Creates a complete GitOps-based operational stack on your Kubernetes clusters:
- Deployment: GitOps via Argo CD with a ready-to-use repo structure
- Monitoring: Prometheus and Grafana
- Secrets Management: Vault and External Secrets Operator
- Notifications/Alerts: Grafana and ArgoCD can be predefined with either an external mailserver or MailHog for demo purposes.
- Pipelines: Example applications using Jenkins with the gitops-build-lib and SCM-Manager
- Ingress Controller: ingress-nginx
- Certificate Management: (planned)
- Runs on:
- local cluster (try it with only one command),
- in the public cloud,
- and even air-gapped environments (work in progress).
The gitops-playground is derived from our experiences in consulting, operating the myCloudogu platform and is used in our GitOps trainings for both Flux and ArgoCD.
You can try the GitOps Playground on a local Kubernetes cluster by running a single command:
bash <(curl -s \
https://raw.githubusercontent.com/cloudogu/gitops-playground/main/scripts/init-cluster.sh) \
&& docker run --rm -t --pull=always -u $(id -u) \
-v ~/.config/k3d/kubeconfig-gitops-playground.yaml:/home/.kube/config \
--net=host \
ghcr.io/cloudogu/gitops-playground --yes --argocd --ingress-nginx --base-url=http://localhost
# If you want to try all features, you might want to add these params: --mail --monitoring --vault=dev
Note that on some linux distros like debian do not support subdomains of localhost.
There you might have to use --base-url=http://local.gd
(see local ingresses).
See the list of applications to get started.
We recommend running this command as an unprivileged user, that is inside the docker group.
- What is the GitOps Playground?
- Installation
- Stack
- Development
The GitOps Playground provides a reproducible environment for setting up a GitOps-Stack. It provides an image for automatically setting up a Kubernetes Cluster including CI-server (Jenkins), source code management (SCM-Manager), Monitoring and Alerting (Prometheus, Grafana, MailHog), Secrets Management (Hashicorp Vault and External Secrets Operator) and of course Argo CD as GitOps operator.
The playground also deploys a number of example applications.
The GitOps Playground lowers the barriers for operating your application on Kubernetes using GitOps.
It creates a complete GitOps-based operational stack on your Kubernetes clusters.
No need to read lots of books and operator
docs, getting familiar with CLIs, ponder about GitOps Repository folder structures and promotion to different environments, etc.
The GitOps Playground is a pre-configured environment to see GitOps in motion, including more advanced use cases like
notifications, monitoring and secret management.
In addition to creating an operational stack in production, you can run the playground locally, for learning and developing new features.
We aim to be compatible with various environments, e.g. OpenShift and in an air-gapped network. The support for these is work in progress.
There a several options for running the GitOps playground
- on a local k3d cluster Works best on Linux, but is possible on Windows and Mac.
- on a remote k8s cluster
- each with the option
- to use an external Jenkins, SCM-Manager and registry (this can be run in production, e.g. with a Cloudogu Ecosystem) or
- to run everything inside the cluster (for demo only)
The diagrams below show an overview of the playground's architecture and three scenarios for running the playground. For a simpler overview including all optional features such as monitoring and secrets management see intro at the very top.
Note that running Jenkins inside the cluster is meant for demo purposes only. The third graphic shows our production scenario with the Cloudogu EcoSystem (CES). Here better security and build performance is achieved using ephemeral Jenkins build agents spawned in the cloud.
Playground on local machine | Production environment with Cloudogu EcoSystem |
---|---|
You can apply the GitOps playground to
- a local k3d cluster (see docs or script for more details):
bash <(curl -s \ https://raw.githubusercontent.com/cloudogu/gitops-playground/main/scripts/init-cluster.sh)
- a remote k8s cluster on Google Kubernetes Engine (e.g. via Terraform, see our docs),
- or almost any k8s cluster.
Note that if you want to deploy Jenkins inside the cluster, you either need Docker as container runtime or set Jenkins up to run its build on an agent that provides Docker.
For the local cluster, you can avoid hitting DockerHub's rate limiting by using a mirror via the --docker-io-registry-mirror
parameter.
For example:
bash <(curl -s \
https://raw.githubusercontent.com/cloudogu/gitops-playground/main/scripts/init-cluster.sh) --docker-io-registry-mirror https://mirror.gcr.io
This parameter is passed on the containerd used by k3d.
In addition, the Jobs run by Jenkins are using the host's Docker daemon.
To avoid rate limits there, you might have to configure a mirror there as well.
This can be done in the /etc/docker/daemon.json
or in the config of Docker Desktop.
For example:
{
"registry-mirrors": ["https://mirror.gcr.io"]
}
You can apply the playground to your cluster using our container image ghcr.io/cloudogu/gitops-playground
.
On success, the container prints a little intro on how to get started with the GitOps playground.
There are several options for running the container:
- For local k3d cluster, we recommend running the image as a local container via
docker
- For remote clusters (e.g. on GKE) you can run the image inside a pod of the target cluster via
kubectl
.
All options offer the same parameters, see below.
When connecting to k3d it is easiest to apply the playground via a local container in the host network and pass k3d's kubeconfig.
CLUSTER_NAME=gitops-playground
docker pull ghcr.io/cloudogu/gitops-playground
docker run --rm -t -u $(id -u) \
-v ~/.config/k3d/kubeconfig-${CLUSTER_NAME}.yaml:/home/.kube/config \
--net=host \
ghcr.io/cloudogu/gitops-playground # additional parameters go here
Note:
docker pull
in advance makes sure you have the newest image, even if you ran this command before.
Of course, you could also specify a specific version of the image.- Using the host network makes it possible to determine
localhost
and to use k3d's kubeconfig without altering, as it access the API server via a port bound to localhost. - We run as the local user in order to avoid file permission issues with the
kubeconfig-${CLUSTER_NAME}.yaml.
- If you experience issues and want to access the full log files, use the following command while the container is running:
docker exec -it \
$(docker ps -q --filter ancestor=ghcr.io/cloudogu/gitops-playground) \
bash -c -- 'tail -f -n +1 /tmp/playground-log-*'
For remote clusters it is easiest to apply the playground via kubectl. You can find info on how to install kubectl here.
# Create a temporary ServiceAccount and authorize via RBAC.
# This is needed to install CRDs, etc.
kubectl create serviceaccount gitops-playground-job-executer -n default
kubectl create clusterrolebinding gitops-playground-job-executer \
--clusterrole=cluster-admin \
--serviceaccount=default:gitops-playground-job-executer
# Then start apply the playground with the following command:
# To access services on remote clusters, add either --remote or --ingress-nginx --base-url=$yourdomain
kubectl run gitops-playground -i --tty --restart=Never \
--overrides='{ "spec": { "serviceAccount": "gitops-playground-job-executer" } }' \
--image ghcr.io/cloudogu/gitops-playground \
-- --yes --argocd # additional parameters go here.
# If everything succeeded, remove the objects
kubectl delete clusterrolebinding/gitops-playground-job-executer \
sa/gitops-playground-job-executer pods/gitops-playground -n default
In general docker run
should work here as well. But GKE, for example, uses gcloud and python in their kubeconfig.
Running inside the cluster avoids these kinds of issues.
The following describes more parameters and use cases.
You can get a full list of all options like so:
docker run -t --rm ghcr.io/cloudogu/gitops-playground --help
You can also use a configuration file to specify the parameters (--config-file
or --config-map
).
That file must be a YAML file.
Note that the config file is not yet a complete replacement for CLI parameters.
You can use --output-config-file
to output the current config as set by defaults and CLI parameters.
In addition, For easier validation and auto-completion, we provide a schema file.
For example in Jetbrains IntelliJ IDEA, you can use the schema for autocompletion and validation when you put the following at the beginning of your config file:
# $schema: https://raw.githubusercontent.com/cloudogu/gitops-playground/main/docs/configuration.schema.json
If you work with an older version, you can use a specific git commit ID instead of main
in the schema URL.
Then use the context assistant to enable coding assistance or fill in all available properties. See here for the full manual.
docker run --rm -t --pull=always -u $(id -u) \
-v ~/.config/k3d/kubeconfig-gitops-playground.yaml:/home/.kube/config \
-v $(pwd)/gitops-playground.yaml:/config/gitops-playground.yaml \
--net=host \
ghcr.io/cloudogu/gitops-playground --yes --argocd --config-file=/config/gitops-playground.yaml
Create the serviceaccount and clusterrolebinding
$ cat config.yaml # for example
features:
monitoring:
active: true
# Convention:
# Find the ConfigMap inside the current namespace for the config map
# From the config map, pick the key "config.yaml"
kubectl create configmap gitops-config --from-file=config.yaml
kubectl run gitops-playground -i --tty --restart=Never \
--overrides='{ "spec": { "serviceAccount": "gitops-playground-job-executer" } }' \
--image ghcr.io/cloudogu/gitops-playground \
-- --yes --argocd --config-map=gitops-config
Afterwards, you might want to do a clean up. In addition, you might want to delete the config-map as well.
kubectl delete cm gitops-config
In the default installation the GitOps-Playground comes without an Ingress-Controller.
We use Nginx as default Ingress-Controller.
It can be enabled via the configfile or parameter --ingress-nginx
.
In order to make use of the ingress controller, it is recommended to use it in conjunction with --base-url
, which will create Ingress
objects for all components of the GitOps playground.
The ingress controller is based on the helm chart ingress-nginx
.
Additional parameters from this chart's values.yaml file can be added to the installation through the gitops-playground configuration file.
Example:
features:
ingressNginx:
active: true
helm:
values:
controller:
replicaCount: 4
In this Example we override the default controller.replicaCount
(GOP's default is 2).
This config file is merged with precedence over the defaults set by
- the GOP and
- the charts itself.
It is possible to deploy Ingress
objects for all components. You can either
- Set a common base url (
--base-url=https://example.com
) or - individual URLS:
--argocd-url https://argocd.example.com
--grafana-url https://grafana.example.com
--vault-url https://vault.example.com
--mailhog-url https://mailhog.example.com
--petclinic-base-domain petclinic.example.com
--nginx-base-domain nginx.example.com
- or both, where the individual URLs take precedence.
Note:
jenkins-url
andscmm-url
are for external services and do not lead to ingresses, but you can set them via--base-url
for now.- In order to make use of the
Ingress
you need an ingress controller. If your cluster does not provide one, the Playground can deploy one for you, via the--ingress-nginx
parameter.
- By default, the ingresses are built as subdomains of
--base-url
. - You can change this behaviour using the parameter
--url-separator-hyphen
. - With this, hyphen are used instead of dots to separate application name from base URL.
- Examples:
--base-url=https://xyz.example.org
:argocd.xyz.example.org
(default)--base-url=https://xyz.example.org
:argocd-xyz.example.org
(--url-separator-hyphen
)
- This is useful when you have a wildcard certificate for the TLD, but use a subdomain as base URL.
Here, browsers accept the validity only for the first level of subdomains.
The ingresses can also be used when running the playground on your local machine:
- Ingresses might be easier to remember than arbitrary port numbers and look better in demos
- With ingresses, we can execute our local clusters in higher isolation or multiple playgrounds concurrently
- Ingresses are required for running on Windows/Mac.
To use them locally,
- init your cluster (
init-cluster.sh
). - apply your playground with the following parameters
--base-url=http://localhost
- this is possible on Windows (tested on 11), Mac (tested on Ventura) or when using Linux with systemd-resolved (default in Ubuntu, not Debian)
As an alternative, you could add all*.localhost
entries to yourhosts
file.
Usekubectl get ingress -A
to get a full list - Then, you can reach argocd on
http://argocd.localhost
, for example
- this is possible on Windows (tested on 11), Mac (tested on Ventura) or when using Linux with systemd-resolved (default in Ubuntu, not Debian)
--base-url=http://local.gd
(or127.0.0.1.nip.io
,127.0.0.1.sslip.io
, or others)- This should work for all other machines that have access to the internet without further config
- Then, you can reach argocd on
http://argocd.local.gd
, for example
- Note that when using port 80, the URLs are shorter, but you run into issues because port 80 is regarded as a privileged port.
Java applications seem not to be able to reach
localhost:80
or even127.0.0.1:80
(NoRouteToHostException
) - You can change the port using
init-cluster.sh --bind-ingress-port=8080
.
When you do, make sure to append the same port when applying the playground:--base-url=http://localhost:8080
- If your setup requires you to bind to a specific interface, you can just pass it with e.g.
--bind-ingress-port=127.0.0.1:80
--argocd
- deploy Argo CD GitOps operator
โ ๏ธ Note that switching between operators is not supported.
That is, expect errors (for example with cluster-resources) if you apply the playground once with Argo CD and the next time without it. We recommend resetting the cluster withinit-cluster.sh
beforehand.
See our Quickstart Guide on how to set up the instance.
Then set the following parameters.
# Note:
# * In this case --password only sets the Argo CD admin password (Jenkins and
# SCMM are external)
# * Insecure is needed, because the local instance will not have a valid cert
--jenkins-url=https://192.168.56.2/jenkins \
--scmm-url=https://192.168.56.2/scm \
--jenkins-username=admin \
--jenkins-password=yourpassword \
--scmm-username=admin \
--scmm-password=yourpassword \
--password=yourpassword \
--insecure
Using Google Container Registry (GCR) fits well with our cluster creation example via Terraform on Google Kubernetes Engine (GKE), see our docs.
Note that you can get a free CES demo instance set up with a Kubernetes Cluster as GitOps Playground here.
# Note: In this case --password only sets the Argo CD admin password (Jenkins
# and SCMM are external)
--jenkins-url=https://your-ecosystem.cloudogu.net/jenkins \
--scmm-url=https://your-ecosystem.cloudogu.net/scm \
--jenkins-username=admin \
--jenkins-password=yourpassword \
--scmm-username=admin \
--scmm-password=yourpassword \
--password=yourpassword \
--registry-url=eu.gcr.io \
--registry-path=yourproject \
--registry-username=_json_key \
--registry-password="$( cat account.json | sed 's/"/\\"/g' )"
Images used by the gitops-build-lib are set in the gitopsConfig
in each Jenkinsfile
of an application like that:
def gitopsConfig = [
...
buildImages : [
helm: 'ghcr.io/cloudogu/helm:3.10.3-1',
kubectl: 'bitnami/kubectl:1.29',
kubeval: 'ghcr.io/cloudogu/helm:3.10.3-1',
helmKubeval: 'ghcr.io/cloudogu/helm:3.10.3-1',
yamllint: 'cytopia/yamllint:1.25-0.7'
],...
To override each image in all the applications you can use following parameters:
--kubectl-image someRegistry/someImage:1.0.0
--helm-image someRegistry/someImage:1.0.0
--kubeval-image someRegistry/someImage:1.0.0
--helmkubeval-image someRegistry/someImage:1.0.0
--yamllint-image someRegistry/someImage:1.0.0
Images used by various tools and exercises can be configured using the following parameters:
--grafana-image someRegistry/someImage:1.0.0
--external-secrets-image someRegistry/someImage:1.0.0
--external-secrets-certcontroller-image someRegistry/someImage:1.0.0
--external-secrets-webhook-image someRegistry/someImage:1.0.0
--vault-image someRegistry/someImage:1.0.0
--nginx-image someRegistry/someImage:1.0.0
Note that specifying a tag is mandatory.
If you are using a remote cluster, you can set the --argocd-url
parameter so that argocd-notification messages have a
link to the corresponding application.
You can specify email addresses for notifications (note that by default, MailHog will not actually send emails)
--argocd-email-from
: Sender E-Mail address. Default is[email protected]
)--argocd-email-to-admin
: Alerts send to admin. Default is[email protected]
)--argocd-email-to-user
: Alerts send to user. Default is[email protected]
)
Set the parameter --monitoring
to enable deployment of monitoring and alerting tools like prometheus, grafana and mailhog.
See Monitoring tools for details.
You can specify email addresses for notifications (note that by default, MailHog will not actually send emails)
--grafana-email-from
: Sender E-Mail address. Default is[email protected]
--grafana-email-to
: Recipient E-Mail address. Default is[email protected]
The gitops-playground uses MailHog to showcase notifications.
Alternatively, you can configure an external mailserver.
Note that you can't use both at the same time.
If you set either --mailhog
or --mail
parameter, MailHog will be installed
If you set --smtp-*
parameters, a external Mailserver will be used and MailHog will not be deployed.
Set the parameter --mailhog
to enable MailHog.
This will deploy MailHog and configure Argo CD and Grafana to send mails to MailHog.
Sender and recipient email addresses can be set via parameters in some applications, e.g. --grafana-email-from
or --argocd-email-to-user
.
Parameters:
--mailhog
: Activate MailHog as internal Mailserver--mailhog-url
: Specify domain name (ingress) under which MailHog will be served
If you want to use an external Mailserver you can set it with these parameters
--smtp-address
: External Mailserver SMTP address or IP--smtp-port
: External Mailserver SMTP port--smtp-user
: External Mailserver login username--smtp-password
: External Mailservers login password. Make sure to put your password in single quotes.
This will configure Argo CD and Grafana to send mails using your external mailserver.
In addition you should set matching sender and recipient email addresses, e.g. --grafana-email-from
or --argocd-email-to-user
.
Set the parameter --vault=[dev|prod]
to enable deployment of secret management tools hashicorp vault and external
secrets operator.
See Secrets management tools for details.
For k3d, you can just k3d cluster delete gitops-playground
. This will delete the whole cluster.
If you want to delete k3d use rm .local/bin/k3d
.
To remove the playground without deleting the cluster, use the option --destroy
.
You need to pass the same parameters when deploying the playground to ensure that the destroy script
can authenticate with all tools.
Note that this option has limitations. It does not remove CRDs, namespaces, locally deployed SCM-Manager, Jenkins and registry, plugins for SCM-Manager and Jenkins
- In general: We cannot use the
host
network, so it's easiest to access via ingress controller and ingresses. --base-url=http://localhost --ingress-nginx
should work on both Windows and Mac.- In case of problems resolving e.g.
jenkins.localhost
, you could try using--base-url=http://local.gd
or similar, as described in local ingresses.
On macOS and when using the Windows Subsystem Linux on Windows (WSL), you can just run our TL;DR command after installing Docker.
For Windows, we recommend using Windows Subsystem for Linux version 2 (WSL2) with a native installation of Docker Engine, because it's easier to set up and less prone to errors.
For macOS, please increase the Memory limit in Docker Desktop (for your DockerVM) to be > 10 GB. Recommendation: 16GB.
bash <(curl -s \
https://raw.githubusercontent.com/cloudogu/gitops-playground/main/scripts/init-cluster.sh) \
&& docker run --rm -t --pull=always -u $(id -u) \
-v ~/.config/k3d/kubeconfig-gitops-playground.yaml:/home/.kube/config \
--net=host \
ghcr.io/cloudogu/gitops-playground --yes --argocd --ingress-nginx --base-url=http://localhost
# If you want to try all features, you might want to add these params: --mail --monitoring --vault=dev
When you encounter errors with port 80 you might want to use e.g.
init-cluster.sh) --bind-ingress-port=8080
and--base-url=http://localhost:8080
instead.
- As mentioned in the previous section, we recommend using WSL2 with a native Docker Engine.
- If you must, you can also run using Docker Desktop from native Windows console (see bellow)
- However, there seems to be a problem when the Jenkins Jobs running the playground access docker, e.g.
$ docker run -t -d -u 0:133 -v ... -e ******** bitnami/kubectl:1.25.4 cat
docker top e69b92070acf3c1d242f4341eb1fa225cc40b98733b0335f7237a01b4425aff3 -eo pid,comm
process apparently never started in /tmp/gitops-playground-jenkins-agent/workspace/xample-apps_petclinic-plain_main/.configRepoTempDir@tmp/durable-7f109066
(running Jenkins temporarily with -Dorg.jenkinsci.plugins.durabletask.BourneShellScript.LAUNCH_DIAGNOSTICS=true might make the problem clearer)
Cannot contact default-1bg7f: java.nio.file.NoSuchFileException: /tmp/gitops-playground-jenkins-agent/workspace/xample-apps_petclinic-plain_main/.configRepoTempDir@tmp/durable-7f109066/output.txt
- In Docker Desktop, it's recommended to use WSL2 as backend.
- Using the Hyper-V backend should also work, but we experienced random
CrashLoopBackoff
s of running pods due to liveness probe timeouts.
Same as for macOS, increasing the Memory limit in Docker Desktop (for your DockerVM) to be > 10 GB might help.
Recommendation: 16GB.
Here is how you can start the playground from a Windows-native PowerShell console:
- Install k3d, see init-cluster.sh for
K3D_VERSION
, e.g. usingwinget
winget install k3d --version x.y.z
- Create k3d cluster.
See
K3S_VERSION
in init-cluster.sh for$image
, then execute
$ingress_port = "80"
$registry_port = "30000"
$image = "rancher/k3s:v1.25.5-k3s2"
# Note that ou can query the image version used by playground like so:
# (Invoke-WebRequest -Uri 'https://raw.githubusercontent.com/cloudogu/gitops-playground/main/scripts/init-cluster.sh').Content -split "`r?`n" | Select-String -Pattern 'K8S_VERSION=|K3S_VERSION='
k3d cluster create gitops-playground `
--k3s-arg=--kube-apiserver-arg=service-node-port-range=8010-65535@server:0 `
-p ${ingress_port}:80@server:0:direct `
-v /var/run/docker.sock:/var/run/docker.sock@server:0 `
--image=${image} `
-p ${registry_port}:30000@server:0:direct
# Write $HOME/.config/k3d/kubeconfig-gitops-playground.yaml
k3d kubeconfig write gitops-playground
- Note that
- You can ignore the warning about docker.sock
- We're mounting the docker socket, so it can be used by the Jenkins Agents for the docker-plugin.
- Windows seems not to provide a group id for the docker socket. So the Jenkins Agents run as root user.
- If you prefer running with an unprivileged user, consider running on WSL2, Mac or Linux
- You could also add
-v gitops-playground-build-cache:/tmp@server:0
to persist the Cache of the Jenkins agent between restarts of k3d containers.
- Apply playground:
Note that when using a$registry_port
other than30000
append the command--internal-registry-port=$registry_port
bellow
docker run --rm -t --pull=always `
-v $HOME/.config/k3d/kubeconfig-gitops-playground.yaml:/home/.kube/config `
--net=host `
ghcr.io/cloudogu/gitops-playground --yes --argocd --ingress-nginx --base-url=http://localhost:$ingress_port # more params go here
As described above the GitOps playground comes with a number of applications. Some of them can be accessed via web.
- Jenkins
- SCM-Manager
- Argo CD
- Prometheus/Grafana
- Vault
- Example applications for each GitOps operator, some with staging and production environments.
The URLs of the applications depend on the environment the playground is deployed to. The following lists all applications and how to find out their respective URLs for a GitOps playground being deployed to local or remote cluster.
For remote clusters you need the external IP, no need to specify the port (everything running on port 80). Basically, you can get the IP address as follows:
kubectl -n "${namespace}" get svc "${serviceName}" \
--template="{{range .status.loadBalancer.ingress}}{{.ip}}{{end}}"
There is also a convenience script scripts/get-remote-url
. The script waits, if externalIP is not present, yet.
You could use this conveniently like so:
bash <(curl -s \
https://raw.githubusercontent.com/cloudogu/gitops-playground/main/scripts/get-remote-url) \
jenkins default
You can open the application in the browser right away, like so for example:
xdg-open $(bash <(curl -s \
https://raw.githubusercontent.com/cloudogu/gitops-playground/main/scripts/get-remote-url) \
jenkins default)
If deployed within the cluster, all applications can be accessed via: admin/admin
Note that you can change (and should for a remote cluster!) the password with the --password
argument.
There also is a --username
parameter, which is ignored for argocd. That is, for now argos username ist always admin
.
Argo CD's web UI is available at
- http://localhost:9092 (k3d)
scripts/get-remote-url argocd-server argocd
(remote k8s)--argocd-url
to specify domain name
Argo CD is installed in a production-ready way, that allows for operating Argo CD with Argo CD, using GitOps and providing a repo per team pattern.
When installing the GitOps playground, the following steps are performed to bootstrap Argo CD:
- The following repos are created and initialized:
argocd
(management and config of Argo CD itself),example-apps
(example for a developer/application team's GitOps repo) andcluster-resources
(example for a cluster admin or infra/platform team's repo; see below for details)
- Argo CD is installed imperatively via a helm chart.
- Two resources are applied imperatively to the cluster: an
AppProject
calledargocd
and anApplication
calledbootstrap
. These are also contained within theargocd
repository.
From there everything is managed via GitOps. This diagram shows how it works.
- The
bootstrap
application manages the folderapplications
, which also containsbootstrap
itself.
With this, changes to thebootstrap
application can be done via GitOps. Thebootstrap
application also deploys other apps (App Of Apps pattern) - The
argocd
application manages the folderargocd
which contains Argo CD's resources as an umbrella helm chart.
The umbrella chart pattern allows describing the actual values invalues.yaml
and deploying additional resources (such as secrets and ingresses) via thetemplates
folder. The actual ArgoCD chart is declared in theChart.yaml
- The
Chart.yaml
contains the Argo CD helm chart asdependency
. It points to a deterministic version of the Chart (pinned viaChart.lock
) that is pulled from the Chart repository on the internet.
This mechanism can be used to upgrade Argo CD via GitOps. See the Readme of the argocd repository for details. - The
projects
application manages theprojects
folder, that contains the followingAppProjects
:- the
argocd
project, used for bootstrapping - the built-in
default
project (which is restricted to eliminate threats to security) - one project per team (to implement least privilege and also notifications per team):
cluster-resources
(for platform admin, needs more access to cluster) andexample-apps
(for developers, needs less access to cluster)
- the
- The
cluster-resources
application points to thecluster-resources
git repository (argocd
folder), which has the typical folder structure of a GitOps repository (explained in the next step). This way, the platform admins use GitOps in the same way as their "customers" (the developers) and can provide better support. - The
example-apps
application points to theexample-apps
git repository (argocd
folder again). Like thecluster-resources
, it also has the typical folder structure of a GitOps repository:apps
- contains the kubernetes resources of all applications (the actual YAML)argocd
- contains Argo CDApplications
that point to subfolders ofapps
(App Of Apps pattern, again)misc
- contains kubernetes resources, that do not belong to specific applications (namespaces, RBAC, resources used by multiple apps, etc.)
- The
misc
application points to themisc
folder - The
my-app-staging
application points to theapps/my-app/staging
folder within the same repo. This provides a folder structure for release promotion. Themy-app-*
applications implement the Environment per App Pattern. This pattern allows each application to have its own environments, e.g. production and staging or none at all. Note that the actual YAML here could either be pushed manually or using the CI server. The applications contain examples that push config changes from the app repo to the GitOps repo using the CI server. This implementation mixes the Repo per Team and Repo per App patterns - The corresponding production environment is realizing using the
my-app-production
application, that points to theapps/my-app/production
folder within the same repo.
Note that it is recommended to protect theproduction
folders from manual access, if supported by the SCM of your choice.
Alternatively, instead of different YAMLs files as used in the diagram, these applications could be realized as- Two applications in the same YAML (implemented in the playground, see e.g.
petclinic-plain.yaml
) - Two application with the same name in different namespaces, when ArgoCD is enabled to search for applications
within different namespaces (implemented in the playground, see
Argo CD's values.yaml -
application.namespaces
setting) - One
ApplicationSet
, using thegit
generator for directories (not used in GitOps playground, yet)
- Two applications in the same YAML (implemented in the playground, see e.g.
To keep things simpler, the GitOps playground only uses one kubernetes cluster, effectively implementing the Standalone
pattern. However, the repo structure could also be used to serve multiple clusters, in a Hub and Spoke pattern:
Additional clusters could either be defined in the vaules.yaml
or as secrets via the templates
folder.
We're also working on an optional implementation of the namespaced pattern, using the Argo CD operator.
And advanced question: Why does the GitOps playground not use the argocd-autopilot?
The short answer is: As of 2023-05, version 0.4.15 it looks far from ready for production.
Here is a diagram that shows how the repo structure created by autopilot looks like:
Here are some thoughts why we deem it not a good fit for production:
- The version of ArgoCD is not pinned.
- Instead, the
kustomization.yaml
(3๏ธ in the diagram) points to abase
within the autopilot repo, which in turn points to thestable
branch of the Argo CD repo. - While it might be possible to pin the version using Kustomize, this is not the default and looks complicated.
- A non-deterministic version calls for trouble. Upgrades of Argo CD might happen unnoticed.
- What about breaking changes? What about disaster recovery?
- Instead, the
- The repository structure autopilot creates is more complicated (i.e. difficult to understand and maintain) than
the one used in the playground
- Why is the
autopilot-bootstrap
application (1๏ธ in the diagram) not within the GitOps repo and lives only in the cluster? - The approach of an
ApplicationSet
within theAppProject
's yaml pointing to aconfig.json
(more difficult to write than YAML) is difficult to grasp (4๏ธ and 6๏ธ in the diagram) - The
cluster-resources
ApplicationSet
is a good approach to multi-cluster but again, requires writing JSON (4๏ธ in the diagram).
- Why is the
- Projects are used to realize environments (6๏ธ and 7๏ธ in the diagram).
How would we separate teams in this monorepo structure?
One idea would be to use multiple Argo CD instances, realising a Standalone pattern. This would mean that every team would have to manage its own ArgoCD instance.
How could this task be delegated to a dedicated platform team? These are the questions that lead to the structure realized in the GitOps playground.
The playground installs cluster-resources (like prometheus, grafana, vault, external secrets operator, etc.) via the repo
argocd/cluster-resources
. See ADR for more details.
When installing without Argo CD, the tools are installed using helm imperatively, we fall back to using imperative helm installation as a kind of neutral ground.
Jenkins is available at
- http://localhost:9090 (k3d)
scripts/get-remote-url jenkins default
(remote k8s)
You can enable browser notifications about build results via a button in the lower right corner of Jenkins Web UI.
Note that this only works when using localhost
or https://
.
You can set an external jenkins server via the following parameters when applying the playground. See parameters for examples.
--jenkins-url
,--jenkins-username
,--jenkins-password
Note that the example applications pipelines will only run on a Jenkins that uses agents that provide
a docker host. That is, Jenkins must be able to run e.g. docker ps
successfully on the agent.
The user has to have the following privileges:
- install plugins
- set credentials
- create jobs
- restarting
SCM-Manager is available at
- http://localhost:9091 (k3d)
scripts/get-remote-url scmm-scm-manager default
(remote k8s)
You can set an external SCM-Manager via the following parameters when applying the playground. See Parameters for examples.
--scmm-url
,--scmm-username
,--scmm-password
The user on the scm has to have privileges to:
- add / edit users
- add / edit permissions
- add / edit repositories
- add / edit proxy
- install plugins
Set the parameter --monitoring
so the kube-prometheus-stack
via its helm-chart
is being deployed including dashboards for
- ArgoCD
- Ingress Nginx Controller
- Prometheus
- SCMManager
- Jenkins.
This leads to the following tools to be exposed:
- Mailhog
- http://localhost:9094 (k3d)
scripts/get-remote-url mailhog monitoring
(remote k8s)--mailhog-url
to specify domain name
- Grafana
- http://localhost:9095 (k3d)
scripts/get-remote-url kube-prometheus-stack-grafana monitoring
(remote k8s)--grafana-url
to specify domain name
Grafana can be used to query and visualize metrics via prometheus. Prometheus is not exposed by default.
In addition, argocd-notifications is set up. Applications deployed with Argo CD now will alert via email to mailhog the sync status failed, for example.
Note that this only works with Argo CD so far
Via the vault
parameter, you can deploy Hashicorp Vault and the External Secrets Operator into your GitOps playground.
With this, the whole flow from secret value in Vault to kubernetes Secret
via External Secrets Operator can be seen in
action:
For this to work, the GitOps playground configures the whole chain in Kubernetes and vault (when dev mode is used):
- In k8s
namespaces
argocd-staging
andargocd-production
:- Creates
SecretStore
andServiceAccount
(used to authenticate with vault) - Creates
ExternalSecrets
- Creates
- In Vault:
- Create secrets for staging and prod
- Create a human user for changing the secrets
- Authorizes the service accounts on those secrets
- Creates an example app that uses the
secrets
For testing you can set the parameter --vault=dev
to deploy vault in development mode. This will lead to
- vault being transient, i.e. all changes during runtime are not persisted. Meaning a restart will reset to default.
- Vault is initialized with some fixed secrets that are used in the example app, see below.
- Vault authorization is initialized with service accounts used in example
SecretStore
s for external secrets operator - Vault is initialized with the usual
admin/admin
account (can be overriden with--username
and--password
)
The secrets are then picked up by the vault-backend
SecretStore
s (connects External Secrets Operator with Vault) in
the namespace argocd-staging
and argocd-production
namespaces
You can reach the vault UI on
- http://localhost:8200 (k3d)
scripts/get-remote-url vault-ui secrets
(remote k8s)--vault-url
to specify domain name- You can log in vie the user account mentioned above.
If necessary, the root token can be found on the log:kubectl logs -n secrets vault-0 | grep 'Root Token'
When using vault=prod
you'll have to initialize vault manually but on the other hand it will persist changes.
If you want the example app to work, you'll have to manually
- set up vault, unseal it and
- authorize the
vault
service accounts inargocd-production
andargocd-staging
namspaces. SeeSecretStore
s and dev-post-start.sh for an example.
With vault in dev
mode and ArgoCD enabled, the example app applications/nginx/argocd/helm-jenkins
will be deployed
in a way that exposes the vault secrets secret/<environment>/nginx-secret
via HTTP on the URL http://<host>/secret
,
for example http://localhost:30024/secret
.
While exposing secrets on the web is a very bad practice, it's very good for demoing auto reload of a secret changed in vault.
To demo this, you could
- change the staging secret
- Wait for the change to show on the web, e.g. like so
while ; do echo -n "$(date '+%Y-%m-%d %H:%M:%S'): " ; \
curl http://localhost:30024/secret/ ; echo; sleep 1; done
This usually takes between a couple of seconds and 1-2 minutes.
This time consists of ExternalSecret
's refreshInterval
, as well as the kubelet sync period
(defaults to 1 Minute)
- cache propagation delay
The following video shows this demo in time-lapse:
secrets-demo-video.mp4
The playground comes with example applications that allow for experimenting with different GitOps features.
All applications are deployed via separated application and GitOps repos:
- Separation of app repo (e.g.
petclinic-plain
) and GitOps repo (e.g.argocd/example-app
) - Config is maintained in app repo,
- CI Server writes to GitOps repo and creates PullRequests.
The applications implement a simple staging mechanism:
- After a successful Jenkins build, the staging application will be deployed into the cluster by the GitOps operator.
- Deployment of production applications can be triggered by accepting pull requests.
- For some applications working without CI Server and committing directly to the GitOps repo is pragmatic
(e.g. 3rd-party-application like NGINX, likeargocd/nginx-helm-umbrella
)
Note that the GitOps-related logic is implemented in the gitops-build-lib for Jenkins. See the README there for more options like
- staging,
- resource creation,
- validation (fail early / shift left).
Please note that it might take about a minute after the pull request has been accepted for the GitOps operator to start deploying. Alternatively, you can trigger the deployment via ArgoCD's UI or CLI.
Jenkinsfile for plain
deployment
- Staging
- local localhost:30020
- remote:
scripts/get-remote-url spring-petclinic-plain argocd-staging
--petclinic-base-domain
to specify base domain. Then usestaging.petclinic-plain.$base-domain
- Production
- local localhost:30021
- remote:
scripts/get-remote-url spring-petclinic-plain argocd-production
--petclinic-base-domain
to specify base domain. Then useproduction.petclinic-plain.$base-domain
Jenkinsfile for helm
deployment
- Staging
- local localhost:30022
- remote:
scripts/get-remote-url spring-petclinic-helm argocd-staging
--petclinic-base-domain
to specify base domain. Then usestaging.petclinic-helm.$base-domain
- Production
- local localhost:30023
- remote:
scripts/get-remote-url spring-petclinic-helm argocd-production
--petclinic-base-domain
to specify base domain. Then useproduction.petclinic-helm.$base-domain
- Staging
- local: localhost:30024
- remote:
scripts/get-remote-url nginx argocd-staging
--nginx-base-domain
to specify base domain. Then usestaging.nginx.$base-domain
- Production
- local: localhost:30025
- remote:
scripts/get-remote-url nginx argocd-production
--nginx-base-domain
to specify base domain. Then useproduction.nginx.$base-domain
- Application name:
nginx-helm-umbrella
- local: localhost:30026
- remote:
scripts/get-remote-url nginx-helm-umbrella argocd-production
--nginx-base-domain
to specify base domain. Then useproduction.nginx-helm-umbrella.$base-domain