Skaffold Pipeline

You can configure Skaffold with the Skaffold configuration file, skaffold.yaml. A single configuration consists of several different components:

You can [learn more]({{< relref "/docs/references/yaml" >}}) about the syntax of skaffold.yaml.

Skaffold normally expects to find the configuration file as skaffold.yaml in the current directory, but the location can be overridden with the --filename flag.

File resolution

The Skaffold configuration file often references other files and directories. These files and directories are resolved relative to the current directory and not to the location of the Skaffold configuration file. There are two important exceptions:

1. Files referenced from a build artifact definition are resolved relative to the build artifact's context directory. When omitted, the context directory defaults to the current directory.
2. For configurations resolved as dependencies, paths are always resolved relative to the directory containing the imported configuration file.

For example, consider a project with the following layout:

.
├── frontend
│   └── Dockerfile
├── helm
│   └── project
│       └── dev-values.yaml
└── skaffold.yaml

The config file might look like:

apiVersion: skaffold/v2beta11
kind: Config
build:
  artifacts:
  - image: app
    context: frontend
    docker:
      dockerfile: "Dockerfile"
deploy:
  helm:
    releases:
    - name: project
      chartPath: helm/project
      valuesFiles:
      - "helm/project/dev-values.yaml"

In this example, the Dockerfile for building app is resolved relative to app's context directory, whereas the the Helm chart's location and its values-files are relative to the current directory in helm/project.

We generally recommend placing the configuration file in the root directory of the Skaffold project.

Multiple configuration support

A single skaffold.yaml file can define multiple skaffold configurations in the schema described above using the separator ---. If these configuration objects define the metadata.name property then we consider them as modules, that can then be activated by name.

Consider a skaffold.yaml defined as:

apiVersion: skaffold/vX
kind: Config
metadata:
  name: cfg1
build:
  # build definition
deploy:
  # deploy definition

---

apiVersion: skaffold/vX
kind: Config
metadata:
  name: cfg2
build:
  # build definition
deploy:
  # deploy definition

Here cfg1 and cfg2 are independent skaffold modules. Running skaffold dev for instance will execute actions from both these modules. You could also run skaffold dev --module cfg1 to only activate the cfg1 module and skip cfg2.

Configuration dependencies

In addition to authoring configurations in a skaffold.yaml file, we can also import other existing configurations as dependencies. Skaffold manages all imported and defined configurations in the same session. It also ensures all artifacts in a required config are built prior to those in current config (provided the artifacts have dependencies defined); and all deploys in required configs are applied prior to those in current config.

{{< alert title="Note:" >}} Running skaffold <command> --module <config-name> will filter to the specified target module, but also include the transitive closure of all other configurations in its dependency graph. For instance, if a module cfg1 imported another module cfg2 as a dependency while cfg2 imported cfg3 and cfg4, then running skaffold dev --module cfg1 would activate all of cfg1, cfg2, cfg3 and cfg4 and execute them in dependency order. {{< /alert >}}

Local config dependency

Consider the same skaffold.yaml defined above. Modules cfg1 and cfg2 from the above file can be imported as dependencies in your current config definition, via:

apiVersion: skaffold/v2beta11
kind: Config
requires:
  - configs: ["cfg1", "cfg2"]
    path: path/to/other/skaffold.yaml 
build:
  # build definition
deploy:
  # deploy definition

If the configs list isn't defined then it imports all the configs defined in the file pointed by path. Additionally, if the path to the configuration isn't defined it assumes that all the required configs are defined in the same file as the current config.

{{< alert title="Note:" >}} In imported configurations, files are resolved relative to the location of imported Skaffold configuration file. {{< /alert >}}

Remote config dependency

The required skaffold config can live in a remote git repository or in Google Cloud Storage:

apiVersion: skaffold/v4beta7
kind: Config
requires:
  - configs: ["cfg1", "cfg2"]
    git:
      repo: http://github.com/GoogleContainerTools/skaffold.git
      path: getting-started/skaffold.yaml
      ref: main
  - configs: ["cfg3"]
    googleCloudStorage:
      source: gs://my-bucket/dir1/*
      path: config/skaffold.yaml

The environment variable SKAFFOLD_REMOTE_CACHE_DIR or flag --remote-cache-dir specifies the download location for all remote dependency contents. If undefined then it defaults to ~/.skaffold/remote-cache. The remote cache directory consists of subdirectories with the contents retrieved from the remote dependency. For git dependencies the subdirectory name is a hash of the repo uri and the branch/ref. For Google Cloud Storage dependencies the subdirectory name is a hash of the source.

The remote config gets treated like a local config after substituting the path with the actual path in the cache directory.

Profile Activation in required configs

Profiles specified by the --profile flag are also propagated to all configurations imported as dependencies, if they define them. This behavior can be disabled by setting the --propagate-profiles flag to false.

You can additionally set up more granular and conditional profile activations across dependencies through the activeProfiles stanza:

apiVersion: skaffold/v2beta11
kind: Config
metadata:
    name: cfg
requires:
  - path: ./path/to/required/skaffold.yaml
    configs: [cfg1, cfg2]                 
    activeProfiles:                                     
     - name: profile1                               
       activatedBy: [profile2, profile3] 

Here, profile1 is a profile that needs to exist in both configs cfg1 and cfg2; while profile2 and profile3 are profiles defined in the current config cfg. If the current config is activated with either profile2 or profile3 then the required configs cfg1 and cfg2 are imported with profile1 applied. If the activatedBy clause is omitted then that profile1 always gets applied for the imported configs.

{{< alert title="Follow up" >}} Take a look at the [tutorial]({{< relref "/docs/tutorials/config-dependencies" >}}) section to see this in action. {{< /alert >}}