Git Repository Contents

Fleet will create bundles from a git repository. This happens either explicitly by specifying paths, or when a fleet.yaml is found.

Each bundle is created from paths in a GitRepo and modified further by reading the discovered fleet.yaml file. Bundle lifecycles are tracked between releases by the helm releaseName field added to each bundle. If the releaseName is not specified within fleet.yaml it is generated from GitRepo.name + path. Long names are truncated and a -<hash> prefix is added.

The git repository has no explicitly required structure. It is important to realize the scanned resources will be saved as a resource in Kubernetes so you want to make sure the directories you are scanning in git do not contain arbitrarily large resources. Right now there is a limitation that the resources deployed must gzip to less than 1MB.

How repos are scanned

Multiple paths can be defined for a GitRepo and each path is scanned independently. Internally each scanned path will become a bundle that Fleet will manage, deploy, and monitor independently.

The following files are looked for to determine the how the resources will be deployed.

File	Location	Meaning
Chart.yaml:	/ relative to path or custom path from fleet.yaml	The resources will be deployed as a Helm chart. Refer to the fleet.yaml for more options.
kustomization.yaml:	/ relative to path or custom path from fleet.yaml	The resources will be deployed using Kustomize. Refer to the fleet.yaml for more options.
fleet.yaml	Any subpath	If any fleet.yaml is found a new bundle will be defined. This allows mixing charts, kustomize, and raw YAML in the same repo
*.yaml	Any subpath	If a Chart.yaml or kustomization.yaml is not found then any .yaml or .yml file will be assumed to be a Kubernetes resource and will be deployed.
overlays/{name}	/ relative to path	When deploying using raw YAML (not Kustomize or Helm) overlays is a special directory for customizations.

Alternative scan, explicitly defined by the user

In addition to the previously described method, Fleet also supports a more direct, user-driven approach for defining Bundles.

In this mode, Fleet will load all resources found within the specified base directory. It will only attempt to locate a fleet.yaml file at the root of that directory if an options file is not explicitly provided. Unlike the traditional scanning method, this one is not recursive and does not attempt to find Bundle definitions other than those explicitly specified by the user.

Example File Structure

driven
     |___helm
     |      |__ fleet.yaml
     |
     |___simple
     |      |__ configmap.yaml
     |      |__ service.yaml
     |
     |___kustomize
            |__ base
            |      |__ kustomization.yaml
            |      |__ secret.yaml
            |
            |__ overlays
            |         |__ dev
            |         |     |__ kustomization.yaml
            |         |     |__ secret.yaml
            |         |__ prod
            |         |     |__ kustomization.yaml
            |         |     |__ secret.yaml
            |         |__ test
            |               |__ kustomization.yaml
            |               |__ secret.yaml
            |__ dev.yaml
            |__ prod.yaml
            |__ test.yaml

Corresponding GitRepo Definition

kind: GitRepo
apiVersion: fleet.cattle.io/v1alpha1
metadata:
  name: driven
  namespace: fleet-local
spec:
  repo: https://github.com/0xavi0/fleet-test-data
  branch: driven-scan-example
  bundles:
  - base: driven/helm
  - base: driven/simple
  - base: driven/kustomize
    options: dev.yaml
  - base: driven/kustomize
    options: test.yaml

In the example above, the user explicitly defines four Bundles to be generated.

• In the first case, the base directory is specified as driven/helm. As shown in the directory structure, this path contains a fleet.yaml file, which will be used to configure the Bundle.

• In the second case, the base directory is driven/simple, which contains only Kubernetes resource manifests (configmap.yaml and service.yaml). Since no fleet.yaml or options file is specified, Fleet will generate a Bundle using the default behavior—simply packaging all resources found within the directory.

• The third and fourth cases both reference the same base directory: driven/kustomize. However, each specifies a different options file (dev.yaml and test.yaml, respectively). These options files define overlay-specific configuration for each environment (e.g., dev, test) by selecting the appropriate kustomize overlay subdirectories and applying them on top of the shared base. Fleet will process these as distinct Bundles, even though they originate from the same base path, because the provided options files point to different configurations.

An example of the files used in the third and fourth Bundles would be the following: (These files follow the exact same format as fleet.yaml, but since we can now reference them by name, we can use one that best suits our needs)

namespace: kustomize-dev
kustomize:
  dir: "overlays/dev"

It is important to note that any path defined in these files must be relative to the base directory used when the Bundle was described.

For example, with the previously mentioned structure, we are defining the base directory as driven/kustomize. That is the directory we need to use as the root for the paths used in Kustomize files.

We could decide to place the dev.yaml file at the path driven/kustomize/overlays/dev (this is supported), and then define the Bundle as:

bundles:
    - base: driven/kustomize
      options: overlays/dev/dev.yaml

However, the path defined within dev.yaml should still be relative to driven/kustomize. This is because when Fleet reads the options files, it always uses the base directory as the root.

In other words, with the previous example... this would be incorrect:

namespace: kustomize-dev
kustomize:
  dir: "."

And the correct definition should still be:

namespace: kustomize-dev
kustomize:
  dir: "overlays/dev"

With this new way of defining Bundles, Fleet becomes much more direct and also simplifies the adoption of deployments using kustomize. In the example, we can see a complete kustomize use case where for each Bundle, we can specify which version we want.

With the previous scanning option, Fleet cannot determine which YAML we want to use to configure the Bundle, so it attempts to find it on its own (Which, at times, does not provide enough flexibility.)

Excluding files and directories from bundles

Fleet supports file and directory exclusion by means of .fleetignore files, in a similar fashion to how .gitignore files behave in git repositories:

• Glob syntax is used to match files or directories, using Golang's filepath.Match
• Empty lines are skipped, and can therefore be used to improve readability
• Characters like white spaces and # can be escaped with a backslash
• Trailing spaces are ignored, unless escaped
• Comments, ie lines starting with unescaped #, are skipped
• A given line can match a file or a directory, even if no separator is provided: eg. subdir/* and subdir are both valid .fleetignore lines, and subdir matches both files and directories called subdir
• A match may be found for a file or directory at any level below the directory where a .fleetignore lives, ie foo.yaml will match ./foo.yaml as well as ./path/to/foo.yaml
• Multiple .fleetignore files are supported. For instance, in the following directory structure, only root/something.yaml, bar/something2.yaml and foo/something.yaml will end up in a bundle:

root/
├── .fleetignore            # contains `ignore-always.yaml'
├── something.yaml
├── bar
│   ├── .fleetignore        # contains `something.yaml`
│   ├── ignore-always.yaml
│   ├── something2.yaml
│   └── something.yaml
└── foo
    ├── ignore-always.yaml
    └── something.yaml

This currently comes with a few limitations, the following not being supported:

• Double asterisks (**)
• Explicit inclusions with !

fleet.yaml

The fleet.yaml is an optional file that can be included in the git repository to change the behavior of how the resources are deployed and customized. The fleet.yaml is always at the root relative to the path of the GitRepo and if a subdirectory is found with a fleet.yaml a new bundle is defined that will then be configured differently from the parent bundle.

caution

Helm chart dependencies: Fleet automatically handles updating Helm chart dependencies, unless flag disableDependencyUpdate (false by default) is set to true.

If automatic dependencies updates are disabled, it is up to the user to fulfill the dependency list for the Helm charts. As such, you must manually run helm dependencies update $chart OR run helm dependencies build $chart prior to install. See the Fleet docs in Rancher for more information.

The available fields are documented in the fleet.yaml reference

For a private Helm repo, users can reference a secret from the git repo resource. See Using Private Helm Repositories for more information.

Using Helm Values

How changes are applied to values.yaml:

• Note that the most recently applied changes to the values.yaml will override any previously existing values.

• When changes are applied to the values.yaml from multiple sources at the same time, the values will update in the following order: helm.values -> helm.valuesFiles -> helm.valuesFrom. That means valuesFrom will take precedence over both, valuesFiles and values.

The targeting step can treat the values as a template and fill in information from the clusters.fleet.cattle.io resource. More information can be found in Helm values templating. This can be turned off in fleet.yaml, by setting disablePreProcess, e.g. to avoid conflicts with other templating languages.

Credentials in Values

If the chart generates certificates or passwords in its templates, these values must be overriden. Otherwise the chart could be continuously deployed as these values change.

Credentials loaded from the downstream cluster with valuesFrom are by default encrypted at rest, when data encryption is enabled in Kubernetes. Credentials contained in the default values.yaml file, or defined via values: or valuesFiles are not, as they are loaded from the repository when the bundle is created.

Hardened clusters should add the Fleet CRDs to the list of resources encrypted at rest, on the upstream cluster, when storing credentials in the bundles.

Using ValuesFrom

These examples showcase the style and format for using valuesFrom. ConfigMaps and Secrets should be created in downstream clusters.

Example ConfigMap:

apiVersion: v1
kind: ConfigMap
metadata:
  name: configmap-values
  namespace: default
data:
  values.yaml: |-
    replication: true
    replicas: 2
    serviceType: NodePort

Example Secret:

apiVersion: v1
kind: Secret
metadata:
  name: secret-values
  namespace: default
stringData:
  values.yaml: |-
    replication: true
    replicas: 3
    serviceType: NodePort

A secret like that, can be created from a YAML file secretdata.yaml by running the following kubectl command: kubectl create secret generic secret-values --from-file=values.yaml=secretdata.yaml

The resources can then be referenced from a fleet.yaml:

helm:
  chart: simple-chart
  valuesFrom:
    - secretKeyRef:
        name: secret-values
        namespace: default
        key: values.yaml
    - configMapKeyRef:
        name: configmap-values
        namespace: default
        key: values.yaml
  values:
    replicas: "4"

Per Cluster Customization

The GitRepo defines which clusters a git repository should be deployed to and the fleet.yaml in the repository determines how the resources are customized per target.

All clusters and cluster groups in the same namespace as the GitRepo will be evaluated against all targets of that GitRepo. The targets list is evaluated one by one and if there is a match the resource will be deployed to the cluster. If no match is made against the target list on the GitRepo then the resources will not be deployed to that cluster. Once a target cluster is matched the fleet.yaml from the git repository is then consulted for customizations. The targetCustomizations in the fleet.yaml will be evaluated one by one and the first match will define how the resource is to be configured. If no match is made the resources will be deployed with no additional customizations.

There are three approaches to matching clusters for both GitRepo targets and fleet.yaml targetCustomizations. One can use cluster selectors, cluster group selectors, or an explicit cluster group name. All criteria is additive so the final match is evaluated as "clusterSelector && clusterGroupSelector && clusterGroup". If any of the three have the default value it is dropped from the criteria. The default value is either null or "". It is important to realize that the value {} for a selector means "match everything."

targetCustomizations:
- name: all
  # Match everything
  clusterSelector: {}
- name: none
  # Selector ignored
  clusterSelector: null

When matching a cluster by name, make sure to use the name of the clusters.fleet.cattle.io resource. The Rancher UI also has a provisioning and a management cluster resource. Since the management cluster resource is not namespaced, its name is different and contains a random suffix.

targetCustomizations:
- name: prod
  clusterName: fleetname

See Mapping to Downstream Clusters for more information and a list of supported customizations.

Raw YAML Resource Customization

When using Kustomize or Helm the kustomization.yaml or the helm.values will control how the resource are customized per target cluster. If you are using raw YAML then the following simple mechanism is built-in and can be used. The overlays/ folder in the git repo is treated specially as folder containing folders that can be selected to overlay on top per target cluster. The resource overlay content uses a file name based approach. This is different from kustomize which uses a resource based approach. In kustomize the resource Group, Kind, Version, Name, and Namespace identify resources and are then merged or patched. For Fleet the overlay resources will override or patch content with a matching file name.

# Base files
deployment.yaml
svc.yaml

# Overlay files

# The following file will be added
overlays/custom/configmap.yaml
# The following file will replace svc.yaml
overlays/custom/svc.yaml
# The following file will patch deployment.yaml
overlays/custom/deployment_patch.yaml

A file named foo will replace a file called foo from the base resources or a previous overlay. In order to patch the contents of a file the convention of adding _patch. (notice the trailing period) to the filename is used. The string _patch. will be replaced with . from the file name and that will be used as the target. For example deployment_patch.yaml will target deployment.yaml. The patch will be applied using JSON Merge, Strategic Merge Patch, or JSON Patch. Which strategy is used is based on the file content. Even though JSON strategies are used, the files can be written using YAML syntax.

Cluster and Bundle State

See Cluster and Bundle state.

Nested GitRepo CRs

Nested GitRepo CRs (defining a GitRepo that points to a repository containing one or more GitRepo resources) is supported. You can use this feature to take advantage of GitOps in your GitRepo resources or, for example, to split complex scenarios into more than one GitRepo resource. When finding a GitRepo in a Bundle Fleet will simply deploy it as any other resource.

See this example.