Git Generator

The Git generator contains two subtypes: the Git directory generator, and Git file generator.

Warning

Git generators are often used to make it easier for (non-admin) developers to create Applications. If the project field in your ApplicationSet is templated, developers may be able to create Applications under Projects with excessive permissions. For ApplicationSets with a templated project field, the source of truth must be controlled by admins - in the case of git generators, PRs must require admin approval.

Git Generator: Directories

The Git directory generator, one of two subtypes of the Git generator, generates parameters using the directory structure of a specified Git repository.

Suppose you have a Git repository with the following directory structure:

  1. ├── argo-workflows
  2. ├── kustomization.yaml
  3. └── namespace-install.yaml
  4. └── prometheus-operator
  5. ├── Chart.yaml
  6. ├── README.md
  7. ├── requirements.yaml
  8. └── values.yaml

This repository contains two directories, one for each of the workloads to deploy:

  • an Argo Workflow controller kustomization YAML file
  • a Prometheus Operator Helm chart

We can deploy both workloads, using this example:

  1. apiVersion: argoproj.io/v1alpha1
  2. kind: ApplicationSet
  3. metadata:
  4. name: cluster-addons
  5. namespace: argocd
  6. spec:
  7. generators:
  8. - git:
  9. repoURL: https://github.com/argoproj/argo-cd.git
  10. revision: HEAD
  11. directories:
  12. - path: applicationset/examples/git-generator-directory/cluster-addons/*
  13. template:
  14. metadata:
  15. name: '{{path.basename}}'
  16. spec:
  17. project: "my-project"
  18. source:
  19. repoURL: https://github.com/argoproj/argo-cd.git
  20. targetRevision: HEAD
  21. path: '{{path}}'
  22. destination:
  23. server: https://kubernetes.default.svc
  24. namespace: '{{path.basename}}'
  25. syncPolicy:
  26. syncOptions:
  27. - CreateNamespace=true

(The full example can be found here.)

The generator parameters are:

  • {{path}}: The directory paths within the Git repository that match the path wildcard.
  • {{path[n]}}: The directory paths within the Git repository that match the path wildcard, split into array elements (n - array index)
  • {{path.basename}}: For any directory path within the Git repository that matches the path wildcard, the right-most path name is extracted (e.g. /directory/directory2 would produce directory2).
  • {{path.basenameNormalized}}: This field is the same as path.basename with unsupported characters replaced with - (e.g. a path of /directory/directory_2, and path.basename of directory_2 would produce directory-2 here).

Note: The right-most path name always becomes {{path.basename}}. For example, for - path: /one/two/three/four, {{path.basename}} is four.

Note: If the pathParamPrefix option is specified, all path-related parameter names above will be prefixed with the specified value and a dot separator. E.g., if pathParamPrefix is myRepo, then the generated parameter name would be myRepo.path instead of path. Using this option is necessary in a Matrix generator where both child generators are Git generators (to avoid conflicts when merging the child generators’ items).

Whenever a new Helm chart/Kustomize YAML/Application/plain subdirectory is added to the Git repository, the ApplicationSet controller will detect this change and automatically deploy the resulting manifests within new Application resources.

As with other generators, clusters must already be defined within Argo CD, in order to generate Applications for them.

Exclude directories

The Git directory generator will automatically exclude directories that begin with . (such as .git).

The Git directory generator also supports an exclude option in order to exclude directories in the repository from being scanned by the ApplicationSet controller:

  1. apiVersion: argoproj.io/v1alpha1
  2. kind: ApplicationSet
  3. metadata:
  4. name: cluster-addons
  5. namespace: argocd
  6. spec:
  7. generators:
  8. - git:
  9. repoURL: https://github.com/argoproj/argo-cd.git
  10. revision: HEAD
  11. directories:
  12. - path: applicationset/examples/git-generator-directory/excludes/cluster-addons/*
  13. - path: applicationset/examples/git-generator-directory/excludes/cluster-addons/exclude-helm-guestbook
  14. exclude: true
  15. template:
  16. metadata:
  17. name: '{{path.basename}}'
  18. spec:
  19. project: "my-project"
  20. source:
  21. repoURL: https://github.com/argoproj/argo-cd.git
  22. targetRevision: HEAD
  23. path: '{{path}}'
  24. destination:
  25. server: https://kubernetes.default.svc
  26. namespace: '{{path.basename}}'

(The full example can be found here.)

This example excludes the exclude-helm-guestbook directory from the list of directories scanned for this ApplicationSet resource.

Exclude rules have higher priority than include rules

If a directory matches at least one exclude pattern, it will be excluded. Or, said another way, exclude rules take precedence over include rules.

As a corollary, which directories are included/excluded is not affected by the order of paths in the directories field list (because, as above, exclude rules always take precedence over include rules).

For example, with these directories:

  1. .
  2. └── d
  3. ├── e
  4. ├── f
  5. └── g

Say you want to include /d/e, but exclude /d/f and /d/g. This will not work:

  1. - path: /d/e
  2. exclude: false
  3. - path: /d/*
  4. exclude: true

Why? Because the exclude /d/* exclude rule will take precedence over the /d/e include rule. When the /d/e path in the Git repository is processed by the ApplicationSet controller, the controller detects that at least one exclude rule is matched, and thus that directory should not be scanned.

You would instead need to do:

  1. - path: /d/*
  2. - path: /d/f
  3. exclude: true
  4. - path: /d/g
  5. exclude: true

Or, a shorter way (using path.Match syntax) would be:

  1. - path: /d/*
  2. - path: /d/[fg]
  3. exclude: true

Root Of Git Repo

The Git directory generator can be configured to deploy from the root of the git repository by providing '*' as the path.

To exclude directories, you only need to put the name/path.Match of the directory you do not want to deploy.

  1. apiVersion: argoproj.io/v1alpha1
  2. kind: ApplicationSet
  3. metadata:
  4. name: cluster-addons
  5. namespace: argocd
  6. spec:
  7. generators:
  8. - git:
  9. repoURL: https://github.com/example/example-repo.git
  10. revision: HEAD
  11. directories:
  12. - path: '*'
  13. - path: donotdeploy
  14. exclude: true
  15. template:
  16. metadata:
  17. name: '{{path.basename}}'
  18. spec:
  19. project: "my-project"
  20. source:
  21. repoURL: https://github.com/example/example-repo.git
  22. targetRevision: HEAD
  23. path: '{{path}}'
  24. destination:
  25. server: https://kubernetes.default.svc
  26. namespace: '{{path.basename}}'

Git Generator: Files

The Git file generator is the second subtype of the Git generator. The Git file generator generates parameters using the contents of JSON/YAML files found within a specified repository.

Suppose you have a Git repository with the following directory structure:

  1. ├── apps
  2. └── guestbook
  3. ├── guestbook-ui-deployment.yaml
  4. ├── guestbook-ui-svc.yaml
  5. └── kustomization.yaml
  6. ├── cluster-config
  7. └── engineering
  8. ├── dev
  9. └── config.json
  10. └── prod
  11. └── config.json
  12. └── git-generator-files.yaml

The directories are:

  • guestbook contains the Kubernetes resources for a simple guestbook application
  • cluster-config contains JSON/YAML files describing the individual engineering clusters: one for dev and one for prod.
  • git-generator-files.yaml is the example ApplicationSet resource that deploys guestbook to the specified clusters.

The config.json files contain information describing the cluster (along with extra sample data):

  1. {
  2. "aws_account": "123456",
  3. "asset_id": "11223344",
  4. "cluster": {
  5. "owner": "cluster-admin@company.com",
  6. "name": "engineering-dev",
  7. "address": "https://1.2.3.4"
  8. }
  9. }

Git commits containing changes to the config.json files are automatically discovered by the Git generator, and the contents of those files are parsed and converted into template parameters. Here are the parameters generated for the above JSON:

  1. aws_account: 123456
  2. asset_id: 11223344
  3. cluster.owner: cluster-admin@company.com
  4. cluster.name: engineering-dev
  5. cluster.address: https://1.2.3.4

And the generated parameters for all discovered config.json files will be substituted into ApplicationSet template:

  1. apiVersion: argoproj.io/v1alpha1
  2. kind: ApplicationSet
  3. metadata:
  4. name: guestbook
  5. namespace: argocd
  6. spec:
  7. generators:
  8. - git:
  9. repoURL: https://github.com/argoproj/argo-cd.git
  10. revision: HEAD
  11. files:
  12. - path: "applicationset/examples/git-generator-files-discovery/cluster-config/**/config.json"
  13. template:
  14. metadata:
  15. name: '{{cluster.name}}-guestbook'
  16. spec:
  17. project: default
  18. source:
  19. repoURL: https://github.com/argoproj/argo-cd.git
  20. targetRevision: HEAD
  21. path: "applicationset/examples/git-generator-files-discovery/apps/guestbook"
  22. destination:
  23. server: '{{cluster.address}}'
  24. namespace: guestbook

(The full example can be found here.)

Any config.json files found under the cluster-config directory will be parameterized based on the path wildcard pattern specified. Within each file JSON fields are flattened into key/value pairs, with this ApplicationSet example using the cluster.address as cluster.name parameters in the template.

As with other generators, clusters must already be defined within Argo CD, in order to generate Applications for them.

In addition to the flattened key/value pairs from the configuration file, the following generator parameters are provided:

  • {{path}}: The path to the directory containing matching configuration file within the Git repository. Example: /clusters/clusterA, if the config file was /clusters/clusterA/config.json
  • {{path[n]}}: The path to the matching configuration file within the Git repository, split into array elements (n - array index). Example: path[0]: clusters, path[1]: clusterA
  • {{path.basename}}: Basename of the path to the directory containing the configuration file (e.g. clusterA, with the above example.)
  • {{path.basenameNormalized}}: This field is the same as path.basename with unsupported characters replaced with - (e.g. a path of /directory/directory_2, and path.basename of directory_2 would produce directory-2 here).
  • {{path.filename}}: The matched filename. e.g., config.json in the above example.
  • {{path.filenameNormalized}}: The matched filename with unsupported characters replaced with -.

Note: The right-most directory name always becomes {{path.basename}}. For example, from - path: /one/two/three/four/config.json, {{path.basename}} will be four. The filename can always be accessed using {{path.filename}}.

Note: If the pathParamPrefix option is specified, all path-related parameter names above will be prefixed with the specified value and a dot separator. E.g., if pathParamPrefix is myRepo, then the generated parameter name would be myRepo.path instead of path. Using this option is necessary in a Matrix generator where both child generators are Git generators (to avoid conflicts when merging the child generators’ items).

Webhook Configuration

When using a Git generator, ApplicationSet polls Git repositories every three minutes to detect changes. To eliminate this delay from polling, the ApplicationSet webhook server can be configured to receive webhook events. ApplicationSet supports Git webhook notifications from GitHub and GitLab. The following explains how to configure a Git webhook for GitHub, but the same process should be applicable to other providers.

Note

ApplicationSet exposes the webhook server as a service of type ClusterIP. An Ingress resource needs to be created to expose this service to the webhook source.

1. Create the webhook in the Git provider

In your Git provider, navigate to the settings page where webhooks can be configured. The payload URL configured in the Git provider should use the /api/webhook endpoint of your ApplicationSet instance (e.g. https://applicationset.example.com/api/webhook). If you wish to use a shared secret, input an arbitrary value in the secret. This value will be used when configuring the webhook in the next step.

Add Webhook

Note

When creating the webhook in GitHub, the “Content type” needs to be set to “application/json”. The default value “application/x-www-form-urlencoded” is not supported by the library used to handle the hooks

2. Configure ApplicationSet with the webhook secret (Optional)

Configuring a webhook shared secret is optional, since ApplicationSet will still refresh applications generated by Git generators, even with unauthenticated webhook events. This is safe to do since the contents of webhook payloads are considered untrusted, and will only result in a refresh of the application (a process which already occurs at three-minute intervals). If ApplicationSet is publicly accessible, then configuring a webhook secret is recommended to prevent a DDoS attack.

In the argocd-secret kubernetes secret, include the Git provider’s webhook secret configured in step 1.

Edit the Argo CD kubernetes secret:

  1. kubectl edit secret argocd-secret -n argocd

TIP: for ease of entering secrets, kubernetes supports inputting secrets in the stringData field, which saves you the trouble of base64 encoding the values and copying it to the data field. Simply copy the shared webhook secret created in step 1, to the corresponding GitHub/GitLab/BitBucket key under the stringData field:

  1. apiVersion: v1
  2. kind: Secret
  3. metadata:
  4. name: argocd-secret
  5. namespace: argocd
  6. type: Opaque
  7. data:
  8. ...
  9. stringData:
  10. # github webhook secret
  11. webhook.github.secret: shhhh! it's a github secret
  12. # gitlab webhook secret
  13. webhook.gitlab.secret: shhhh! it's a gitlab secret

After saving, please restart the ApplicationSet pod for the changes to take effect.