Keptn v1 reached EOL December 22, 2023. For more information see


The shipyard is configured in the shipyard.yaml file, which defines the activities to be performed for a Keptn project and the order in which those activities are executed.

Each project must have one, and only one, shipyard.yaml file which is passed to the keptn create project command.

  • Each shipyard contains one or more stages, which can be given any name that is meaningful. Examples are “development”, “hardening”, “production”, and “remediation”. A stage is a grouping of activities to be executed until the project is deployed and, optionally, a “production” stage that defines remediation activities that can be executed in response to issues detected on the production site.

  • Each stage must have one or more sequences, which can be given any name that is meaningful. A sequence defines the tasks to be performed and, optionally, an event that triggers that sequence.

The following Synopsis shows all the constructions that are supported for a shipyard although most projects only use some of the constructions.


kind: "Shipyard"
  name: "shipyard-<project-name>"
    - name: "<stagename-1>"
      - name: "delivery"
        - name: "deployment"
            deploymentstrategy: "direct" | "blue_green_service" | "user_managed"
        - name: "release"
    - name: "hardening"
    - name: "production"
      - name: "<delivery-sequence>"
           - event: "<event>.finished"
         - name: "delivery"
      - name: "remediation-sequence>"
           -event: "<event>"
                evaluation.result: "<result>"
            - name: "get-action"
            - name: "action"
            - name: "evaluation"
              triggeredAfter: "<timeframe>"
                timeframe: "<xx>m"



  • apiVersion: The version of the shipyard specification in the format:
  • kind: is Shipyard
  • metadata: name: Unique name for this shipyard file Typically, this is the string shipyard followed by a dash and the project name; for example, shipyard-sockshop
  • spec: Consists of the property stages.
    • stages: An array of stages, each of which has a name and a sequence of tasks to execute


A stage is named for the particular activity to be performed, such as development, hardening, staging, or production. Each shipyard file must have at least one stage. The name of the stage becomes the name of the branch in the upstream Git repository and the Kubernetes namespace to which services are deployed.

A stage can be given any meaningful name that conforms to the Kubernetes Object Names and IDs specification, meaning:

  • Contains only lowercase alphanumeric characters or -. Most especially, must not contain / or %.

  • Starts and ends with an alphanumeric character.

  • Contains at most 43 characters.

    Kubernetes allows up to 63 character names (because of DNS limitations). For continuous delivery, the helm-service creates Helm releases named <serviceName>-generated, where <serviceName> is the stage name.. Because “-generated” is 10 characters, the stage name can not be longer than 43 characters.

    It is possible that larger values are allowed if you are not using the helm-service but this has not been tested and is not guaranteed.

A stage has the properties:

  • name: A unique name for the stage such as development, hardening, staging, or `production.
  • sequences: An array of sequences that define the tasks to be performed and, optionally, the events that trigger each task.

You can not add or delete stages in the shipyard file for an existing project although you can make other modifications.


A sequence is an ordered list of tasks that are triggered sequentially and are part of a stage. By default, a sequence is a standalone section that runs and finishes, unless you specify the triggeredOn property to form a chain of sequences.

Each stage includes an implicit evaluation sequence with a single task that is also called evaluation. You can define your own evaluation sequence with any set of tasks to override this default sequence but each stage must have either an implicit or explicit evaluation sequence.

Sequences that use the same service cannot be run in parallel. If you simultaneously trigger multiple sequences for the same service, they are queued to run sequentially. Sequences for different services can be run in parallel. This is possible when you have different automation projects or if you have multiple services within a project.

A sequence has the properties:

  • name: A unique name for the sequence

  • tasks: An array of tasks executed by the sequence in the declared order.

  • triggeredOn (optional): An array of events that trigger the sequence. This property can be used to trigger a sequence once another sequence has been finished, essentially forming chains of sequences. In addition to specifying the sequence whose completion should activate the trigger, you can define a selector that defines whether the sequence should be triggered if the preceeding sequence has been executed successfuly, or had a failed or warning result. For example, the following sequence with the name rollback is only triggered if the sequence delivery in production had a result of failed:

      - name: rollback
        - event:
              result: failed

    It is also possible to refer to certain tasks within the preceeding sequence. For example, if match is changed to release.result: failed, the rollback sequence is executed only if the task release of the sequence delivery has a result of failed:

      - name: rollback
        - event:
              release.result: failed

    If no selector is specified, the sequence is triggered only if the preceeding delivery sequence has a result of pass:

      - name: rollback
        - event:


A single task is the smallest executable unit and is contained in a sequences block. A task has the properties:

  • name: A unique name for the task
  • triggeredAfter (optional): Wait time before the task is triggered.
  • properties (optional): Task properties as individual key:value pairs. These properties are properties that the actioning tool requires and are consumed by the tool that executes the task. Typically, properties are passed in at runtime using JSON data rather than having the data hardcoded into the shipyard file.

Keptn supports a set of opinionated tasks for declaring a delivery or remediation sequence. Additional tasks may be defined for the services you integrate.

  • action
  • approval
  • deployment
  • evaluation
  • get-action
  • rollback

In addition, the following two tasks are reserved although they are associated with specific services:

  • release (helm-service only)
  • test (jmeter-service only)

Each of these are discussed below.

  • action

    Indicates that a remediation action should be executed by an action provider that is defined in a remediation configuration.

  • approval

    Intercepts the task sequence and waits for a user to approve or decline the open approval. This task can be added, for example, before deploying an artifact into the next stage. The approval strategy is defined based on the evaluation result pass and warning. The approval strategies for the evaluation results pass and warning can be set to:

    • automatic: Task sequence continues without requesting approval
    • manual: Task sequence requests for approval before continuing


      - name: approval
           pass: automatic
           warning: manual

    This allows combinations as follows:

    Evaluation result: pass Evaluation result: warning Behavior
    Skip approval task: pass:automatic warning:automatic Regardless of the evaluation result, the approval task is skipped
    Depending on evaluation result: pass:automatic warning:manual If the evaluation result is a warning, an approval is required
    Depending on evaluation result: pass:manual warning:automatic If the evaluation result is a pass, an approval is required
    Mandatory approval task: pass:manual warning:manual Regardless of the evaluation result, an approval is required

    By default, an automatic approval strategy is used for the pass and warning evaluation results.

  • evaluation

    Defines the quality evaluation that is executed to verify the quality of a deplyoment based on its SLOs/SLIs. Every stage must include an evaluation sequence. If you do not define an evaluation sequence for a stage, an “implicit” evaluation sequence with a single task that is also called evaluation is used. This default evaluation sequence is overridden by any explicit evaluation sequence that you define for your project.

    Use the optional triggeredAfter parameter to specify when to trigger the evaluation.

    Set the timeframe property to specify the timespan to be evaluated. For example, timeframe: 5m says that the quality gate evaluation looks at the previous five minutes.

    timeframe must be specified but you can specify the timeframe as part of the JSON payload and pass it in when you trigger the sequence using curl rather than hard-coding it in the file. This makes the timeframe value dynamic. For example:

       "type": "sh.keptn.event.SomeStage.MySequence.triggered",
       # Other fields removed for brevity
       "data": {
         "evaluation": {
           "timeframe": "5m"
  • get-action Extracts the desired remediation action from a remediation configuration.

  • release

    Defines the releasing task that is executed after a successful deployment occurs. This task shifts production trafic towards the new deployment.

  • rollback

    Defines the rollback task that is executed when a rollback is triggered.

  • remediation

    Defines whether remediation actions are enabled or not.

For historical reasons and backward compatibility, the following tasks are reserved in Keptn although they are actually associated with services that run on the execution plane rather than on the control plane.

  • deployment

    Defines the deployment strategy used to deploy a new version of a service. This is part of helm-service, which assumes that Istio is installed on the cluster unless the Job Executor Service is installed in the cluster. The deployment strategy is set to one of the following:

    • direct: Deploys a new version of a service by replacing the old version of the service.
    • blue_green_service: Deploys a new version of a service next to the old one. After a successful validation of this new version, it replaces the old one and is marked as stable.
    • user_managed: Deploys a new version of a service by fetching the current Helm chart from the Git repo and updating appropriate values.

    This is discussed more in Deployment with Helm

  • test

    Defines the test strategy used to validate a deployment with the jmeter-service. The jmeter-service supports setting the teststrategy to one of the following:

    • functional: Test a deployment based on functional tests.
    • performance: Test a deployment based on performance/load tests.

    Task properties are a way to pass additional Key / Value metadata that actioning integration uses as part of it’s logic. In this case, the jmeter-service expects either teststrategy: functional or teststrategy: performance to be set. If you action the test task with another service you may not need to set these details.

    Failed tests result in an automatic rollback of the latest deployment when using a blue/green deployment strategy.

      - name: test
          teststrategy: functional | performance

    For example, you may run functional tests in a dev stage and performance tests in a hardening or `staging stage.


A shipyard is defined at the level of a project. This means that all services in a project share the same shipyard definition.

You can not add or delete stages in the shipyard file for an existing project although you can make other modifications.

As a workaround, you can temporarily skip the execution of a particular stage by doing either of the following:

  • Temporarily remove the triggeredOn attribute for the stage you want to skip
  • Add an approval step as the first task of the stage you want to skip and then Deny the approvals


The See also section of this page references other pages in this documentation set that contain annotated examples for accomplishing various tasks.


  • Your shipyard is stored in the upstage Git repo for the project and named to match the value of the Metadata name field in the shipyard file.

Differences between versions

See also