Config Management Plugins

Argo CD’s “native” config management tools are Helm, Jsonnet, and Kustomize. If you want to use a different config management tools, or if Argo CD’s native tool support does not include a feature you need, you might need to turn to a Config Management Plugin (CMP).

The Argo CD “repo server” component is in charge of building Kubernetes manifests based on some source files from a Helm, OCI, or git repository. When a config management plugin is correctly configured, the repo server may delegate the task of building manifests to the plugin.

The following sections will describe how to create, install, and use plugins. Check out the example plugins for additional guidance.

Warning

Plugins are granted a level of trust in the Argo CD system, so it is important to implement plugins securely. Argo CD administrators should only install plugins from trusted sources, and they should audit plugins to weigh their particular risks and benefits.

Installing a config management plugin

Sidecar plugin

An operator can configure a plugin tool via a sidecar to repo-server. The following changes are required to configure a new plugin:

Write the plugin configuration file

Plugins will be configured via a ConfigManagementPlugin manifest located inside the plugin container.

  1. apiVersion: argoproj.io/v1alpha1
  2. kind: ConfigManagementPlugin
  3. metadata:
  4. # The name of the plugin must be unique within a given Argo CD instance.
  5. name: my-plugin
  6. spec:
  7. # The version of your plugin. Optional. If specified, the Application's spec.source.plugin.name field
  8. # must be <plugin name>-<plugin version>.
  9. version: v1.0
  10. # The init command runs in the Application source directory at the beginning of each manifest generation. The init
  11. # command can output anything. A non-zero status code will fail manifest generation.
  12. init:
  13. # Init always happens immediately before generate, but its output is not treated as manifests.
  14. # This is a good place to, for example, download chart dependencies.
  15. command: [sh]
  16. args: [-c, 'echo "Initializing..."']
  17. # The generate command runs in the Application source directory each time manifests are generated. Standard output
  18. # must be ONLY valid Kubernetes Objects in either YAML or JSON. A non-zero exit code will fail manifest generation.
  19. # To write log messages from the command, write them to stderr, it will always be displayed.
  20. # Error output will be sent to the UI, so avoid printing sensitive information (such as secrets).
  21. generate:
  22. command: [sh, -c]
  23. args:
  24. - |
  25. echo "{\"kind\": \"ConfigMap\", \"apiVersion\": \"v1\", \"metadata\": { \"name\": \"$ARGOCD_APP_NAME\", \"namespace\": \"$ARGOCD_APP_NAMESPACE\", \"annotations\": {\"Foo\": \"$ARGOCD_ENV_FOO\", \"KubeVersion\": \"$KUBE_VERSION\", \"KubeApiVersion\": \"$KUBE_API_VERSIONS\",\"Bar\": \"baz\"}}}"
  26. # The discovery config is applied to a repository. If every configured discovery tool matches, then the plugin may be
  27. # used to generate manifests for Applications using the repository. If the discovery config is omitted then the plugin
  28. # will not match any application but can still be invoked explicitly by specifying the plugin name in the app spec.
  29. # Only one of fileName, find.glob, or find.command should be specified. If multiple are specified then only the
  30. # first (in that order) is evaluated.
  31. discover:
  32. # fileName is a glob pattern (https://pkg.go.dev/path/filepath#Glob) that is applied to the Application's source
  33. # directory. If there is a match, this plugin may be used for the Application.
  34. fileName: "./subdir/s*.yaml"
  35. find:
  36. # This does the same thing as fileName, but it supports double-start (nested directory) glob patterns.
  37. glob: "**/Chart.yaml"
  38. # The find command runs in the repository's root directory. To match, it must exit with status code 0 _and_
  39. # produce non-empty output to standard out.
  40. command: [sh, -c, find . -name env.yaml]
  41. # The parameters config describes what parameters the UI should display for an Application. It is up to the user to
  42. # actually set parameters in the Application manifest (in spec.source.plugin.parameters). The announcements _only_
  43. # inform the "Parameters" tab in the App Details page of the UI.
  44. parameters:
  45. # Static parameter announcements are sent to the UI for _all_ Applications handled by this plugin.
  46. # Think of the `string`, `array`, and `map` values set here as "defaults". It is up to the plugin author to make
  47. # sure that these default values actually reflect the plugin's behavior if the user doesn't explicitly set different
  48. # values for those parameters.
  49. static:
  50. - name: string-param
  51. title: Description of the string param
  52. tooltip: Tooltip shown when the user hovers the
  53. # If this field is set, the UI will indicate to the user that they must set the value.
  54. required: false
  55. # itemType tells the UI how to present the parameter's value (or, for arrays and maps, values). Default is
  56. # "string". Examples of other types which may be supported in the future are "boolean" or "number".
  57. # Even if the itemType is not "string", the parameter value from the Application spec will be sent to the plugin
  58. # as a string. It's up to the plugin to do the appropriate conversion.
  59. itemType: ""
  60. # collectionType describes what type of value this parameter accepts (string, array, or map) and allows the UI
  61. # to present a form to match that type. Default is "string". This field must be present for non-string types.
  62. # It will not be inferred from the presence of an `array` or `map` field.
  63. collectionType: ""
  64. # This field communicates the parameter's default value to the UI. Setting this field is optional.
  65. string: default-string-value
  66. # All the fields above besides "string" apply to both the array and map type parameter announcements.
  67. - name: array-param
  68. # This field communicates the parameter's default value to the UI. Setting this field is optional.
  69. array: [default, items]
  70. collectionType: array
  71. - name: map-param
  72. # This field communicates the parameter's default value to the UI. Setting this field is optional.
  73. map:
  74. some: value
  75. collectionType: map
  76. # Dynamic parameter announcements are announcements specific to an Application handled by this plugin. For example,
  77. # the values for a Helm chart's values.yaml file could be sent as parameter announcements.
  78. dynamic:
  79. # The command is run in an Application's source directory. Standard output must be JSON matching the schema of the
  80. # static parameter announcements list.
  81. command: [echo, '[{"name": "example-param", "string": "default-string-value"}]']
  82. # If set to `true` then the plugin receives repository files with original file mode. Dangerous since the repository
  83. # might have executable files. Set to true only if you trust the CMP plugin authors.
  84. preserveFileMode: false

Note

While the ConfigManagementPlugin looks like a Kubernetes object, it is not actually a custom resource. It only follows kubernetes-style spec conventions.

The generate command must print a valid Kubernetes YAML or JSON object stream to stdout. Both init and generate commands are executed inside the application source directory.

The discover.fileName is used as glob pattern to determine whether an application repository is supported by the plugin or not.

  1. discover:
  2. find:
  3. command: [sh, -c, find . -name env.yaml]

If discover.fileName is not provided, the discover.find.command is executed in order to determine whether an application repository is supported by the plugin or not. The find command should return a non-error exit code and produce output to stdout when the application source type is supported.

Place the plugin configuration file in the sidecar

Argo CD expects the plugin configuration file to be located at /home/argocd/cmp-server/config/plugin.yaml in the sidecar.

If you use a custom image for the sidecar, you can add the file directly to that image.

  1. WORKDIR /home/argocd/cmp-server/config/
  2. COPY plugin.yaml ./

If you use a stock image for the sidecar or would rather maintain the plugin configuration in a ConfigMap, just nest the plugin config file in a ConfigMap under the plugin.yaml key and mount the ConfigMap in the sidecar (see next section).

  1. apiVersion: v1
  2. kind: ConfigMap
  3. metadata:
  4. name: my-plugin-config
  5. data:
  6. plugin.yaml: |
  7. apiVersion: argoproj.io/v1alpha1
  8. kind: ConfigManagementPlugin
  9. metadata:
  10. name: my-plugin
  11. spec:
  12. version: v1.0
  13. init:
  14. command: [sh, -c, 'echo "Initializing..."']
  15. generate:
  16. command: [sh, -c, 'echo "{\"kind\": \"ConfigMap\", \"apiVersion\": \"v1\", \"metadata\": { \"name\": \"$ARGOCD_APP_NAME\", \"namespace\": \"$ARGOCD_APP_NAMESPACE\", \"annotations\": {\"Foo\": \"$ARGOCD_ENV_FOO\", \"KubeVersion\": \"$KUBE_VERSION\", \"KubeApiVersion\": \"$KUBE_API_VERSIONS\",\"Bar\": \"baz\"}}}"']
  17. discover:
  18. fileName: "./subdir/s*.yaml"

Register the plugin sidecar

To install a plugin, patch argocd-repo-server to run the plugin container as a sidecar, with argocd-cmp-server as its entrypoint. You can use either off-the-shelf or custom-built plugin image as sidecar image. For example:

  1. containers:
  2. - name: my-plugin
  3. command: [/var/run/argocd/argocd-cmp-server] # Entrypoint should be Argo CD lightweight CMP server i.e. argocd-cmp-server
  4. image: ubuntu # This can be off-the-shelf or custom-built image
  5. securityContext:
  6. runAsNonRoot: true
  7. runAsUser: 999
  8. volumeMounts:
  9. - mountPath: /var/run/argocd
  10. name: var-files
  11. - mountPath: /home/argocd/cmp-server/plugins
  12. name: plugins
  13. # Remove this volumeMount if you've chosen to bake the config file into the sidecar image.
  14. - mountPath: /home/argocd/cmp-server/config/plugin.yaml
  15. subPath: plugin.yaml
  16. name: my-plugin-config
  17. # Starting with v2.4, do NOT mount the same tmp volume as the repo-server container. The filesystem separation helps
  18. # mitigate path traversal attacks.
  19. - mountPath: /tmp
  20. name: cmp-tmp
  21. volumes:
  22. - configMap:
  23. name: my-plugin-config
  24. name: my-plugin-config
  25. - emptyDir: {}
  26. name: cmp-tmp

Double-check these items

  1. Make sure to use /var/run/argocd/argocd-cmp-server as an entrypoint. The argocd-cmp-server is a lightweight GRPC service that allows Argo CD to interact with the plugin.
  2. Make sure that sidecar container is running as user 999.
  3. Make sure that plugin configuration file is present at /home/argocd/cmp-server/config/plugin.yaml. It can either be volume mapped via configmap or baked into image.

Using environment variables in your plugin

Plugin commands have access to

  1. The system environment variables of the sidecar
  2. Standard build environment variables
  3. Variables in the Application spec (References to system and build variables will get interpolated in the variables’ values):

    1. apiVersion: argoproj.io/v1alpha1
    2. kind: Application
    3. spec:
    4. source:
    5. plugin:
    6. env:
    7. - name: FOO
    8. value: bar
    9. - name: REV
    10. value: test-$ARGOCD_APP_REVISION

    Before reaching the init.command, generate.command, and discover.find.command commands, Argo CD prefixes all user-supplied environment variables (#3 above) with ARGOCD_ENV_. This prevents users from directly setting potentially-sensitive environment variables.

  4. Parameters in the Application spec:

    1. apiVersion: argoproj.io/v1alpha1
    2. kind: Application
    3. spec:
    4. source:
    5. plugin:
    6. parameters:
    7. - name: values-files
    8. array: [values-dev.yaml]
    9. - name: helm-parameters
    10. map:
    11. image.tag: v1.2.3

    The parameters are available as JSON in the ARGOCD_APP_PARAMETERS environment variable. The example above would produce this JSON:

    1. [{"name": "values-files", "array": ["values-dev.yaml"]}, {"name": "helm-parameters", "map": {"image.tag": "v1.2.3"}}]

    Note

    Parameter announcements, even if they specify defaults, are not sent to the plugin in ARGOCD_APP_PARAMETERS. Only parameters explicitly set in the Application spec are sent to the plugin. It is up to the plugin to apply the same defaults as the ones announced to the UI.

    The same parameters are also available as individual environment variables. The names of the environment variables follows this convention:

    1. - name: some-string-param
    2. string: some-string-value
    3. # PARAM_SOME_STRING_PARAM=some-string-value
    4. - name: some-array-param
    5. value: [item1, item2]
    6. # PARAM_SOME_ARRAY_PARAM_0=item1
    7. # PARAM_SOME_ARRAY_PARAM_1=item2
    8. - name: some-map-param
    9. map:
    10. image.tag: v1.2.3
    11. # PARAM_SOME_MAP_PARAM_IMAGE_TAG=v1.2.3

Sanitize/escape user input

As part of Argo CD’s manifest generation system, config management plugins are treated with a level of trust. Be sure to escape user input in your plugin to prevent malicious input from causing unwanted behavior.

Using a config management plugin with an Application

You may leave the name field empty in the plugin section for the plugin to be automatically matched with the Application based on its discovery rules. If you do mention the name make sure it is either <metadata.name>-<spec.version> if version is mentioned in the ConfigManagementPlugin spec or else just <metadata.name>. When name is explicitly specified only that particular plugin will be used iff its discovery pattern/command matches the provided application repo.

  1. apiVersion: argoproj.io/v1alpha1
  2. kind: Application
  3. metadata:
  4. name: guestbook
  5. namespace: argocd
  6. spec:
  7. project: default
  8. source:
  9. repoURL: https://github.com/argoproj/argocd-example-apps.git
  10. targetRevision: HEAD
  11. path: guestbook
  12. plugin:
  13. env:
  14. - name: FOO
  15. value: bar

If you don’t need to set any environment variables, you can set an empty plugin section.

  1. plugin: {}

Important

If your CMP command runs too long, the command will be killed, and the UI will show an error. The CMP server respects the timeouts set by the server.repo.server.timeout.seconds and controller.repo.server.timeout.seconds items in argocd-cm. Increase their values from the default of 60s.

Each CMP command will also independently timeout on the ARGOCD_EXEC_TIMEOUT set for the CMP sidecar. The default is 90s. So if you increase the repo server timeout greater than 90s, be sure to set ARGOCD_EXEC_TIMEOUT on the sidecar.

Note

Each Application can only have one config management plugin configured at a time. If you’re converting an existing plugin configured through the argocd-cm ConfigMap to a sidecar, make sure to update the plugin name to either <metadata.name>-<spec.version> if version was mentioned in the ConfigManagementPlugin spec or else just use <metadata.name>. You can also remove the name altogether and let the automatic discovery to identify the plugin.

Note

If a CMP renders blank manfiests, and prune is set to true, Argo CD will automatically remove resources. CMP plugin authors should ensure errors are part of the exit code. Commonly something like kustomize build . | cat won’t pass errors because of the pipe. Consider setting set -o pipefail so anything piped will pass errors on failure.

Debugging a CMP

If you are actively developing a sidecar-installed CMP, keep a few things in mind:

  1. If you are mounting plugin.yaml from a ConfigMap, you will have to restart the repo-server Pod so the plugin will pick up the changes.
  2. If you have baked plugin.yaml into your image, you will have to build, push, and force a re-pull of that image on the repo-server Pod so the plugin will pick up the changes. If you are using :latest, the Pod will always pull the new image. If you’re using a different, static tag, set imagePullPolicy: Always on the CMP’s sidecar container.
  3. CMP errors are cached by the repo-server in Redis. Restarting the repo-server Pod will not clear the cache. Always do a “Hard Refresh” when actively developing a CMP so you have the latest output.
  4. Verify your sidecar has started properly by viewing the Pod and seeing that two containers are running kubectl get pod -l app.kubernetes.io/component=repo-server -n argocd
  5. Write log message to stderr and set the --loglevel=info flag in the sidecar. This will print everything written to stderr, even on successfull command execution.

Other Common Errors

Error MessageCause
no matches for kind “ConfigManagementPlugin” in version “argoproj.io/v1alpha1”The ConfigManagementPlugin CRD was deprecated in Argo CD 2.4 and removed in 2.8. This error means you’ve tried to put the configuration for your plugin directly into Kubernetes as a CRD. Refer to this section of documentation for how to write the plugin configuration file and place it properly in the sidecar.

Plugin tar stream exclusions

In order to increase the speed of manifest generation, certain files and folders can be excluded from being sent to your plugin. We recommend excluding your .git folder if it isn’t necessary. Use Go’s filepatch.Match syntax. For example, .git/* to exclude .git folder.

You can set it one of three ways:

  1. The --plugin-tar-exclude argument on the repo server.
  2. The reposerver.plugin.tar.exclusions key if you are using argocd-cmd-params-cm
  3. Directly setting ARGOCD_REPO_SERVER_PLUGIN_TAR_EXCLUSIONS environment variable on the repo server.

For option 1, the flag can be repeated multiple times. For option 2 and 3, you can specify multiple globs by separating them with semicolons.

Migrating from argocd-cm plugins

Installing plugins by modifying the argocd-cm ConfigMap is deprecated as of v2.4 and has been completely removed starting in v2.8.

CMP plugins work by adding a sidecar to argocd-repo-server along with a configuration in that sidecar located at /home/argocd/cmp-server/config/plugin.yaml. A argocd-cm plugin can be easily converted with the following steps.

Convert the ConfigMap entry into a config file

First, copy the plugin’s configuration into its own YAML file. Take for example the following ConfigMap entry:

  1. data:
  2. configManagementPlugins: |
  3. - name: pluginName
  4. init: # Optional command to initialize application source directory
  5. command: ["sample command"]
  6. args: ["sample args"]
  7. generate: # Command to generate Kubernetes Objects in either YAML or JSON
  8. command: ["sample command"]
  9. args: ["sample args"]
  10. lockRepo: true # Defaults to false. See below.

The pluginName item would be converted to a config file like this:

  1. apiVersion: argoproj.io/v1alpha1
  2. kind: ConfigManagementPlugin
  3. metadata:
  4. name: pluginName
  5. spec:
  6. init: # Optional command to initialize application source directory
  7. command: ["sample command"]
  8. args: ["sample args"]
  9. generate: # Command to generate Kubernetes Objects in either YAML or JSON
  10. command: ["sample command"]
  11. args: ["sample args"]

Note

The lockRepo key is not relevant for sidecar plugins, because sidecar plugins do not share a single source repo directory when generating manifests.

Next, we need to decide how this yaml is going to be added to the sidecar. We can either bake the yaml directly into the image, or we can mount it from a ConfigMap.

If using a ConfigMap, our example would look like this:

  1. apiVersion: v1
  2. kind: ConfigMap
  3. metadata:
  4. name: pluginName
  5. namespace: argocd
  6. data:
  7. pluginName.yaml: |
  8. apiVersion: argoproj.io/v1alpha1
  9. kind: ConfigManagementPlugin
  10. metadata:
  11. name: pluginName
  12. spec:
  13. init: # Optional command to initialize application source directory
  14. command: ["sample command"]
  15. args: ["sample args"]
  16. generate: # Command to generate Kubernetes Objects in either YAML or JSON
  17. command: ["sample command"]
  18. args: ["sample args"]

Then this would be mounted in our plugin sidecar.

Write discovery rules for your plugin

Sidecar plugins can use either discovery rules or a plugin name to match Applications to plugins. If the discovery rule is omitted then you have to explicitly specify the plugin by name in the app spec or else that particular plugin will not match any app.

If you want to use discovery instead of the plugin name to match applications to your plugin, write rules applicable to your plugin using the instructions above and add them to your configuration file.

To use the name instead of discovery, update the name in your application manifest to <metadata.name>-<spec.version> if version was mentioned in the ConfigManagementPlugin spec or else just use <metadata.name>. For example:

  1. apiVersion: argoproj.io/v1alpha1
  2. kind: Application
  3. metadata:
  4. name: guestbook
  5. spec:
  6. source:
  7. plugin:
  8. name: pluginName # Delete this for auto-discovery (and set `plugin: {}` if `name` was the only value) or use proper sidecar plugin name

Make sure the plugin has access to the tools it needs

Plugins configured with argocd-cm ran on the Argo CD image. This gave it access to all the tools installed on that image by default (see the Dockerfile for base image and installed tools).

You can either use a stock image (like ubuntu, busybox, or alpine/k8s) or design your own base image with the tools your plugin needs. For security, avoid using images with more binaries installed than what your plugin actually needs.

Test the plugin

After installing the plugin as a sidecar according to the directions above, test it out on a few Applications before migrating all of them to the sidecar plugin.

Once tests have checked out, remove the plugin entry from your argocd-cm ConfigMap.

Additional Settings

Preserve repository files mode

By default, config management plugin receives source repository files with reset file mode. This is done for security reasons. If you want to preserve original file mode, you can set preserveFileMode to true in the plugin spec:

Warning

Make sure you trust the plugin you are using. If you set preserveFileMode to true then the plugin might receive files with executable permissions which can be a security risk.

  1. apiVersion: argoproj.io/v1alpha1
  2. kind: ConfigManagementPlugin
  3. metadata:
  4. name: pluginName
  5. spec:
  6. init:
  7. command: ["sample command"]
  8. args: ["sample args"]
  9. generate:
  10. command: ["sample command"]
  11. args: ["sample args"]
  12. preserveFileMode: true