could use here https://hub.tekton.dev/tekton/task/yq
But this doesn't work currently: https://stackoverflow.com/questions/70944069/tekton-yq-task-gives-safelyrenamefile-erro-failed-copying-from-tmp-temp-e
It produces the following errors (without braking the pipeline, which is double sad):
16:50:43 safelyRenameFile [ERRO] Failed copying from /tmp/temp3555913516 to /workspace/source/deployment/deployment.yml
16:50:43 safelyRenameFile [ERRO] open /workspace/source/deployment/deployment.yml: permission denied
As https://stackoverflow.com/a/70944070/4964553 suggests we could use an older version of the Task - or write our own, which is preferred here - since also the older task wasn't able to evaluate the expression with 2 parameters like "$(params.IMAGE):$(params.SOURCE_REVISION)"
.
So we created our own custom replace-image-name-with-yq.yml:
apiVersion: tekton.dev/v1beta1
kind: Task
metadata:
name: replace-image-name-with-yq
spec:
workspaces:
- name: source
description: A workspace that contains the file which need to be dumped.
params:
- name: IMAGE_NAME
description: The image name to substitute
- name: FILE_PATH
description: The file path relative to the workspace dir.
- name: YQ_VERSION
description: Version of https://github.com/mikefarah/yq
default: v4.2.0
steps:
- name: substitute-with-yq
image: alpine
workingDir: $(workspaces.source.path)
command:
- /bin/sh
args:
- '-c'
- |
set -ex
echo "--- Download yq & add to path"
wget https://github.com/mikefarah/yq/releases/download/$(params.YQ_VERSION)/yq_linux_amd64 -O /usr/bin/yq &&\
chmod +x /usr/bin/yq
echo "--- Run yq expression"
yq e ".spec.template.spec.containers[0].image = \"$(params.IMAGE_NAME)\"" -i $(params.FILE_PATH)
echo "--- Show file with replacement"
cat $(params.FILE_PATH)
resources: {}
The cat $(params.FILE_PATH)
even shows the substitution in the Tekton output and/or Dashboard for convenience.
Now we need to commit and push the new image tag to our config repository. Therefore we can use the Tekton git-cli task https://hub.tekton.dev/tekton/task/git-cli
Maybe we can use the already existing GitHub PAT we created for the GitHub Container Registry access?!
But we need to create a new basic-auth
Secret in our GitHub Actions pipeline. Therefore we create a gitlab-push-secret.yml:
apiVersion: v1
kind: Secret
metadata:
name: gitlab-push-secret
annotations:
tekton.dev/git-0: https://gitlab.com
type: kubernetes.io/basic-auth
stringData:
username: gitlab-push-token
password: {{GITLAB_PUSH_TOKEN}}
and substitute the {{GITLAB_PUSH_TOKEN}}
using sed
in our GitHub Actions pipeline:
- name: Create Secret for GitHub based configuration repository
run: |
echo "--- Create Secret for GitHub based configuration repository"
sed "s#{{GITLAB_PUSH_TOKEN}}#${{ secrets.GITLAB_PUSH_TOKEN }}#g" tekton/misc/gitlab-push-secret.yml | kubectl apply -f -
Also we need to add the new Secret to our ServiceAccount
inside buildpacks-service-account-gitlab.yml:
apiVersion: v1
kind: ServiceAccount
metadata:
name: buildpacks-service-account-gitlab
secrets:
- name: gitlab-container-registry
- name: gitlab-push-secret
Finally we can use the git-cli
Task from the Tekton Hub https://hub.tekton.dev/tekton/task/git-cli to add, commit and push the deployment.yml
including it's replaced image
name containing the original GitLab commit Hash as image tag to the configuration repository.
- name: commit-and-push-to-config-repo
taskRef:
name: git-cli
runAfter:
- replace-config-image-name
workspaces:
- name: source
workspace: config-workspace
params:
- name: GIT_USER_NAME
value: "tekton"
- name: GIT_USER_EMAIL
value: "[email protected]"
- name: GIT_SCRIPT
value: |
git checkout -b "$(params.CONFIG_REVISION)"
git status
git add .
git commit -m "Update to $(params.IMAGE):$(params.SOURCE_REVISION)"
git push --set-upstream origin "$(params.CONFIG_REVISION)"
Mostly the GIT_SCRIPT
is important (the GIT_USER_NAME
and GIT_USER_EMAIL
are neat to define) - here we need to checkout the correct branch or revision, which is defined in $(params.CONFIG_REVISION)
.
Then we add a git status
to have some info printed into the Tekton logs. Also we add
and then commit
the deployment.yml
with a useful comment.
Finally we push the change to our configuration repository. To not run into errors we also need to set the upstream branch/revision via --set-upstream origin "$(params.CONFIG_REVISION)"
.
Now if we created our Argo application already with:
argocd app create microservice-api-spring-boot --repo https://gitlab.com/jonashackt/microservice-api-spring-boot-config.git --path deployment --dest-server https://kubernetes.default.svc --dest-namespace default --revision argocd --sync-policy auto
we should see the application beeing deployed through Argo after a maximum of 3 minutes:
The Hub task https://hub.tekton.dev/tekton/task/argocd-task-sync-and-wait HAZ NO APP CREATE!
So let's create our own simple Task https://github.com/codecentric/tekton-catalog/tree/main/task/argocd-task-create-sync-wait/0.3 derived from https://hub.tekton.dev/tekton/task/argocd-task-sync-and-wait and https://github.com/tektoncd/catalog/pull/903/files (since the v0.1
uses the old ArgoCD version 1.x
):
apiVersion: tekton.dev/v1beta1
kind: Task
metadata:
name: argocd-task-create-sync-and-wait
labels:
app.kubernetes.io/version: "0.2"
annotations:
tekton.dev/pipelines.minVersion: "0.12.1"
tekton.dev/categories: Deployment
tekton.dev/tags: deploy
tekton.dev/displayName: "argocd"
tekton.dev/platforms: "linux/amd64"
tekton.dev/deprecated: "true"
spec:
description: >-
This task creates and syncs (deploys) an Argo CD application and waits for it to be healthy.
(derived from https://hub.tekton.dev/tekton/task/argocd-task-sync-and-wait, which isn't able to create an app)
To do so, it requires the address of the Argo CD server and some form of
authentication either a username/password or an authentication token.
params:
- name: application-name
description: name of the application to sync
- name: config-repository
description: the applications config repository
- name: config-path
description: the path to the K8s deployment and service files in the applications config repository
- name: config-revision
description: the revision of the config repository to sync to
default: HEAD
- name: destination-namespace
description: the namespace to deploy the application to
- name: argo-appproject
description: the AppProject which contains the role with permissions to create and sync the application
- name: flags
description: Any flags to add to the command. Defaulting to --insecure here because of no proper certificate setup here
default: "--insecure"
- name: argocd-version
default: v2.2.2
steps:
- name: login-create-sync
image: quay.io/argoproj/argocd:$(params.argocd-version)
script: |
if [ -z "$ARGOCD_AUTH_TOKEN" ]; then
yes | argocd login "$ARGOCD_SERVER" --username="$ARGOCD_USERNAME" --password="$ARGOCD_PASSWORD";
fi
argocd app create "$(params.application-name)" --repo "$(params.config-repository)" --path "$(params.config-path)" --project "$(params.argo-appproject)" --dest-server https://kubernetes.default.svc --dest-namespace "$(params.destination-namespace)" --revision "$(params.config-revision)" --sync-policy auto "$(params.flags)"
argocd app sync "$(params.application-name)" --revision "$(params.config-revision)" "$(params.flags)"
argocd app wait "$(params.application-name)" --health "$(params.flags)"
envFrom:
- configMapRef:
name: argocd-env-configmap # used for server address
- secretRef:
name: argocd-env-secret # used for authentication (username/password or auth token)
And apply it with
kubectl apply -f https://raw.githubusercontent.com/codecentric/tekton-catalog/main/task/argocd-task-create-sync-wait/0.3/argocd-task-create-sync-wait.yml
https://hub.tekton.dev/tekton/task/argocd-task-sync-and-wait
kubectl create configmap argocd-env-configmap --from-literal="ARGOCD_SERVER=argocd.tekton-argocd.de"
Omit error Failed to establish connection to xyz.com:443: x509: certificate is valid for localhost, argocd-server, not xyz.com
time="2022-02-04T08:02:13Z" level=fatal msg="Failed to establish connection to a5f715808162c48c1af54069ba37db0e-1371850981.eu-central-1.elb.amazonaws.com:443: x509: certificate is valid for localhost, argocd-server, argocd-server.argocd, argocd-server.argocd.svc, argocd-server.argocd.svc.cluster.local, not a5f715808162c48c1af54069ba37db0e-1371850981.eu-central-1.elb.amazonaws.com"
Can't we simply access the ArgoCD server from within our cluster - since it's all deployed on the same K8s cluster?!
Since argocd-server
is not enough and produces a Failed to establish connection to argocd-server:443: dial tcp: lookup argocd-server on 10.100.0.10:53: no such host
, we should go with argocd-server.argocd.svc.cluster.local
(see https://stackoverflow.com/a/44329470/4964553):
kubectl create configmap argocd-env-configmap --from-literal="ARGOCD_SERVER=argocd-server.argocd.svc.cluster.local"
As this also gives us a Failed to establish connection to argocd-server.argocd.svc.cluster.local:443: x509: certificate signed by unknown authority
we should try the --insecure
flag, wwhich is described as:
--insecure Skip server certificate and domain verification
Now the argocd command should reach the ArgoCD server as expected.
Tackling the error:
error rpc error: code = PermissionDenied desc = permission denied: applications, create, default/jonashackt/microservice-api-spring-boot, sub: tekton, iat: 2022-02-03T16:36:48Z
So maybe we have the following issue described in https://argo-cd.readthedocs.io/en/stable/operator-manual/user-management/#local-usersaccounts-v15
When you create local users, each of those users will need additional RBAC rules set up, otherwise they will fall back to the default policy specified by policy.default field of the argocd-rbac-cm ConfigMap.
Here ArgoCD Projects come into play:
Projects provide a logical grouping of applications - [they] restrict what may be deployed (trusted Git source repositories)
ArgoCD projects have the ability to define Project roles:
Projects include a feature called roles that enable automated access to a project's applications. These can be used to give a CI pipeline a restricted set of permissions. For example, a CI system may only be able to sync a single app (but not change its source or destination).
So let's get our hands dirty and create a ArgoCD project:
argocd proj create apps2deploy -d https://kubernetes.default.svc,default --src "*"
We create it with the --src "*"
as a wildcard for any git repository (as described here).
Now we create a Project role called create-sync
via:
argocd proj role create apps2deploy create-sync --description "project role to create and sync apps from a CI/CD pipeline"
You can check the new role has been created with argocd proj role list apps2deploy
.
Now we need to create a token for the new Project role create-sync
, which can be created via:
argocd proj role create-token apps2deploy create-sync
Directly update the ARGOCD_AUTH_TOKEN
in the argocd-env-secret
secret:
kubectl create secret generic argocd-env-secret \
--from-literal=ARGOCD_AUTH_TOKEN=INSERT_TOKEN_HERE \
--namespace default \
--save-config --dry-run=client -o yaml | kubectl apply -f -
Now we need to give permissions for Tekton to be able to create and sync our application in ArgoCD. Therefore use (for more details see):
argocd proj role add-policy apps2deploy create-sync --action get --permission allow --object "*"
argocd proj role add-policy apps2deploy create-sync --action create --permission allow --object "*"
argocd proj role add-policy apps2deploy create-sync --action sync --permission allow --object "*"
argocd proj role add-policy apps2deploy create-sync --action update --permission allow --object "*"
argocd proj role add-policy apps2deploy create-sync --action delete --permission allow --object "*"
Have a look on the role policies with argocd proj role get apps2deploy create-sync
, which should look somehow like this:
$ argocd proj role get apps2deploy create-sync
Role Name: create-sync
Description: project role to create and sync apps from a CI/CD pipeline
Policies:
p, proj:apps2deploy:create-sync, projects, get, apps2deploy, allow
p, proj:apps2deploy:create-sync, applications, get, apps2deploy/*, allow
p, proj:apps2deploy:create-sync, applications, create, apps2deploy/*, allow
p, proj:apps2deploy:create-sync, applications, update, apps2deploy/*, allow
p, proj:apps2deploy:create-sync, applications, delete, apps2deploy/*, allow
p, proj:apps2deploy:create-sync, applications, sync, apps2deploy/*, allow
JWT Tokens:
ID ISSUED-AT EXPIRES-AT
1644166189 2022-02-06T17:49:49+01:00 (2 hours ago) <none>
Now we finally need to add our application to the AppProject
we created.
We add it to our argocd-task-app-create.yml argocd app create
command as --project "$(params.argo-appproject)"
with a new parameter argo-appproject
.
Finally we need to introduce a new parameter containing only the project name, since the REPO_PATH_ONLY
parameter containing e.g. jonashackt/microservice-api-spring-boot
produces an error like rpc error: code = Unknown desc = invalid resource name \"jonashackt/microservice-api-spring-boot\": [may not contain '/']
.
So let's introduce PROJECT_NAME
to our pipeline.yml, which we can also retrieve easily in our EventListener / Tekton Trigger solution implemented in gitlab-push-listener.yml.
There we can use $(body.project.name)
inside the TriggerBinding to retrieve the project name from the payload (see gitlab-push-test-event.json) and use it later in the parameter definition.
Mind the spec params definition of project_name
also to not run into '$(tt.params.project_name)' must be declared in spec.params
errors. Now the parameter can finally be used as:
- name: PROJECT_NAME
value: $(tt.params.project_name)
In the end our pipeline should be able to create our app and sync/wait for it to be deployed:
So we're doing CI/CD for our CI/CD process here, right?! So let's also automate all the steps inside our provision.yml.
- name: Install ArgoCD CLI
run: brew install argocd
Then we need to install our custom task:
echo "--- Install the argocd-task-create-sync-and-wait task"
kubectl apply -f tasks/argocd-task-app-create.yml
Let's do the argocd login
command without human interaction (see https://stackoverflow.com/a/71030112/4964553):
We already know how to extract the password for argo with kubectl -n argocd get secret argocd-initial-admin-secret -o jsonpath="{.data.password}" | base64 -d; echo
(we should not change it here inside our CI/CD process in order to be able to use it for ArgoCD configuration).
We also know how to obtain the ArgoCD server's hostname with kubectl get service argocd-server -n argocd --output=jsonpath='{.status.loadBalancer.ingress[0].hostname}'
.
Now as the argocd login
command has the parameters --username
and --password
, we can craft our login command like this:
argocd login $(kubectl get service argocd-server -n argocd --output=jsonpath='{.status.loadBalancer.ingress[0].hostname}') --username admin --password $(kubectl -n argocd get secret argocd-initial-admin-secret -o jsonpath="{.data.password}" | base64 -d; echo) --insecure
Mind the --insecure
to prevent the argo CLI from asking things like WARNING: server certificate had error: x509: certificate is valid for localhost, argocd-server, argocd-server.argocd, argocd-server.argocd.svc, argocd-server.argocd.svc.cluster.local, not a5f715808162c48c1af54069ba37db0e-1371850981.eu-central-1.elb.amazonaws.com. Proceed insecurely (y/n)?
.
Prevent error FATA[0000] dial tcp: lookup a965bfb530e8449f5a355f221b2fd107-598531793.eu-central-1.elb.amazonaws.com on 8.8.8.8:53: no such host
see https://stackoverflow.com/a/71030112/4964553
The problem arises if the argocd-server
Kubernetes service is freshly installed right before the argocd login
command is run.
Then the argocd login
command failes for some time until it finally will work correctly. Assuming some DNS propagation issues we can prevent this error from breaking our CI pipeline by wrapping our argocd login
command into an until
like already done in this answer https://stackoverflow.com/a/70108997/4964553
The full command will then look like this:
until argocd login $(kubectl get service argocd-server -n argocd --output=jsonpath='{.status.loadBalancer.ingress[0].hostname}') --username admin --password $(kubectl -n argocd get secret argocd-initial-admin-secret -o jsonpath="{.data.password}" | base64 -d; echo) --insecure; do : ; done
In GitHub Actions this will then look somehow like this:
--- Login argocd CLI - now wrapped in until to prevent FATA[0000] dial tcp: lookup 12345.eu-central-1.elb.amazonaws.com on 8.8.8.8:53: no such host
time="2022-02-21T12:57:32Z" level=fatal msg="dial tcp: lookup a071bed7e9ea14747951b04360133141-459093397.eu-central-1.elb.amazonaws.com on 127.0.0.53:53: no such host"
time="2022-02-21T12:57:35Z" level=fatal msg="dial tcp: lookup a071bed7e9ea14747951b04360133141-459093397.eu-central-1.elb.amazonaws.com on 127.0.0.53:53: no such host"
time="2022-02-21T12:57:37Z" level=fatal msg="dial tcp: lookup a071bed7e9ea14747951b04360133141-459093397.eu-central-1.elb.amazonaws.com on 127.0.0.53:53: no such host"
[...]
time="2022-02-21T12:58:27Z" level=fatal msg="dial tcp: lookup a071bed7e9ea14747951b04360133141-459093397.eu-central-1.elb.amazonaws.com on 127.0.0.53:53: no such host"
time="2022-02-21T12:58:30Z" level=fatal msg="dial tcp: lookup a071bed7e9ea14747951b04360133141-459093397.eu-central-1.elb.amazonaws.com on 127.0.0.53:53: no such host"
time="2022-02-21T12:58:32Z" level=fatal msg="dial tcp: lookup a071bed7e9ea14747951b04360133141-459093397.eu-central-1.elb.amazonaws.com on 127.0.0.53:53: no such host"
'admin:login' logged in successfully
Context 'a071bed7e9ea14747951b04360133141-459093397.eu-central-1.elb.amazonaws.com' updated
We need to create the ConfigMap
idempotently - so a simple kubectl create configmap
would crash in GitHub Actions the second time it runs. So let's redesign it to use kubectl apply -f -
style like that:
echo "--- Create ConfigMap to point argocd CLI to our argocd-server"
kubectl create configmap argocd-env-configmap \
--from-literal="ARGOCD_SERVER=$(kubectl get service argocd-server -n argocd --output=jsonpath='{.status.loadBalancer.ingress[0].hostname}')" \
--namespace default \
--save-config --dry-run=client -o yaml | kubectl apply -f -
Since all those commands are quite bloated, we should better go with AppProject
yaml manifest like https://argo-cd.readthedocs.io/en/stable/user-guide/projects/#configuring-rbac-with-projects and argoproj/argo-cd#5382
So let's create a manifest file like argocd-appproject-apps2deploy.yml:
apiVersion: argoproj.io/v1alpha1
kind: AppProject
metadata:
name: apps2deploy
namespace: argocd
spec:
destinations:
- namespace: default
server: https://kubernetes.default.svc
sourceRepos:
- '*'
roles:
- description: project role to create and sync apps from a CI/CD pipeline
name: create-sync
policies:
- p, proj:apps2deploy:create-sync, applications, get, apps2deploy/*, allow
- p, proj:apps2deploy:create-sync, applications, create, apps2deploy/*, allow
- p, proj:apps2deploy:create-sync, applications, update, apps2deploy/*, allow
- p, proj:apps2deploy:create-sync, applications, delete, apps2deploy/*, allow
- p, proj:apps2deploy:create-sync, applications, sync, apps2deploy/*, allow
Let's apply it with
kubectl apply -f argocd/argocd-appproject-apps2deploy.yml
We also check the new role has been created with argocd proj role list apps2deploy
.
Now we need to create a token for the AppProject role create-sync
- but not with the bloated output! We only need the token. Luckily there's a parameter -t, --token-only Output token only - for use in scripts.
. So argocd proj role create-token apps2deploy create-sync --token-only
creates only the token.
We can directly combine the token generation with the Secret creation like this:
kubectl create secret generic argocd-env-secret \
--from-literal=ARGOCD_AUTH_TOKEN=$(argocd proj role create-token apps2deploy create-sync --token-only) \
--namespace default \
--save-config --dry-run=client -o yaml | kubectl apply -f -
Let's add a feature branch name based deployment in ArgoCD - ideally through namespaces. If we have a look at our app https://gitlab.com/jonashackt/microservice-api-spring-boot/ and assume one creates a new feature branch called greatfeature
, we want our app to be deployed in a separate Kubernetes Deployment & Service named greatfeature
.
So how do we accomplish this?
In our full Tekton / Argo architecture we have the Tekton Triggers EventListener first. Here we need to extract the branch name
from the GitLab WebHook event json (we have an example in gitlab-push-test-event.json). The branch name
could be found in ref
field:
{
"object_kind": "push",
"event_name": "push",
"before": "5bbc8580432fc7a16f50be27eb513db42aad0860",
"after": "c25a74c8f919a72e3f00928917dc4ab2944ab061",
"ref": "refs/heads/greatfeature",
"checkout_sha": "c25a74c8f919a72e3f00928917dc4ab2944ab061",
...
"project": {
"id": 30444286,
"name": "microservice-api-spring-boot",
"description": "Forked from https://github.com/jonashackt/microservice-api-spring-boot",
"web_url": "https://gitlab.com/jonashackt/microservice-api-spring-boot",
...
Since the ref
field includes refs/heads/
inside the string we need to somehow strip these out. Therefore we can use the CEL
interceptor https://tekton.dev/vault/triggers-main/interceptors/#cel-interceptors in our Tekton Triggers configuration:
CEL Interceptors support overlays, which are CEL expressions that Tekton Triggers adds to the event payload in the top-level extensions field. overlays are accessible from TriggerBindings.
Let's add the interceptor in our gitlab-push-listener.yml:
...
triggers:
- name: gitlab-push-events-trigger
interceptors:
- name: "verify-gitlab-payload"
ref:
name: "gitlab"
...
- name: "split-ref-heads-from-branch-name"
ref:
name: cel
params:
- name: "overlays"
value:
- key: branch_name
expression: "body.ref.split('/')[2]"
...
Now inside the bindings
configuration we can access the CEL splitted branch_name
via $(extensions.branch_name)
instead of using the already known $(body.ref)
notation:
...
bindings:
- name: gitrevision
value: $(body.checkout_sha)
- name: gitbranch
value: $(extensions.branch_name)
...
Also don't forget to add a new parameter gitbranch
to the Template and use it as a parameter for the PipelineRun
(like SOURCE_BRANCH
).
We need to make sure the feature-branch is present or created inside our application configuration repository, which should contain the freshly replaced Deployment
and Service
manifests.
Because our ArgoCD deployment environments are based on these branches. Therefore create a new git-cli
based Task which creates a new branch - or uses the already existing one - prior to our manifests changes.
If the branch already exists, we need to first do a git fetch
from origin - otherwise we get the following error:
error: pathspec 'your-branch-name-here' did not match any file(s) known to git
If the branch DOES NOT exits, we need to create a new branch using git checkout
. Both can be accomplished using the ||
operator:
git fetch origin "$(params.SOURCE_BRANCH)" && git checkout "$(params.SOURCE_BRANCH)" || git checkout -b "$(params.SOURCE_BRANCH)"
The full Task switch-config-repository-branch
using git-cli
in the pipeline.yml looks like this:
- name: switch-config-repository-branch
taskRef:
name: git-cli
runAfter:
- fetch-config-repository
workspaces:
- name: source
workspace: config-workspace
params:
- name: GIT_USER_NAME
value: "tekton"
- name: GIT_USER_EMAIL
value: "[email protected]"
- name: GIT_SCRIPT
value: |
git fetch origin "$(params.SOURCE_BRANCH)" && git checkout "$(params.SOURCE_BRANCH)" || git checkout -b "$(params.SOURCE_BRANCH)"
Add branch name in Deployment & Service (metadata.name, spec.selector.matchLabels.branch, spec.template.metadata.labels.branch, spec.selector.branch)
Now that we have our branch name
extracted via Tekton Triggers CEL interceptor and available for the Pipeline as a parameter, we should add it to our application's Deployment
and Service
manifests.
Therefore our replace-image-name-with-yq.yml needs to be redesigned, since right now it only replaces the image
tag. So first rename it to replace-yaml-value-with-yq.yml
and then we may start in our pipeline.yml to see what interface our task should have:
- name: replace-deployment-name-branch-image
taskRef:
name: replace-yaml-value-with-yq
runAfter:
- fetch-config-repository
workspaces:
- name: source
workspace: config-workspace
params:
- name: YQ_EXPRESSIONS
value:
- ".metadata.name = \"$(params.PROJECT_NAME)-$(params.SOURCE_BRANCH)\""
- ".spec.template.spec.containers[0].image = \"$(params.IMAGE):$(params.SOURCE_REVISION)\""
- ".spec.selector.matchLabels.branch = \"$(params.SOURCE_BRANCH)\""
- ".spec.template.metadata.labels.branch = \"$(params.SOURCE_BRANCH)\""
- name: FILE_PATH
value: "./deployment/deployment.yml"
In our application configuration repository https://gitlab.com/jonashackt/microservice-api-spring-boot-config inside the deployment/deployment.yml
we need to replace:
metadata.name
to contain ourPROJECT_NAME-SOURCE_BRANCH
- for examplemicroservice-api-spring-boot-trigger-tekton-via-webhook
spec.template.spec.containers[0].image
must contain the correct image name as already implementedspec.selector.matchLabels.branch
should contain thebranch name
spec.template.metadata.labels.branch
should also contain thebranch name
And in the application configuration repository's deployment/service.yml
we need to replace:
metadata.name
to containPROJECT_NAME-SOURCE_BRANCH
- just like in our Deploymentspec.selector.branch
should contain thebranch name
- also very similar to our Deployment
which inside our pipeline.yml looks like:
- name: replace-service-name-branch
taskRef:
name: replace-yaml-value-with-yq
runAfter:
- replace-deployment-name-branch-image
workspaces:
- name: source
workspace: config-workspace
params:
- name: YQ_EXPRESSIONS
value:
- ".metadata.name = \"$(params.PROJECT_NAME)-$(params.SOURCE_BRANCH)\""
- ".spec.selector.branch = \"$(params.SOURCE_BRANCH)\""
- name: FILE_PATH
value: "./deployment/service.yml"
Therefore the task replace-image-name-with-yq.yml was redesigned to support multiple yq expressions as array list:
apiVersion: tekton.dev/v1beta1
kind: Task
metadata:
name: replace-yaml-value-with-yq
spec:
workspaces:
- name: source
description: A workspace that contains the file which need to be dumped.
params:
- name: YQ_EXPRESSIONS
type: array
description: "The yq expression yaml selector to choose a key for replacement like .spec.template.spec.containers[0].image = \"$(params.IMAGE):$(params.SOURCE_REVISION)\""
- name: FILE_PATH
description: The file path relative to the workspace dir.
- name: YQ_VERSION
description: Version of https://github.com/mikefarah/yq
default: v4.2.0
steps:
- name: substitute-with-yq
image: alpine
workingDir: $(workspaces.source.path)
args: ["$(params.YQ_EXPRESSIONS[*])"]
script: |
echo "--- Download yq & add to path"
wget https://github.com/mikefarah/yq/releases/download/$(params.YQ_VERSION)/yq_linux_amd64 -O /usr/bin/yq
chmod +x /usr/bin/yq
echo "--- Run yq expressions"
for expression in "$@"
do
yq e "$expression" -i $(params.FILE_PATH)
done
echo "--- Show file with replacement"
cat $(params.FILE_PATH)
resources: {}
The traefik-ingress-route.yml
will also be added to our application configuration repository https://gitlab.com/jonashackt/microservice-api-spring-boot-config in the deployment
directory. So now it can be also deployed using Argo.
Now we simply use our replace-yaml-value-with-yq.yml a 3rd time in our pipeline:
- name: replace-ingress-name-route
taskRef:
name: replace-yaml-value-with-yq
runAfter:
- replace-service-name-branch
workspaces:
- name: source
workspace: config-workspace
params:
- name: YQ_EXPRESSIONS
value:
- ".metadata.name = \"$(params.PROJECT_NAME)-$(params.SOURCE_BRANCH)-ingressroute\""
- ".spec.routes[0].match = \"Host(`$(params.PROJECT_NAME)-$(params.SOURCE_BRANCH).$(params.TRAEFIK_DOMAIN)`)\""
- ".spec.routes[0].services[0].name = \"$(params.PROJECT_NAME)-$(params.SOURCE_BRANCH)\""
- name: FILE_PATH
value: "./deployment/traefik-ingress-route.yml"
Now ArgoCD should deploy the Traefik IngressRoute
which matches the Service and Branch name exactly to our cluster:
Try to access the app after a successful pipeline run using your Browser:
In order to replace all needed fields in Deployment, Service and IngressRoute using the refactored replace task we still have 3 big tasks with lot's of yq expressions, we need to maintain in the future.
But we can switch over to Kustomize - see https://stackoverflow.com/questions/71704023/how-to-use-kustomize-to-configure-traefik-2-x-ingressroute-metadata-name-spec
If you're interested how this works have a look into the application configuration repository: https://gitlab.com/jonashackt/microservice-api-spring-boot-config/-/blob/main/README.md#configuration-with-kustomize
When our application configuration repository is Kustomize-ready (it at least needs a kustomization.yaml
), we can refactor our tekton/pipelines/pipeline.yml to use a custom task, which replaces the 3 replace-with-yq tasks:
- name: kustomize-manifests
taskRef:
name: kustomize-manifests
runAfter:
- switch-config-repository-branch
workspaces:
- name: source
workspace: config-workspace
params:
- name: KUSTOMIZATION_PATH
value: "deployment"
- name: APPLICATION_NAME
value: "$(params.PROJECT_NAME)"
- name: TRAEFIK_DOMAIN
value: "$(params.TRAEFIK_DOMAIN)"
- name: BRANCH_NAME
value: "$(params.SOURCE_BRANCH)"
- name: IMAGE_NAME
value: "$(params.IMAGE):$(params.SOURCE_REVISION)"
Inside the new custom task kustomize-manifests.yml, we simply use the official Kustomize container image https://kubectl.docs.kubernetes.io/installation/kustomize/docker/ to issue our kustomize edit set
commands:
apiVersion: tekton.dev/v1beta1
kind: Task
metadata:
name: kustomize-manifests
spec:
workspaces:
- name: source
description: The workspace containing the manifests and kustomization.yaml
params:
- name: KUSTOMIZATION_PATH
description: The path where the root kustomization.yaml can be found
- name: APPLICATION_NAME
description: The application or project name - e.g. microservice-api-spring-boot
- name: TRAEFIK_DOMAIN
description: The domain part of the Traefik IngressRoutes .spec.routes.match Host - e.g. tekton-argocd.de
- name: BRANCH_NAME
description: The branch name to configure to the manifests with Kustomize
default: main
- name: IMAGE_NAME
description: The image name used in the deployments .spec.template.spec.containers[0].image
- name: KUSTOMIZE_VERSION
description: Version of https://kubectl.docs.kubernetes.io/installation/kustomize/docker/
default: v4.5.4
steps:
- name: kustomize-them-all
image: k8s.gcr.io/kustomize/kustomize:$(params.KUSTOMIZE_VERSION)
workingDir: $(workspaces.source.path)
script: |
echo "--- cd into the kustomization root folder"
cd $(params.KUSTOMIZATION_PATH)
echo "--- Create ingressroute-patch.yml with correct spec.routes.match: Host() name for Traefik IngressRoute - see https://stackoverflow.com/a/71704024/4964553"
cat > ./ingressroute-patch.yml <<EOF
apiVersion: traefik.containo.us/v1alpha1
kind: IngressRoute
metadata:
name: $(params.APPLICATION_NAME)-ingressroute
namespace: default
spec:
entryPoints:
- web
routes:
- match: Host(\`$(params.APPLICATION_NAME)-$(params.BRANCH_NAME).$(params.TRAEFIK_DOMAIN)\`)
kind: Rule
services:
- name: $(params.APPLICATION_NAME)
port: 80
EOF
echo "--- Run kustomize edits"
kustomize edit set namesuffix -- -$(params.BRANCH_NAME)
kustomize edit set label branch:$(params.BRANCH_NAME)
kustomize edit set image $(params.IMAGE_NAME)
echo "--- Show output of Kustomization for better insights"
kustomize build .
resources: {}
Also managed by Kustomize the task get's installed right in our GitHub Actions pipeline by
kubectl apply -k tekton/tasks
We already made sure the feature-branch is present or created inside our application configuration repository. Now we need to push the freshly replaced Deployment
and Service
manifests to the branch also.
Because our ArgoCD deployment environments are based on these branches. Therefore we enhance our git-cli
task to push to the branch via git push --set-upstream origin "$(params.SOURCE_BRANCH)"
:
- name: commit-and-push-to-config-repo
taskRef:
name: git-cli
runAfter:
- replace-service-name-branch
workspaces:
- name: source
workspace: config-workspace
params:
- name: GIT_USER_NAME
value: "tekton"
- name: GIT_USER_EMAIL
value: "[email protected]"
- name: GIT_SCRIPT
value: |
git status
git add .
git commit -m "Update to $(params.IMAGE):$(params.SOURCE_REVISION) on branch $(params.SOURCE_BRANCH)" && git push --set-upstream origin "$(params.SOURCE_BRANCH)" --force
The ArgoCD application-name should contain the feature-branch with "$(params.PROJECT_NAME)-$(params.SOURCE_BRANCH)"
. Also the config-revision
should use the$(params.SOURCE_BRANCH)
parameter.
- name: argo-create-app-sync-wait
taskRef:
name: argocd-task-create-sync-and-wait
runAfter:
- commit-and-push-to-config-repo
params:
- name: application-name
value: "$(params.PROJECT_NAME)-$(params.SOURCE_BRANCH)"
- name: config-repository
value: "$(params.CONFIG_URL)"
- name: config-path
value: deployment
- name: config-revision
value: "$(params.SOURCE_BRANCH)"
- name: destination-namespace
value: default
- name: argo-appproject
value: apps2deploy
This will lead to an argocd app create
inside our argocd-task-app-create.yml that uses the correct name
and --revision
.
Also we need to add 2 configuration flags:
--upsert
: To be able to also upgrade already existing applications (Allows to override application with the same name even if supplied application spec is different from existing spec
)--auto-prune
: To automatically delete apps whose feature-branch doesn't exist in the application configuration repository anymore (Set automatic pruning when sync is automated
)
Now if we run our Pipeline using different branches in GitLab, we should see our application getting deployed multiple times:
In order to enable Renovate to keep all our manifests up-to-date, we need a mechanism and a format renovate can read.
As we already use Kustomize to install and configure ArgoCD, we could use it to install all needed remote manifests for us. Kustomize is also supported by Renovate: https://docs.renovatebot.com/modules/manager/kustomize/
Inside our GitHub Actions workflow provision.yml Kustomize is used through kubectl apply -k
:
- name: Install ArgoCD
run: |
echo "--- Create argo namespace and install it"
kubectl create namespace argocd --dry-run=client -o yaml | kubectl apply -f -
echo "--- Install & configure ArgoCD via Kustomize - see https://stackoverflow.com/a/71692892/4964553"
kubectl apply -k argocd/install
...
- name: Install Tekton Pipelines, Dashboard, Triggers
run: |
echo "--- Install Tekton Pipelines, Dashboard, Triggers via Kustomize"
kubectl apply -k tekton/install
...
- name: Install Tekton Hub & local Tasks via Kustomize
run: |
kubectl apply -k installation/tekton-tasks
Sadly Renovate doesn't seem to work out of the box with our kustomization.yaml
s - right now it simply does nothing to update Tekton, ArgoCD etc.
But Renovate should somehow support Kustomize: https://docs.renovatebot.com/modules/manager/kustomize/ (they link to Kustomize docs https://github.com/kubernetes-sigs/kustomize/blob/master/examples/remoteBuild.md) - it seems that git ref references are supported.
So how does this work? For example,
This url inside a kustomization.yaml
:
https://github.com/kubernetes-sigs/kustomize/tree/v1.0.6/examples/multibases/dev
has to be rebuild to this
https://github.com/kubernetes-sigs/kustomize//examples/multibases/dev/?ref=v1.0.6
which then only works inside a kustomization.yaml
:
cat > ./kustomization.yaml <<EOF
apiVersion: kustomize.config.k8s.io/v1beta1
kind: Kustomization
resources:
- https://github.com/kubernetes-sigs/kustomize//examples/multibases/dev/?ref=v1.0.6
EOF
And then kustomize build .
works:
kustomize build .
apiVersion: v1
kind: Pod
metadata:
labels:
app: myapp
name: dev-myapp-pod
spec:
containers:
- image: nginx:1.7.9
name: nginx
So how can we adapt this to the ArgoCD installation for example?
https://raw.githubusercontent.com/argoproj/argo-cd/v2.3.3/manifests/install.yaml
which references:
https://github.com/argoproj/argo-cd/blob/master/manifests/install.yaml
So does our kustomization.yaml
look like this now?
cat > ./kustomization.yaml <<EOF
apiVersion: kustomize.config.k8s.io/v1beta1
kind: Kustomization
resources:
- https://github.com/argoproj/argo-cd//manifests/install.yaml?ref=v2.3.3
EOF
But this gives us a:
$ kustomize build .
Error: accumulating resources: accumulation err='accumulating resources from 'https://github.com/argoproj/argo-cd//manifests/install.yaml?ref=v2.3.3':
URL is a git repository': '/private/var/folders/5p/l1cc1kqd69n_qxrftgln7xdm0000gn/T/kustomize-1732093150/manifests/install.yaml' refers to file 'install.yaml'; expecting directory
So the problem is, when the Kustomization remote has no kustomization.yaml
, but instead a different file like install.yaml
.
Ok, so we need kustomization.yamls
in the remote source!
https://gitlab.com/MShekow/gitops-with-monitoring/-/blob/main/argocd-kustomize/kustomization.yaml
With this I realized, that there are multiple directories in https://github.com/argoproj/argo-cd/tree/master/manifests featuring a kustomization.yaml
! Which isn't really documented.
But it seems that the https://github.com/argoproj/argo-cd/tree/master/manifests/cluster-install is the same package as the install.yaml
- which is referred to as the "Standard Argo CD installation with cluster-admin access."
The url that should work uses the
apiVersion: kustomize.config.k8s.io/v1beta1
kind: Kustomization
resources:
- https://github.com/argoproj/argo-cd.git/manifests/cluster-install?ref=v2.3.2
## changes to config maps
patchesStrategicMerge:
- argocd-cmd-params-cm-patch.yml
namespace: argocd
The Tekton pipeline failed and I had to dig into the Pod logs to find the error (see this log):
As described in https://stackoverflow.com/a/55514852/4964553 and the section Statefull applications
in https://vorozhko.net/120-days-of-aws-eks-kubernetes-in-staging two nodes are provisioned on other AWS availability zones as the persistent volume (PV), which is created by applying our PersistendVolumeClaim in buildpacks-source-pvc.yml.
To double check that, you need to look into/describe your nodes:
k get nodes
NAME STATUS ROLES AGE VERSION
ip-172-31-10-186.eu-central-1.compute.internal Ready <none> 2d16h v1.21.5-eks-bc4871b
ip-172-31-20-83.eu-central-1.compute.internal Ready <none> 2d16h v1.21.5-eks-bc4871b
and have a look at the Label
section:
$ k describe node ip-172-77-88-99.eu-central-1.compute.internal
Name: ip-172-77-88-99.eu-central-1.compute.internal
Roles: <none>
Labels: beta.kubernetes.io/arch=amd64
beta.kubernetes.io/instance-type=t2.medium
beta.kubernetes.io/os=linux
failure-domain.beta.kubernetes.io/region=eu-central-1
failure-domain.beta.kubernetes.io/zone=eu-central-1b
kubernetes.io/arch=amd64
kubernetes.io/hostname=ip-172-77-88-99.eu-central-1.compute.internal
kubernetes.io/os=linux
node.kubernetes.io/instance-type=t2.medium
topology.kubernetes.io/region=eu-central-1
topology.kubernetes.io/zone=eu-central-1b
Annotations: node.alpha.kubernetes.io/ttl: 0
...
In my case the node ip-172-77-88-99.eu-central-1.compute.internal
has failure-domain.beta.kubernetes.io/region
defined as eu-central-1
and the az with failure-domain.beta.kubernetes.io/zone
to `eu-central-1b``
And the other node defines az eu-central-1a
:
$ k describe nodes ip-172-31-10-186.eu-central-1.compute.internal
Name: ip-172-31-10-186.eu-central-1.compute.internal
Roles: <none>
Labels: beta.kubernetes.io/arch=amd64
beta.kubernetes.io/instance-type=t2.medium
beta.kubernetes.io/os=linux
failure-domain.beta.kubernetes.io/region=eu-central-1
failure-domain.beta.kubernetes.io/zone=eu-central-1a
kubernetes.io/arch=amd64
kubernetes.io/hostname=ip-172-31-10-186.eu-central-1.compute.internal
kubernetes.io/os=linux
node.kubernetes.io/instance-type=t2.medium
topology.kubernetes.io/region=eu-central-1
topology.kubernetes.io/zone=eu-central-1a
Annotations: node.alpha.kubernetes.io/ttl: 0
...
Now looking into our PersistentVolume
automatically provisioned after applying our PersistentVolumeClaim
with buildpacks-source-pvc.yml, we see the problem already:
$ k get pv
NAME CAPACITY ACCESS MODES RECLAIM POLICY STATUS CLAIM STORAGECLASS REASON AGE
pvc-93650993-6154-4bd0-bd1c-6260e7df49d3 1Gi RWO Delete Bound default/buildpacks-source-pvc gp2 21d
$ k describe pv pvc-93650993-6154-4bd0-bd1c-6260e7df49d3
Name: pvc-93650993-6154-4bd0-bd1c-6260e7df49d3
Labels: topology.kubernetes.io/region=eu-central-1
topology.kubernetes.io/zone=eu-central-1c
Annotations: kubernetes.io/createdby: aws-ebs-dynamic-provisioner
...
The PersistentVolume
was provisioned to topology.kubernetes.io/zone
in az eu-central-1c
, which makes our Pods complain about not finding their volume - since they are in a completely different az.
As stated in the Kubernetes docs one solution to the problem is to add a allowedTopologies
configuration to the StorageClass
like this:
apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
name: gp2
parameters:
fsType: ext4
type: gp2
provisioner: kubernetes.io/aws-ebs
reclaimPolicy: Delete
volumeBindingMode: WaitForFirstConsumer
allowedTopologies:
- matchLabelExpressions:
- key: failure-domain.beta.kubernetes.io/zone
values:
- eu-central-1a
- eu-central-1b
If you already provisioned a EKS cluster like me, you need to show your already defined StorageClass
with
k get storageclasses gp2 -o yaml
and add the allowedTopologies
configuration with:
k apply -f tekton/misc/storage-class.yml
As you see the allowedTopologies
configuration defines that the failure-domain.beta.kubernetes.io/zone
of the PersistentVolume
must be either in eu-central-1a
or eu-central-1b
- not eu-central-1c
!
Next apply this StorageClass
and delete the PersistentVolumeClaim
. Now add storageClassName: gp2
to the PersistendVolumeClaim definition in buildpacks-source-pvc.yml:
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
name: buildpacks-source-pvc
spec:
accessModes:
- ReadWriteOnce
resources:
requests:
storage: 500Mi
storageClassName: gp2
and then re-applying it will resolve the problem.
The command k describe pvc buildpacks-source-pvc
gives the following event messages:
Name: buildpacks-source-pvc
Namespace: default
StorageClass: gp2
Status: Pending
Volume:
Labels: <none>
Annotations: volume.beta.kubernetes.io/storage-provisioner: ebs.csi.aws.com
volume.kubernetes.io/selected-node: ip-172-31-18-82.eu-central-1.compute.internal
volume.kubernetes.io/storage-provisioner: ebs.csi.aws.com
Finalizers: [kubernetes.io/pvc-protection]
Capacity:
Access Modes:
VolumeMode: Filesystem
Used By: affinity-assistant-0b3d266b91-0
affinity-assistant-53a7c08baf-0
affinity-assistant-a375f28de3-0
affinity-assistant-e8cb1a6e15-0
buildpacks-test-pipeline-run-9rz4l-fetch-repository-pod
Events:
Type Reason Age From Message
---- ------ ---- ---- -------
Normal ExternalProvisioning 3m43s (x561 over 143m) persistentvolume-controller waiting for a volume to be created, either by external provisioner "ebs.csi.aws.com" or manually created by system administrator
The thing from Kubernetes 1.23+ on is that we need to additionally install the Amazon EBS CSI driver in order to get PersistentVolumes working again.
The aws docs tell us what to do and the aws-ebs-csi-driver GitHub readme:
See https://docs.aws.amazon.com/eks/latest/userguide/csi-iam-role.html
The easiest way to enable the IAM OIDC provider and create the IAM role for the EBS CSI driver is to use eksctl. To use eksctl
we need to install it first. On a Mac use brew like:
brew tap weaveworks/tap
brew install weaveworks/tap/eksctl
or on Linux use:
curl --silent --location "https://github.com/weaveworks/eksctl/releases/latest/download/eksctl_$(uname -s)_amd64.tar.gz" | tar xz -C /tmp
sudo mv /tmp/eksctl /usr/local/bin
A prerequisite for the EBS CSI driver to work is to have an existing AWS Identity and Access Management (IAM) OpenID Connect (OIDC) provider for your cluster. This IAM OIDC provider can be enabled with the following command:
eksctl utils associate-iam-oidc-provider --region=eu-central-1 --cluster=$(pulumi stack output clusterName) --approve
Now having eksctl
in place, create the IAM role:
eksctl create iamserviceaccount \
--name ebs-csi-controller-sa \
--namespace kube-system \
--cluster $(pulumi stack output clusterName) \
--attach-policy-arn arn:aws:iam::aws:policy/service-role/AmazonEBSCSIDriverPolicy \
--approve \
--role-only \
--role-name AmazonEKS_EBS_CSI_DriverRole
As you can see AWS maintains a managed policy for us we can simply use (AWS maintains a managed policy, available at ARN arn:aws:iam::aws:policy/service-role/AmazonEBSCSIDriverPolicy
). Only if you use encrypted EBS drives you need to additionally add configuration to the policy.
The command...
...deploys an AWS CloudFormation stack that creates an IAM role, attaches the IAM policy to it, and annotates the existing ebs-csi-controller-sa service account with the Amazon Resource Name (ARN) of the IAM role.
Now we can finally add the EBS CSI add-on. Therefor we also need the AWS Account id which we can obtain by running aws sts get-caller-identity --query Account --output text
(see https://stackoverflow.com/questions/33791069/quick-way-to-get-aws-account-number-from-the-aws-cli-tools). Now the eksctl create addon
command looks like this:
eksctl create addon --name aws-ebs-csi-driver --cluster $(pulumi stack output clusterName) --service-account-role-arn arn:aws:iam::$(aws sts get-caller-identity --query Account --output text):role/AmazonEKS_EBS_CSI_DriverRole --force
https://hub.tekton.dev/tekton/task/create-gitlab-release
https://gitlab.com/gitlab-org/gitlab/-/issues/213360
https://gitlab.com/gitlab-org/gitlab-runner/-/issues/28051
https://www.youtube.com/watch?v=skcLi9-WTkA
https://cloud.google.com/tekton
Why not use a Chat bot to do the job?
-> If I push, a bot asks me to assign another person for the pull request
- see Kubernetes github project: prow comments on every pull request
https://archive.fosdem.org/2021/schedule/event/ci_on_gitlab_ringing_gitlab_tekton_and_prow_together/
prow is GitHub specific, so what's the alternative?
rather than being GitHub specific Lighthouse uses jenkins-x/go-scm so it can support any Git provider (while Lighthouse reuses the Prow plugin source code and a bunch of plugins from Prow)
https://jenkins-x.io/v3/about/what/#chatops
With the ever growing number of microservices needing automation, Jenkins X provides the ability to interact with pipelines via comments on pull requests. Lighthouse has evolved from Prow which is used heavily in the Kubernetes ecosystem to provide a consistent developer workflow for triggering tests, approvals, hold and other common commands developers use in their everyday activities
https://github.com/jenkins-x/lighthouse
Lighthouse is a lightweight ChatOps based webhook handler which can trigger Jenkins X Pipelines, Tekton Pipelines or Jenkins Jobs based on webhooks from multiple git providers such as GitHub, GitHub Enterprise, BitBucket Server and GitLab.
https://github.com/jenkins-x/lighthouse/blob/main/docs/install_lighthouse_with_tekton.md
lighthouse-foghorn: watches execution of Tekton PipelineRun
triggered by lighthouse and updates AND BLOCKS pull requests from beeing merged until Tekton pipelines succeeded
And more:
--> Lighthouse Tekton integration misses JSON-payload proceeding to Tekton Pipelines/Tasks!
https://github.com/tektoncd/experimental/tree/main/commit-status-tracker
https://github.com/tektoncd/hub#deploy-your-own-instance
You can deploy your own instance of Tekton Hub. You can find the documentation https://github.com/tektoncd/hub/blob/main/docs/DEPLOYMENT.md
https://piotrminkowski.com/2021/08/05/kubernetes-ci-cd-with-tekton-and-argocd/