kubernetes

Description

The kubernetes provider allows you to deploy container modules to Kubernetes clusters, and adds the helm and kubernetes module types.

For usage information, please refer to the guides section. A good place to start is the Remote Kubernetes guide guide if you're connecting to remote clusters. The Quickstart guide guide is also helpful as an introduction.

Note that if you're using a local Kubernetes cluster (e.g. minikube or Docker Desktop), the local-kubernetes provider simplifies (and automates) the configuration and setup quite a bit.

Below is the full schema reference for the provider configuration. For an introduction to configuring a Garden project with providers, please look at our configuration guide.

The reference is divided into two sections. The first section contains the complete YAML schema, and the second section describes each schema key.

Complete YAML Schema

The values in the schema below are the default values.

providers:
  - # List other providers that should be resolved before this one.
    dependencies: []

    # If specified, this provider will only be used in the listed environments. Note that an empty array effectively
    # disables the provider. To use a provider in all environments, omit this field.
    environments:

    # Choose the mechanism for building container images before deploying. By default your local Docker daemon is
    # used, but you can set it to `cluster-buildkit` or `kaniko` to sync files to the cluster, and build container
    # images there. This removes the need to run Docker locally, and allows you to share layer and image caches
    # between multiple developers, as well as between your development and CI workflows.
    #
    # For more details on all the different options and what makes sense to use for your setup, please check out the
    # [in-cluster building
    # guide](https://docs.garden.io/v/acorn-0.12/kubernetes-plugins/advanced/in-cluster-building).
    #
    # **Note:** The `cluster-docker` mode has been deprecated and will be removed in a future release!
    buildMode: local-docker

    # Configuration options for the `cluster-buildkit` build mode.
    clusterBuildkit: {}
      # Use the `cache` configuration to customize the default cluster-buildkit cache behaviour.
      #
      # The default value is:
      # clusterBuildkit:
      #   cache:
      #     - type: registry
      #       mode: auto
      #
      # For every build, this will
      # - import cached layers from a docker image tag named `_buildcache`
      # - when the build is finished, upload cache information to `_buildcache`
      #
      # For registries that support it, `mode: auto` (the default) will enable the buildkit `mode=max`
      # option.
      #
      # See the following table for details on our detection mechanism:
      #
      # | Registry Name                   | Registry Domain         | Assumed `mode=max` support |
      # |---------------------------------|-------------------------|------------------------------|
      # | Google Cloud Artifact Registry  | `pkg.dev`             | Yes                          |
      # | Azure Container Registry        | `azurecr.io`          | Yes                          |
      # | GitHub Container Registry       | `ghcr.io`             | Yes                          |
      # | DockerHub                       | `hub.docker.com`     | Yes                          |
      # | Garden In-Cluster Registry      |                         | Yes                          |
      # | Any other registry              |                         | No                           |
      #
      # In case you need to override the defaults for your registry, you can do it like so:
      #
      # clusterBuildkit:
      #   cache:
      #     - type: registry
      #       mode: max
      #
      # When you add multiple caches, we will make sure to pass the `--import-cache` options to buildkit in the same
      # order as provided in the cache configuration. This is because buildkit will not actually use all imported
      # caches
      # for every build, but it will stick with the first cache that yields a cache hit for all the following layers.
      #
      # An example for this is the following:
      #
      # clusterBuildkit:
      #   cache:
      #     - type: registry
      #       tag: _buildcache-${slice(kebabCase(git.branch), "0", "30")}
      #     - type: registry
      #       tag: _buildcache-main
      #       export: false
      #
      # Using this cache configuration, every build will first look for a cache specific to your feature branch.
      # If it does not exist yet, it will import caches from the main branch builds (`_buildcache-main`).
      # When the build is finished, it will only export caches to your feature branch, and avoid polluting the `main`
      # branch caches.
      # A configuration like that may improve your cache hit rate and thus save time.
      #
      # If you need to disable caches completely you can achieve that with the following configuration:
      #
      # clusterBuildkit:
      #   cache: []
      cache:
        - # Use the Docker registry configured at `deploymentRegistry` to retrieve and store buildkit cache
          # information.
          #
          # See also the [buildkit registry cache
          # documentation](https://github.com/moby/buildkit#registry-push-image-and-cache-separately)
          type:

          # The registry from which the cache should be imported from, or which it should be exported to.
          #
          # If not specified, use the configured `deploymentRegistry` in your kubernetes provider config, or the
          # internal in-cluster registry in case `deploymentRegistry` is not set.
          #
          # Important: You must make sure `imagePullSecrets` includes authentication with the specified cache
          # registry, that has the appropriate write privileges (usually full write access to the configured
          # `namespace`).
          registry:
            # The hostname (and optionally port, if not the default port) of the registry.
            hostname:

            # The port where the registry listens on, if not the default.
            port:

            # The registry namespace. Will be placed between hostname and image name, like so:
            # <hostname>/<namespace>/<image name>
            namespace: _

            # Set to true to allow insecure connections to the registry (without SSL).
            insecure: false

          # This is the buildkit cache mode to be used.
          #
          # The value `inline` ensures that garden is using the buildkit option `--export-cache inline`. Cache
          # information will be inlined and co-located with the Docker image itself.
          #
          # The values `min` and `max` ensure that garden passes the `mode=max` or `mode=min` modifiers to the
          # buildkit `--export-cache` option. Cache manifests will only be
          # stored stored in the configured `tag`.
          #
          # `auto` is the same as `max` for some registries that are known to support it. Garden will fall back to
          # `inline` for all other registries.
          #  See the [clusterBuildkit cache option](#providers-.clusterbuildkit.cache) for a description of the
          # detection mechanism.
          #
          # See also the [buildkit export cache documentation](https://github.com/moby/buildkit#export-cache)
          mode: auto

          # This is the Docker registry tag name buildkit should use for the registry build cache. Default is
          # `_buildcache`
          #
          # **NOTE**: `tag` can only be used together with the `registry` cache type
          tag: _buildcache

          # If this is false, only pass the `--import-cache` option to buildkit, and not the `--export-cache` option.
          # Defaults to true.
          export: true

      # Enable rootless mode for the cluster-buildkit daemon, which runs the daemon with decreased privileges.
      # Please see [the buildkit docs](https://github.com/moby/buildkit/blob/master/docs/rootless.md) for caveats when
      # using this mode.
      rootless: false

      # Exposes the `nodeSelector` field on the PodSpec of the BuildKit deployment. This allows you to constrain the
      # BuildKit daemon to only run on particular nodes.
      #
      # [See here](https://kubernetes.io/docs/concepts/configuration/assign-pod-node/) for the official Kubernetes
      # guide to assigning Pods to nodes.
      nodeSelector: {}

      # Specify tolerations to apply to cluster-buildkit daemon. Useful to control which nodes in a cluster can run
      # builds.
      tolerations:
        - # "Effect" indicates the taint effect to match. Empty means match all taint effects. When specified,
          # allowed values are "NoSchedule", "PreferNoSchedule" and "NoExecute".
          effect:

          # "Key" is the taint key that the toleration applies to. Empty means match all taint keys.
          # If the key is empty, operator must be "Exists"; this combination means to match all values and all keys.
          key:

          # "Operator" represents a key's relationship to the value. Valid operators are "Exists" and "Equal".
          # Defaults to
          # "Equal". "Exists" is equivalent to wildcard for value, so that a pod can tolerate all taints of a
          # particular category.
          operator: Equal

          # "TolerationSeconds" represents the period of time the toleration (which must be of effect "NoExecute",
          # otherwise this field is ignored) tolerates the taint. By default, it is not set, which means tolerate
          # the taint forever (do not evict). Zero and negative values will be treated as 0 (evict immediately)
          # by the system.
          tolerationSeconds:

          # "Value" is the taint value the toleration matches to. If the operator is "Exists", the value should be
          # empty,
          # otherwise just a regular string.
          value:

      # Specify annotations to apply to both the Pod and Deployment resources associated with cluster-buildkit.
      # Annotations may have an effect on the behaviour of certain components, for example autoscalers.
      annotations:

    # Setting related to Jib image builds.
    jib:
      # In some cases you may need to push images built with Jib to the remote registry via Kubernetes cluster, e.g.
      # if you don't have connectivity or access from where Garden is being run. In that case, set this flag to true,
      # but do note that the build will take considerably take longer to complete! Only applies when using in-cluster
      # building.
      pushViaCluster: false

    # Configuration options for the `kaniko` build mode.
    kaniko:
      # Specify extra flags to use when building the container image with kaniko. Flags set on `container` modules
      # take precedence over these.
      extraFlags:

      # Change the kaniko image (repository/image:tag) to use when building in kaniko mode.
      image: 'gcr.io/kaniko-project/executor:v1.11.0-debug'

      # Choose the namespace where the Kaniko pods will be run. Set to `null` to use the project namespace.
      #
      # **IMPORTANT: The default namespace will change to the project namespace instead of the garden-system namespace
      # in an upcoming release!**
      namespace: garden-system

      # Exposes the `nodeSelector` field on the PodSpec of the Kaniko pods. This allows you to constrain the Kaniko
      # pods to only run on particular nodes. The same nodeSelector will be used for each util pod unless they are
      # specifically set under `util.nodeSelector`.
      #
      # [See here](https://kubernetes.io/docs/concepts/configuration/assign-pod-node/) for the official Kubernetes
      # guide to assigning pods to nodes.
      nodeSelector:

      # Specify tolerations to apply to each Kaniko builder pod. Useful to control which nodes in a cluster can run
      # builds. The same tolerations will be used for each util pod unless they are specifically set under
      # `util.tolerations`
      tolerations:
        - # "Effect" indicates the taint effect to match. Empty means match all taint effects. When specified,
          # allowed values are "NoSchedule", "PreferNoSchedule" and "NoExecute".
          effect:

          # "Key" is the taint key that the toleration applies to. Empty means match all taint keys.
          # If the key is empty, operator must be "Exists"; this combination means to match all values and all keys.
          key:

          # "Operator" represents a key's relationship to the value. Valid operators are "Exists" and "Equal".
          # Defaults to
          # "Equal". "Exists" is equivalent to wildcard for value, so that a pod can tolerate all taints of a
          # particular category.
          operator: Equal

          # "TolerationSeconds" represents the period of time the toleration (which must be of effect "NoExecute",
          # otherwise this field is ignored) tolerates the taint. By default, it is not set, which means tolerate
          # the taint forever (do not evict). Zero and negative values will be treated as 0 (evict immediately)
          # by the system.
          tolerationSeconds:

          # "Value" is the taint value the toleration matches to. If the operator is "Exists", the value should be
          # empty,
          # otherwise just a regular string.
          value:

      # Specify annotations to apply to each Kaniko builder pod. Annotations may have an effect on the behaviour of
      # certain components, for example autoscalers. The same annotations will be used for each util pod unless they
      # are specifically set under `util.annotations`
      annotations:

      util:
        # Specify tolerations to apply to each garden-util pod.
        tolerations:
          - # "Effect" indicates the taint effect to match. Empty means match all taint effects. When specified,
            # allowed values are "NoSchedule", "PreferNoSchedule" and "NoExecute".
            effect:

            # "Key" is the taint key that the toleration applies to. Empty means match all taint keys.
            # If the key is empty, operator must be "Exists"; this combination means to match all values and all keys.
            key:

            # "Operator" represents a key's relationship to the value. Valid operators are "Exists" and "Equal".
            # Defaults to
            # "Equal". "Exists" is equivalent to wildcard for value, so that a pod can tolerate all taints of a
            # particular category.
            operator: Equal

            # "TolerationSeconds" represents the period of time the toleration (which must be of effect "NoExecute",
            # otherwise this field is ignored) tolerates the taint. By default, it is not set, which means tolerate
            # the taint forever (do not evict). Zero and negative values will be treated as 0 (evict immediately)
            # by the system.
            tolerationSeconds:

            # "Value" is the taint value the toleration matches to. If the operator is "Exists", the value should be
            # empty,
            # otherwise just a regular string.
            value:

        # Specify annotations to apply to each garden-util pod and deployments.
        annotations:

        # Specify the nodeSelector constraints for each garden-util pod.
        nodeSelector:

    # A default hostname to use when no hostname is explicitly configured for a service.
    defaultHostname:

    # Sets the deployment strategy for `container` services.
    #
    # The default is `"rolling"`, which performs rolling updates. There is also experimental support for blue/green
    # deployments (via the `"blue-green"` strategy).
    #
    # Note that this setting only applies to `container` services (and not, for example,  `kubernetes` or `helm`
    # services).
    deploymentStrategy: rolling

    # Configuration options for dev mode.
    devMode:
      # Specifies default settings for dev mode syncs (e.g. for `container`, `kubernetes` and `helm` services).
      #
      # These are overridden/extended by the settings of any individual dev mode sync specs for a given module or
      # service.
      #
      # Dev mode is enabled when running the `garden dev` command, and by setting the `--dev` flag on the `garden
      # deploy` command.
      #
      # See the [Code Synchronization guide](https://docs.garden.io/v/acorn-0.12/guides/code-synchronization-dev-mode)
      # for more information.
      defaults:
        # Specify a list of POSIX-style paths or glob patterns that should be excluded from the sync.
        #
        # Any exclusion patterns defined in individual dev mode sync specs will be applied in addition to these
        # patterns.
        #
        # `.git` directories and `.garden` directories are always ignored.
        exclude:

        # The default permission bits, specified as an octal, to set on files at the sync target. Defaults to 0600
        # (user read/write). See the [Mutagen
        # docs](https://mutagen.io/documentation/synchronization/permissions#permissions) for more information.
        fileMode:

        # The default permission bits, specified as an octal, to set on directories at the sync target. Defaults to
        # 0700 (user read/write). See the [Mutagen
        # docs](https://mutagen.io/documentation/synchronization/permissions#permissions) for more information.
        directoryMode:

        # Set the default owner of files and directories at the target. Specify either an integer ID or a string name.
        # See the [Mutagen docs](https://mutagen.io/documentation/synchronization/permissions#owners-and-groups) for
        # more information.
        owner:

        # Set the default group on files and directories at the target. Specify either an integer ID or a string name.
        # See the [Mutagen docs](https://mutagen.io/documentation/synchronization/permissions#owners-and-groups) for
        # more information.
        group:

    # Require SSL on all `container` module services. If set to true, an error is raised when no certificate is
    # available for a configured hostname on a `container` module.
    forceSsl: false

    # References to `docker-registry` secrets to use for authenticating with remote registries when pulling
    # images. This is necessary if you reference private images in your module configuration, and is required
    # when configuring a remote Kubernetes environment with buildMode=local.
    imagePullSecrets:
      - # The name of the Kubernetes secret.
        name:

        # The namespace where the secret is stored. If necessary, the secret may be copied to the appropriate
        # namespace before use.
        namespace: default

    # References to secrets you need to have copied into all namespaces deployed to. These secrets will be
    # ensured to exist in the namespace before deploying any service.
    copySecrets:
      - # The name of the Kubernetes secret.
        name:

        # The namespace where the secret is stored. If necessary, the secret may be copied to the appropriate
        # namespace before use.
        namespace: default

    # Resource requests and limits for the in-cluster builder, container registry and code sync service. (which are
    # automatically installed and used when `buildMode` is `cluster-docker` or `kaniko`).
    resources:
      # Resource requests and limits for the in-cluster builder. It's important to consider which build mode you're
      # using when configuring this.
      #
      # When `buildMode` is `kaniko`, this refers to _each Kaniko pod_, i.e. each individual build, so you'll want to
      # consider the requirements for your individual image builds, with your most expensive/heavy images in mind.
      #
      # When `buildMode` is `cluster-buildkit`, this applies to the BuildKit deployment created in _each project
      # namespace_. So think of this as the resource spec for each individual user or project namespace.
      #
      # When `buildMode` is `cluster-docker`, this applies to the single Docker Daemon that is installed and run
      # cluster-wide. This is shared across all users and builds in the cluster, so it should be resourced
      # accordingly, factoring in how many concurrent builds you expect and how heavy your builds tend to be. **Note
      # that the cluster-docker build mode has been deprecated!**
      builder:
        limits:
          # CPU limit in millicpu.
          cpu: 4000

          # Memory limit in megabytes.
          memory: 8192

          # Ephemeral storage limit in megabytes.
          ephemeralStorage:

        requests:
          # CPU request in millicpu.
          cpu: 100

          # Memory request in megabytes.
          memory: 512

          # Ephemeral storage request in megabytes.
          ephemeralStorage:

      # Resource requests and limits for the in-cluster image registry. Built images are pushed to this registry,
      # so that they are available to all the nodes in your cluster.
      #
      # This is shared across all users and builds, so it should be resourced accordingly, factoring
      # in how many concurrent builds you expect and how large your images tend to be.
      registry:
        limits:
          # CPU limit in millicpu.
          cpu: 2000

          # Memory limit in megabytes.
          memory: 4096

          # Ephemeral storage limit in megabytes.
          ephemeralStorage:

        requests:
          # CPU request in millicpu.
          cpu: 200

          # Memory request in megabytes.
          memory: 512

          # Ephemeral storage request in megabytes.
          ephemeralStorage:

      # Resource requests and limits for the util pod for in-cluster builders.
      # This pod is used to get, start, stop and inquire the status of the builds.
      #
      # This pod is created in each garden namespace.
      util:
        limits:
          # CPU limit in millicpu.
          cpu: 256

          # Memory limit in megabytes.
          memory: 512

          # Ephemeral storage limit in megabytes.
          ephemeralStorage:

        requests:
          # CPU request in millicpu.
          cpu: 256

          # Memory request in megabytes.
          memory: 512

          # Ephemeral storage request in megabytes.
          ephemeralStorage:

    # Storage parameters to set for the in-cluster builder, container registry and code sync persistent volumes
    # (which are automatically installed and used when `buildMode` is `cluster-docker` or `kaniko`).
    #
    # These are all shared cluster-wide across all users and builds, so they should be resourced accordingly,
    # factoring in how many concurrent builds you expect and how large your images and build contexts tend to be.
    storage:
      # Storage parameters for the in-cluster Docker registry volume. Built images are stored here, so that they
      # are available to all the nodes in your cluster.
      #
      # Only applies when `buildMode` is set to `cluster-docker` or `kaniko`, ignored otherwise.
      registry:
        # Volume size in megabytes.
        size: 20480

        # Storage class to use for the volume.
        storageClass: null

    # One or more certificates to use for ingress.
    tlsCertificates:
      - # A unique identifier for this certificate.
        name:

        # A list of hostnames that this certificate should be used for. If you don't specify these, they will be
        # automatically read from the certificate.
        hostnames:

        # A reference to the Kubernetes secret that contains the TLS certificate and key for the domain.
        secretRef:
          # The name of the Kubernetes secret.
          name:

          # The namespace where the secret is stored. If necessary, the secret may be copied to the appropriate
          # namespace before use.
          namespace: default

    # Exposes the `nodeSelector` field on the PodSpec of system services. This allows you to constrain the system
    # services to only run on particular nodes.
    #
    # [See here](https://kubernetes.io/docs/concepts/configuration/assign-pod-node/) for the official Kubernetes guide
    # to assigning Pods to nodes.
    systemNodeSelector: {}

    # For setting tolerations on the registry-proxy when using in-cluster building.
    # The registry-proxy is a DaemonSet that proxies connections to the docker registry service on each node.
    #
    # Use this only if you're doing in-cluster building and the nodes in your cluster
    # have [taints](https://kubernetes.io/docs/concepts/configuration/taint-and-toleration/).
    registryProxyTolerations:
      - # "Effect" indicates the taint effect to match. Empty means match all taint effects. When specified,
        # allowed values are "NoSchedule", "PreferNoSchedule" and "NoExecute".
        effect:

        # "Key" is the taint key that the toleration applies to. Empty means match all taint keys.
        # If the key is empty, operator must be "Exists"; this combination means to match all values and all keys.
        key:

        # "Operator" represents a key's relationship to the value. Valid operators are "Exists" and "Equal". Defaults
        # to
        # "Equal". "Exists" is equivalent to wildcard for value, so that a pod can tolerate all taints of a
        # particular category.
        operator: Equal

        # "TolerationSeconds" represents the period of time the toleration (which must be of effect "NoExecute",
        # otherwise this field is ignored) tolerates the taint. By default, it is not set, which means tolerate
        # the taint forever (do not evict). Zero and negative values will be treated as 0 (evict immediately)
        # by the system.
        tolerationSeconds:

        # "Value" is the taint value the toleration matches to. If the operator is "Exists", the value should be
        # empty,
        # otherwise just a regular string.
        value:

    # The name of the provider plugin to use.
    name: kubernetes

    # The kubectl context to use to connect to the Kubernetes cluster.
    context:

    # The registry where built containers should be pushed to, and then pulled to the cluster when deploying services.
    #
    # Important: If you specify this in combination with in-cluster building, you must make sure `imagePullSecrets`
    # includes authentication with the specified deployment registry, that has the appropriate write privileges
    # (usually full write access to the configured `deploymentRegistry.namespace`).
    deploymentRegistry:
      # The hostname (and optionally port, if not the default port) of the registry.
      hostname:

      # The port where the registry listens on, if not the default.
      port:

      # The registry namespace. Will be placed between hostname and image name, like so: <hostname>/<namespace>/<image
      # name>
      namespace: _

      # Set to true to allow insecure connections to the registry (without SSL).
      insecure: false

    # The ingress class to use on configured Ingresses (via the `kubernetes.io/ingress.class` annotation)
    # when deploying `container` services. Use this if you have multiple ingress controllers in your cluster.
    ingressClass:

    # The external HTTP port of the cluster's ingress controller.
    ingressHttpPort: 80

    # The external HTTPS port of the cluster's ingress controller.
    ingressHttpsPort: 443

    # Path to kubeconfig file to use instead of the system default.
    kubeconfig:

    # Set a specific path to a kubectl binary, instead of having Garden download it automatically as required.
    #
    # It may be useful in some scenarios to allow individual users to set this, e.g. with an environment variable. You
    # could configure that with something like `kubectlPath: ${local.env.GARDEN_KUBECTL_PATH}?`.
    #
    # **Warning**: Garden may make some assumptions with respect to the kubectl version, so it is suggested to only
    # use this when necessary.
    kubectlPath:

    # Specify which namespace to deploy services to, and optionally annotations/labels to apply to the namespace.
    #
    # You can specify a string as a shorthand for `name: <name>`. Defaults to `<project name>-<environment
    # namespace>`.
    #
    # Note that the framework may generate other namespaces as well with this name as a prefix. Also note that if the
    # namespace previously exists, Garden will attempt to add the specified labels and annotations. If the user does
    # not have permissions to do so, a warning is shown.
    namespace:
      # A valid Kubernetes namespace name. Must be a valid RFC1035/RFC1123 (DNS) label (may contain lowercase letters,
      # numbers and dashes, must start with a letter, and cannot end with a dash) and must not be longer than 63
      # characters.
      name:

      # Map of annotations to apply to the namespace when creating it.
      annotations:

      # Map of labels to apply to the namespace when creating it.
      labels:

    # Set this to `nginx` to install/enable the NGINX ingress controller.
    setupIngressController: false

Configuration Keys

providers[]

TypeDefaultRequired

array[object]

[]

No

providers[].dependencies[]

providers > dependencies

List other providers that should be resolved before this one.

TypeDefaultRequired

array[string]

[]

No

Example:

providers:
  - dependencies:
      - exec

providers[].environments[]

providers > environments

If specified, this provider will only be used in the listed environments. Note that an empty array effectively disables the provider. To use a provider in all environments, omit this field.

TypeRequired

array[string]

No

Example:

providers:
  - environments:
      - dev
      - stage

providers[].buildMode

providers > buildMode

Choose the mechanism for building container images before deploying. By default your local Docker daemon is used, but you can set it to cluster-buildkit or kaniko to sync files to the cluster, and build container images there. This removes the need to run Docker locally, and allows you to share layer and image caches between multiple developers, as well as between your development and CI workflows.

For more details on all the different options and what makes sense to use for your setup, please check out the in-cluster building guide.

Note: The cluster-docker mode has been deprecated and will be removed in a future release!

TypeAllowed ValuesDefaultRequired

string

"local-docker", "cluster-docker", "kaniko", "cluster-buildkit"

"local-docker"

Yes

providers[].clusterBuildkit

providers > clusterBuildkit

Configuration options for the cluster-buildkit build mode.

TypeDefaultRequired

object

{}

No

providers[].clusterBuildkit.cache[]

providers > clusterBuildkit > cache

Use the cache configuration to customize the default cluster-buildkit cache behaviour.

The default value is:

clusterBuildkit:
  cache:
    - type: registry
      mode: auto

For every build, this will

  • import cached layers from a docker image tag named _buildcache

  • when the build is finished, upload cache information to _buildcache

For registries that support it, mode: auto (the default) will enable the buildkit mode=max option.

See the following table for details on our detection mechanism:

Registry Name

Registry Domain

Assumed mode=max support

Google Cloud Artifact Registry

pkg.dev

Yes

Azure Container Registry

azurecr.io

Yes

GitHub Container Registry

ghcr.io

Yes

DockerHub

hub.docker.com

Yes

Garden In-Cluster Registry

Yes

Any other registry

No

In case you need to override the defaults for your registry, you can do it like so:

clusterBuildkit:
  cache:
    - type: registry
      mode: max

When you add multiple caches, we will make sure to pass the --import-cache options to buildkit in the same order as provided in the cache configuration. This is because buildkit will not actually use all imported caches for every build, but it will stick with the first cache that yields a cache hit for all the following layers.

An example for this is the following:

clusterBuildkit:
  cache:
    - type: registry
      tag: _buildcache-${slice(kebabCase(git.branch), "0", "30")}
    - type: registry
      tag: _buildcache-main
      export: false

Using this cache configuration, every build will first look for a cache specific to your feature branch. If it does not exist yet, it will import caches from the main branch builds (_buildcache-main). When the build is finished, it will only export caches to your feature branch, and avoid polluting the main branch caches. A configuration like that may improve your cache hit rate and thus save time.

If you need to disable caches completely you can achieve that with the following configuration:

clusterBuildkit:
  cache: []
TypeDefaultRequired

array[object]

[{"type":"registry","mode":"auto","tag":"_buildcache","export":true}]

No

providers[].clusterBuildkit.cache[].type

providers > clusterBuildkit > cache > type

Use the Docker registry configured at deploymentRegistry to retrieve and store buildkit cache information.

See also the buildkit registry cache documentation

TypeAllowed ValuesRequired

string

"registry"

Yes

providers[].clusterBuildkit.cache[].registry

providers > clusterBuildkit > cache > registry

The registry from which the cache should be imported from, or which it should be exported to.

If not specified, use the configured deploymentRegistry in your kubernetes provider config, or the internal in-cluster registry in case deploymentRegistry is not set.

Important: You must make sure imagePullSecrets includes authentication with the specified cache registry, that has the appropriate write privileges (usually full write access to the configured namespace).

TypeRequired

object

No

providers[].clusterBuildkit.cache[].registry.hostname

providers > clusterBuildkit > cache > registry > hostname

The hostname (and optionally port, if not the default port) of the registry.

TypeRequired

string

Yes

Example:

providers:
  - clusterBuildkit:
      ...
      cache:
        - registry:
            ...
            hostname: "gcr.io"

providers[].clusterBuildkit.cache[].registry.port

providers > clusterBuildkit > cache > registry > port

The port where the registry listens on, if not the default.

TypeRequired

number

No

providers[].clusterBuildkit.cache[].registry.namespace

providers > clusterBuildkit > cache > registry > namespace

The registry namespace. Will be placed between hostname and image name, like so: //

TypeDefaultRequired

string

"_"

No

Example:

providers:
  - clusterBuildkit:
      ...
      cache:
        - registry:
            ...
            namespace: "my-project"

providers[].clusterBuildkit.cache[].registry.insecure

providers > clusterBuildkit > cache > registry > insecure

Set to true to allow insecure connections to the registry (without SSL).

TypeDefaultRequired

boolean

false

No

providers[].clusterBuildkit.cache[].mode

providers > clusterBuildkit > cache > mode

This is the buildkit cache mode to be used.

The value inline ensures that garden is using the buildkit option --export-cache inline. Cache information will be inlined and co-located with the Docker image itself.

The values min and max ensure that garden passes the mode=max or mode=min modifiers to the buildkit --export-cache option. Cache manifests will only be stored stored in the configured tag.

auto is the same as max for some registries that are known to support it. Garden will fall back to inline for all other registries. See the clusterBuildkit cache option for a description of the detection mechanism.

See also the buildkit export cache documentation

TypeAllowed ValuesDefaultRequired

string

"auto", "min", "max", "inline"

"auto"

Yes

providers[].clusterBuildkit.cache[].tag

providers > clusterBuildkit > cache > tag

This is the Docker registry tag name buildkit should use for the registry build cache. Default is _buildcache

NOTE: tag can only be used together with the registry cache type

TypeDefaultRequired

string

"_buildcache"

No

providers[].clusterBuildkit.cache[].export

providers > clusterBuildkit > cache > export

If this is false, only pass the --import-cache option to buildkit, and not the --export-cache option. Defaults to true.

TypeDefaultRequired

boolean

true

No

providers[].clusterBuildkit.rootless

providers > clusterBuildkit > rootless

Enable rootless mode for the cluster-buildkit daemon, which runs the daemon with decreased privileges. Please see the buildkit docs for caveats when using this mode.

TypeDefaultRequired

boolean

false

No

providers[].clusterBuildkit.nodeSelector

providers > clusterBuildkit > nodeSelector

Exposes the nodeSelector field on the PodSpec of the BuildKit deployment. This allows you to constrain the BuildKit daemon to only run on particular nodes.

See here for the official Kubernetes guide to assigning Pods to nodes.

TypeDefaultRequired

object

{}

No

Example:

providers:
  - clusterBuildkit:
      ...
      nodeSelector:
          disktype: ssd

providers[].clusterBuildkit.tolerations[]

providers > clusterBuildkit > tolerations

Specify tolerations to apply to cluster-buildkit daemon. Useful to control which nodes in a cluster can run builds.

TypeDefaultRequired

array[object]

[]

No

providers[].clusterBuildkit.tolerations[].effect

providers > clusterBuildkit > tolerations > effect

"Effect" indicates the taint effect to match. Empty means match all taint effects. When specified, allowed values are "NoSchedule", "PreferNoSchedule" and "NoExecute".

TypeRequired

string

No

providers[].clusterBuildkit.tolerations[].key

providers > clusterBuildkit > tolerations > key

"Key" is the taint key that the toleration applies to. Empty means match all taint keys. If the key is empty, operator must be "Exists"; this combination means to match all values and all keys.

TypeRequired

string

No

providers[].clusterBuildkit.tolerations[].operator

providers > clusterBuildkit > tolerations > operator

"Operator" represents a key's relationship to the value. Valid operators are "Exists" and "Equal". Defaults to "Equal". "Exists" is equivalent to wildcard for value, so that a pod can tolerate all taints of a particular category.

TypeDefaultRequired

string

"Equal"

No

providers[].clusterBuildkit.tolerations[].tolerationSeconds

providers > clusterBuildkit > tolerations > tolerationSeconds

"TolerationSeconds" represents the period of time the toleration (which must be of effect "NoExecute", otherwise this field is ignored) tolerates the taint. By default, it is not set, which means tolerate the taint forever (do not evict). Zero and negative values will be treated as 0 (evict immediately) by the system.

TypeRequired

string

No

providers[].clusterBuildkit.tolerations[].value

providers > clusterBuildkit > tolerations > value

"Value" is the taint value the toleration matches to. If the operator is "Exists", the value should be empty, otherwise just a regular string.

TypeRequired

string

No

providers[].clusterBuildkit.annotations

providers > clusterBuildkit > annotations

Specify annotations to apply to both the Pod and Deployment resources associated with cluster-buildkit. Annotations may have an effect on the behaviour of certain components, for example autoscalers.

TypeRequired

object

No

Example:

providers:
  - clusterBuildkit:
      ...
      annotations:
          cluster-autoscaler.kubernetes.io/safe-to-evict: 'false'

providers[].clusterDocker

providers > clusterDocker

Deprecated: This field will be removed in a future release.

Configuration options for the cluster-docker build mode.

TypeDefaultRequired

object

{}

No

providers[].clusterDocker.enableBuildKit

providers > clusterDocker > enableBuildKit

Deprecated: This field will be removed in a future release.

Enable BuildKit support. This should in most cases work well and be more performant, but we're opting to keep it optional until it's enabled by default in Docker.

TypeDefaultRequired

boolean

false

No

providers[].jib

providers > jib

Setting related to Jib image builds.

TypeRequired

object

No

providers[].jib.pushViaCluster

providers > jib > pushViaCluster

In some cases you may need to push images built with Jib to the remote registry via Kubernetes cluster, e.g. if you don't have connectivity or access from where Garden is being run. In that case, set this flag to true, but do note that the build will take considerably take longer to complete! Only applies when using in-cluster building.

TypeDefaultRequired

boolean

false

No

providers[].kaniko

providers > kaniko

Configuration options for the kaniko build mode.

TypeRequired

object

No

providers[].kaniko.extraFlags[]

providers > kaniko > extraFlags

Specify extra flags to use when building the container image with kaniko. Flags set on container modules take precedence over these.

TypeRequired

array[string]

No

providers[].kaniko.image

providers > kaniko > image

Change the kaniko image (repository/image:tag) to use when building in kaniko mode.

TypeDefaultRequired

string

"gcr.io/kaniko-project/executor:v1.11.0-debug"

No

providers[].kaniko.namespace

providers > kaniko > namespace

Choose the namespace where the Kaniko pods will be run. Set to null to use the project namespace.

IMPORTANT: The default namespace will change to the project namespace instead of the garden-system namespace in an upcoming release!

TypeDefaultRequired

string

"garden-system"

No

providers[].kaniko.nodeSelector

providers > kaniko > nodeSelector

Exposes the nodeSelector field on the PodSpec of the Kaniko pods. This allows you to constrain the Kaniko pods to only run on particular nodes. The same nodeSelector will be used for each util pod unless they are specifically set under util.nodeSelector.

See here for the official Kubernetes guide to assigning pods to nodes.

TypeRequired

object

No

providers[].kaniko.tolerations[]

providers > kaniko > tolerations

Specify tolerations to apply to each Kaniko builder pod. Useful to control which nodes in a cluster can run builds. The same tolerations will be used for each util pod unless they are specifically set under util.tolerations

TypeDefaultRequired

array[object]

[]

No

providers[].kaniko.tolerations[].effect

providers > kaniko > tolerations > effect

"Effect" indicates the taint effect to match. Empty means match all taint effects. When specified, allowed values are "NoSchedule", "PreferNoSchedule" and "NoExecute".

TypeRequired

string

No

providers[].kaniko.tolerations[].key

providers > kaniko > tolerations > key

"Key" is the taint key that the toleration applies to. Empty means match all taint keys. If the key is empty, operator must be "Exists"; this combination means to match all values and all keys.

TypeRequired

string

No

providers[].kaniko.tolerations[].operator

providers > kaniko > tolerations > operator

"Operator" represents a key's relationship to the value. Valid operators are "Exists" and "Equal". Defaults to "Equal". "Exists" is equivalent to wildcard for value, so that a pod can tolerate all taints of a particular category.

TypeDefaultRequired

string

"Equal"

No

providers[].kaniko.tolerations[].tolerationSeconds

providers > kaniko > tolerations > tolerationSeconds

"TolerationSeconds" represents the period of time the toleration (which must be of effect "NoExecute", otherwise this field is ignored) tolerates the taint. By default, it is not set, which means tolerate the taint forever (do not evict). Zero and negative values will be treated as 0 (evict immediately) by the system.

TypeRequired

string

No

providers[].kaniko.tolerations[].value

providers > kaniko > tolerations > value

"Value" is the taint value the toleration matches to. If the operator is "Exists", the value should be empty, otherwise just a regular string.

TypeRequired

string

No

providers[].kaniko.annotations

providers > kaniko > annotations

Specify annotations to apply to each Kaniko builder pod. Annotations may have an effect on the behaviour of certain components, for example autoscalers. The same annotations will be used for each util pod unless they are specifically set under util.annotations

TypeRequired

object

No

Example:

providers:
  - kaniko:
      ...
      annotations:
          cluster-autoscaler.kubernetes.io/safe-to-evict: 'false'

providers[].kaniko.util

providers > kaniko > util

TypeRequired

object

No

providers[].kaniko.util.tolerations[]

providers > kaniko > util > tolerations

Specify tolerations to apply to each garden-util pod.

TypeDefaultRequired

array[object]

[]

No

providers[].kaniko.util.tolerations[].effect

providers > kaniko > util > tolerations > effect

"Effect" indicates the taint effect to match. Empty means match all taint effects. When specified, allowed values are "NoSchedule", "PreferNoSchedule" and "NoExecute".

TypeRequired

string

No

providers[].kaniko.util.tolerations[].key

providers > kaniko > util > tolerations > key

"Key" is the taint key that the toleration applies to. Empty means match all taint keys. If the key is empty, operator must be "Exists"; this combination means to match all values and all keys.

TypeRequired

string

No

providers[].kaniko.util.tolerations[].operator

providers > kaniko > util > tolerations > operator

"Operator" represents a key's relationship to the value. Valid operators are "Exists" and "Equal". Defaults to "Equal". "Exists" is equivalent to wildcard for value, so that a pod can tolerate all taints of a particular category.

TypeDefaultRequired

string

"Equal"

No

providers[].kaniko.util.tolerations[].tolerationSeconds

providers > kaniko > util > tolerations > tolerationSeconds

"TolerationSeconds" represents the period of time the toleration (which must be of effect "NoExecute", otherwise this field is ignored) tolerates the taint. By default, it is not set, which means tolerate the taint forever (do not evict). Zero and negative values will be treated as 0 (evict immediately) by the system.

TypeRequired

string

No

providers[].kaniko.util.tolerations[].value

providers > kaniko > util > tolerations > value

"Value" is the taint value the toleration matches to. If the operator is "Exists", the value should be empty, otherwise just a regular string.

TypeRequired

string

No

providers[].kaniko.util.annotations

providers > kaniko > util > annotations

Specify annotations to apply to each garden-util pod and deployments.

TypeRequired

object

No

Example: