docs/content/master/composition/compositions.md

title	weight	aliases	description
Compositions	30	composition composition-functions /knowledge-base/guides/composition-functions	Define which resources to create and how

Compositions are a template for creating multiple Kubernetes resources as a single composite resource.

A Composition composes individual resources together into a larger, reusable, solution.

An example Composition may combine a virtual machine, storage resources and networking policies. A Composition template links all these individual resources together.

Here's an example Composition. When you create an {{}}AcmeBucket{{}} composite resource (XR) that uses this Composition, Crossplane uses the template to create the Amazon S3 {{}}Bucket{{}} managed resource.

apiVersion: apiextensions.crossplane.io/v1
kind: Composition
metadata:
  name: example
spec:
  compositeTypeRef:
    apiVersion: custom-api.example.org/v1alpha1
    kind: AcmeBucket
  mode: Pipeline
  pipeline:
  - step: patch-and-transform
    functionRef:
      name: function-patch-and-transform
    input:
      apiVersion: pt.fn.crossplane.io/v1beta1
      kind: Resources
      resources:
      - name: storage-bucket
        base:
          apiVersion: s3.aws.m.upbound.io/v1beta1
          kind: Bucket
          spec:
            forProvider:
              region: "us-east-2"

{{<expand "What are XRs, XRDs and Compositions?" >}} A [composite resource]({{<ref "./composite-resources">}}) or XR is a custom API.

You use two Crossplane types to create a new custom API:

• A [Composite Resource Definition]({{<ref "./composite-resource-definitions">}}) (XRD) - Defines the XR's schema.
• A Composition - This page. Configures how the XR creates other resources. {{}}

Create a composition

Creating a Composition consists of:

• Using composition functions to define the resources to create.
• Enabling composite resources to use the Composition template.

A Composition is a pipeline of composition functions.

Composition functions (or just functions, for short) are Crossplane extensions that template Crossplane resources. Crossplane calls the composition functions to determine what resources it should create when you create a composite resource (XR).

{{<hint "tip" >}} Crossplane has functions that let you template composed resources using YAML [patch and transforms]({{<ref "../guides/function-patch-and-transform">}}). Helm-like YAML templates, CUE, KCL, or Python.

You can also write your own function using Go or Python. {{< /hint >}}

Install a composition function

Installing a Function creates a function pod. Crossplane sends requests to this pod to ask it what resources to create when you create a composite resource.

Install a Function with a Crossplane {{}}Function{{}} object setting the {{}}spec.package{{}} value to the location of the function package.

For example, to install [Function Patch and Transform]({{<ref "../guides/function-patch-and-transform">}}),

apiVersion: pkg.crossplane.io/v1
kind: Function
metadata:
  name: function-patch-and-transform
spec:
  package: xpkg.crossplane.io/crossplane-contrib/function-patch-and-transform:v0.8.2

{{< hint "tip" >}} Functions are Crossplane Packages. Read more about Packages in the [Packages documentation]({{<ref "../packages/functions" >}}). {{< /hint >}}

By default, the Function pod installs in the same namespace as Crossplane (crossplane-system).

Verify a composition function

View the status of a Function with kubectl get functions

During the install a Function reports INSTALLED as True and HEALTHY as Unknown.

kubectl get functions
NAME                              INSTALLED   HEALTHY   PACKAGE                                                                  AGE
function-patch-and-transform      True        Unknown   xpkg.crossplane.io/crossplane-contrib/function-patch-and-transform:v0.8.2   10s

After the Function install completes and it's ready for use the HEALTHY status reports True.

Use a function in a composition

Crossplane calls a Function to determine what resources it should create when you create a composite resource. The Function also tells Crossplane what to do with these resources when you update a composite resource.

{{<hint "note" >}} Composition functions don't run when you delete a composite resource. Crossplane handles deletion of composed resources automatically. {{< /hint >}}

When Crossplane calls a Function it sends it the current state of the composite resource. It also sends it the current state of any resources the composite resource owns.

Crossplane knows what Function to call when a composite resource changes by looking at the Composition the composite resource uses.

To use composition functions set the Composition {{}}mode{{}} to {{}}Pipeline{{}}.

Define a {{}}pipeline{{}} of {{}}steps{{}}. Each {{}}step{{}} calls a Function.

Each {{}}step{{}} uses a {{}}functionRef{{}} to reference the {{}}name{{}} of the Function to call.

Some Functions also allow you to specify an {{}}input{{}}. The function defines the {{}}kind{{}} of input.

This example uses [Function Patch and Transform]({{<ref "../guides/function-patch-and-transform">}}). Function Patch and Transform implements Crossplane resource templates. The input kind is {{}}Resources{{}}, and it accepts {{}}resources{{}} as input.

apiVersion: apiextensions.crossplane.io/v1
kind: Composition
# Removed for Brevity
spec:
  # Removed for Brevity
  mode: Pipeline
  pipeline:
  - step: patch-and-transform
    functionRef:
      name: function-patch-and-transform
    input:
      apiVersion: pt.fn.crossplane.io/v1beta1
      kind: Resources
      resources:
      - name: storage-bucket
        base:
          apiVersion: s3.aws.m.upbound.io/v1beta1
          kind: Bucket
          spec:
            forProvider:
              region: "us-east-2"

Use a pipeline of functions in a composition

Crossplane can ask more than one Function what to do when a composite resource changes. When a Composition has a pipeline of two or more steps, Crossplane calls them all. It calls them in the order they appear in the pipeline.

Crossplane passes each Function in the pipeline the result of the previous Function. This enables powerful combinations of Functions. In this example, Crossplane calls {{}}function-cue{{}} to create an S3 bucket. Crossplane then passes the bucket to {{}}function-auto-ready{{}}, which marks the composite resource as ready when the bucket becomes ready.

apiVersion: apiextensions.crossplane.io/v1
kind: Composition
# Removed for Brevity
spec:
  # Removed for Brevity
  mode: Pipeline
  pipeline:
  - step: cue-export-resources
    functionRef:
      name: function-cue
    input:
      apiVersion: cue.fn.crossplane.io/v1beta1
      kind: CUEInput
      name: storage-bucket
      export:
        target: Resources
        value: |
          apiVersion: "s3.aws.m.upbound.io/v1beta1"
          kind: "Bucket"
          spec: forProvider: region: "us-east-2"          
  - step: automatically-detect-readiness
    functionRef:
      name: function-auto-ready

Match composite resources

A Composition is only a template defining how to create composed resources. A Composition limits which kind of composite resource (XR) can use this template.

A Composition's {{}}compositeTypeRef{{}} defines which Composite Resource type can use this Composition.

{{<hint "note" >}} Read more about Composite Resources in the [Composite Resources page]({{<ref "./composite-resources" >}}). {{< /hint >}}

Inside a Composition's {{}}spec{{}} define the Composite Resource {{}}apiVersion{{}} and {{}}kind{{}} that the Composition allows to use this template.

apiVersion: apiextensions.crossplane.io/v1
kind: Composition
metadata:
  name: dynamodb-with-bucket
spec:
  compositeTypeRef:
    apiVersion: custom-api.example.org/v1alpha1
    kind: database
  # Removed for brevity

Grant access to composed resources

Crossplane uses its service account to create the composed resources that a function pipeline returns.

Crossplane's service account has access to create, update, and delete any resource installed by a [provider]({{<ref "../packages/providers">}}), or defined by an XRD. This includes all [MRs]({{<ref "../managed-resources/managed-resources">}}) and [XRs]({{<ref "composite-resources">}}). It also has access to some types of Kubernetes resources that it needs to function - for example it can create deployments.

You must grant Crossplane access to compose any other kind of resource. You do this by creating an RBAC ClusterRole.

The ClusterRole must aggregate to Crossplane's primary ClusterRole using ClusterRole aggregation.

Here's a ClusterRole that grants Crossplane access to manage CloudNativePG PostgreSQL clusters.

apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRole
metadata:
  name: cnpg:aggregate-to-crossplane
  labels:
    rbac.crossplane.io/aggregate-to-crossplane: "true"
rules:
- apiGroups:
  - postgresql.cnpg.io
  resources:
  - clusters
  verbs:
  - "*"

The rbac.crossplane.io/aggregate-to-crossplane: "true" label is critical. It configures the role to aggregate to Crossplane's primary cluster role.

{{<hint "note" >}} The [RBAC manager]({{<ref "../guides/pods#rbac-manager-pod">}}) automatically grants Crossplane access to MRs and XRs. The RBAC manager uses escalate access to grant Crossplane access that the RBAC manager doesn't have.

The RBAC manager is an optional Crossplane component that's enabled by default. If you disable the RBAC manager, you must manually grant Crossplane access to any kind of resource you wish to compose - including XRs and MRs. {{< /hint >}}

Test a composition

You can preview the output of any composition using the Crossplane CLI. You don't need a Crossplane control plane to do this. The Crossplane CLI uses Docker Engine to run functions.

{{<hint "tip">}} See the [Crossplane CLI docs]({{<ref "../cli">}}) to learn how to install and use the Crossplane CLI. {{< /hint >}}

{{<hint "important">}} Running crossplane render requires Docker. {{< /hint >}}

Provide a composite resource, composition and composition functions to render the output locally.

crossplane render xr.yaml composition.yaml functions.yaml

crossplane render prints resources as YAML to stdout. It prints the composite resource first, followed by the resources the composition functions created.

---
apiVersion: example.crossplane.io/v1
kind: Bucket
metadata:
  name: example-render
---
apiVersion: s3.aws.m.upbound.io/v1beta1
kind: Bucket
metadata:
  annotations:
    crossplane.io/composition-resource-name: storage-bucket
  generateName: example-render-
  labels:
    crossplane.io/composite: example-render
  ownerReferences:
  - apiVersion: example.crossplane.io/v1
    blockOwnerDeletion: true
    controller: true
    kind: Bucket
    name: example-render
    uid: ""
spec:
  forProvider:
    region: us-east-2

{{<expand "The xr.yaml, composition.yaml and function.yaml files used in the example">}}

You can recreate the output below by running crossplane render with these files.

The xr.yaml file contains the composite resource to render:

apiVersion: example.crossplane.io/v1
kind: Bucket
metadata:
  name: example-render
spec:
  bucketRegion: us-east-2

The composition.yaml file contains the Composition to use to render the composite resource:

apiVersion: apiextensions.crossplane.io/v1
kind: Composition
metadata:
  name: example-render
spec:
  compositeTypeRef:
    apiVersion: example.crossplane.io/v1
    kind: Bucket
  mode: Pipeline
  pipeline:
  - step: patch-and-transform
    functionRef:
      name: function-patch-and-transform
    input:
      apiVersion: pt.fn.crossplane.io/v1beta1
      kind: Resources
      resources:
      - name: storage-bucket
        base:
          apiVersion: s3.aws.m.upbound.io/v1beta1
          kind: Bucket
        patches:
        - type: FromCompositeFieldPath
          fromFieldPath: spec.bucketRegion
          toFieldPath: spec.forProvider.region

The functions.yaml file contains the Functions the Composition references in its pipeline steps:

---
apiVersion: pkg.crossplane.io/v1
kind: Function
metadata:
  name: function-patch-and-transform
spec:
  package: xpkg.crossplane.io/crossplane-contrib/function-patch-and-transform:v0.8.2

{{}}

The Crossplane CLI uses Docker Engine to run functions. You can change how the Crossplane CLI runs a function by adding an annotation in functions.yaml. Add the render.crossplane.io/runtime annotation to a Function to change how it's run.

crossplane render supports two render.crossplane.io/runtime values:

• Docker (the default) connects to Docker Engine. It uses Docker to pull and run a function runtime.
• Development connects to a function runtime you have run manually.

When you use the {{}}Development{{}} runtime the Crossplane CLI ignores the Function's {{}}package{{}}. Instead it expects you to make sure the function is listening on localhost port 9443. The function must be listening without gRPC transport security. Most function SDKs let you run a function with the --insecure flag to disable transport security. For example you can run a Go function locally using go run . --insecure.

apiVersion: pkg.crossplane.io/v1
kind: Function
metadata:
  name: function-patch-and-transform
  annotations:
    render.crossplane.io/runtime: Development
spec:
  package: xpkg.crossplane.io/crossplane-contrib/function-patch-and-transform:v0.8.2

{{<hint "tip">}} Use the Development runtime when you write a composition function to test your function end-to-end. {{}}

crossplane render also supports the following Function annotations. These annotations affect how it runs Functions:

• render.crossplane.io/runtime-docker-cleanup - When using the Docker runtime this annotation specifies whether the CLI should stop the function container after it calls the function. It supports the values Stop, to stop the container, and Orphan, to leave it running.
• render.crossplane.io/runtime-docker-pull-policy - When using the Docker runtime this annotation specifies when the CLI should pull the Function's package. It supports the values Always, Never, and IfNotPresent.
• render.crossplane.io/runtime-development-target - When using the Development runtime this annotation tells the CLI to connect to a Function running at the specified target. It uses gRPC target syntax.

Verify a composition

View all available Compositions with kubectl get composition.

kubectl get composition
NAME                                       XR-KIND        XR-APIVERSION                         AGE
xapps.aws.platformref.upbound.io           XApp           aws.platformref.upbound.io/v1alpha1   123m
xclusters.aws.platformref.upbound.io       XCluster       aws.platformref.upbound.io/v1alpha1   123m
xeks.aws.platformref.upbound.io            XEKS           aws.platformref.upbound.io/v1alpha1   123m
xnetworks.aws.platformref.upbound.io       XNetwork       aws.platformref.upbound.io/v1alpha1   123m
xservices.aws.platformref.upbound.io       XServices      aws.platformref.upbound.io/v1alpha1   123m
xsqlinstances.aws.platformref.upbound.io   XSQLInstance   aws.platformref.upbound.io/v1alpha1   123m

The XR-KIND lists the Composite Resource kind that's allowed to use the Composition template. The XR-APIVERSION lists the Composite Resource API versions allowed to use the Composition template.

{{<hint "note" >}} The output of kubectl get composition is different than kubectl get composite.

kubectl get composition lists all available Compositions.

kubectl get composite lists all created Composite Resources and their related Composition. {{< /hint >}}

Write a composition function

Composition functions let you replace complicated Compositions with code written in your programming language of choice. Crossplane has tools, software development kits (SDKs) and templates to help you write a function.

Here's an example of a tiny, hello world function. This example is written in Go.

func (f *Function) RunFunction(_ context.Context, req *fnv1.RunFunctionRequest) (*fnv1.RunFunctionResponse, error) {
        rsp := response.To(req, response.DefaultTTL)
        response.Normal(rsp, "Hello world!")
        return rsp, nil
}

Crossplane has [language specific guides]({{<ref "../guides">}}) to writing a composition function. Refer to the guide for your preferred language to learn how to write a composition function.

When you're writing a composition function it's useful to know how composition functions work. Read the next section to learn how composition functions work.

How composition functions work

Each composition function is actually a gRPC server. gRPC is a high performance, open source remote procedure call (RPC) framework. When you install a function Crossplane deploys the function as a gRPC server. Crossplane encrypts and authenticates all gRPC communication.

You don't have to be a gRPC expert to write a function. Crossplane's function SDKs setup gRPC for you. It's useful to understand how Crossplane calls your function though, and how your function should respond.

sequenceDiagram
    User->>+API Server: Create composite resource
    Crossplane Pod->>+API Server: Observe composite resource
    Crossplane Pod->>+Function Pod: gRPC RunFunctionRequest
    Function Pod->>+Crossplane Pod: gRPC RunFunctionResponse
    loop Extra resources needed?
      Crossplane Pod->>+API Server: Get Extra resources
      Crossplane Pod->>+Function Pod: gRPC RunFunctionRequest
      Function Pod->>+Crossplane Pod: gRPC RunFunctionResponse
    end
    Crossplane Pod->>+API Server: Apply desired composed resources

When you create or update a composite resource that uses composition functions Crossplane calls each function in the order they appear in the Composition's pipeline. Crossplane calls each function by sending it a gRPC RunFunctionRequest. The function must respond with a gRPC RunFunctionResponse.

{{<hint "tip">}} You can find detailed schemas for the RunFunctionRequest and RunFunctionResponse RPCs in the Buf Schema Registry. {{}}

When Crossplane calls a function the first time it includes four important things in the RunFunctionRequest.

1. The observed state of the composite resource, and any composed resources.
2. The desired state of the composite resource, and any composed resources.
3. The function's input.
4. The function pipeline's context.

A function's main job is to update the desired state and return it to Crossplane. It does this by returning a RunFunctionResponse.

Most composition functions read the observed state of the composite resource, and use it to add composed resources to the desired state. This tells Crossplane which composed resources it should create or update.

If the function needs required resources to determine the desired state it can request any cluster-scoped or namespaced resource Crossplane already has access to, either by name or labels through the returned RunFunctionResponse. Crossplane then calls the function again including the requested required resources and the context returned by the Function itself alongside the same input, observed and desired state of the previous RunFunctionRequest. Functions can iteratively request required resources if needed, but to avoid endlessly looping Crossplane limits the number of iterations to 5. Crossplane considers the function satisfied as soon as the required resources requests become stable, so the Function returns the same exact request two times in a row. Crossplane errors if stability isn't reached after 5 iterations.

{{<hint "tip">}} A composed resource is a resource created by a composite resource. Composed resources can be any kind of Kubernetes resource. {{}}

Observed state

When you create a composite resource like this one, Crossplane observes it and sends it to the composition function as part of the observed state.

apiVersion: example.crossplane.io/v1
kind: Bucket
metadata:
  name: example-render
spec:
  bucketRegion: us-east-2

If any composed resources already exist, Crossplane observes them and sends them to your function as part of the observed state.

Crossplane also observes the connection details of your composite resource and any composed resources. It sends them to your function as part of the observed state.

Crossplane observes the composite resource and any composed resources once, right before it starts calling the functions in the pipeline. This means that Crossplane sends every function in the pipeline the same observed state.

Desired state

Desired state is the set of the changes the function pipeline wants to make to the composite resource and any composed resources. When a function adds composed resources to the desired state Crossplane creates them.

A function can change:

• The status of the composite resource.
• The metadata and spec of any composed resource.

A function can also change the connection details and readiness of the composite resource. A function indicates that the composite resource is ready by telling Crossplane whether its composed resources are ready. When the function pipeline tells Crossplane that all composed resources are ready, Crossplane marks the composite resource as ready.

A function can't change:

• The metadata or spec of the composite resource.
• The status of any composed resource.
• The connection details of any composed resource.

A pipeline of functions accumulates desired state. This means that each function builds upon the desired state of previous functions in the pipeline. Crossplane sends a function the desired state accumulated by all previous functions in the pipeline. The function adds to or updates the desired state and then passes it on. When the last function in the pipeline has run, Crossplane applies the desired state it returns.

{{<hint "important">}} A function must copy all desired state from its RunFunctionRequest to its RunFunctionResponse. If a function adds a resource to its desired state the next function must copy it to its desired state. If it doesn't, Crossplane doesn't apply the resource. If the resource exists, Crossplane deletes it.

A function can intentionally choose not to copy parts of the desired state. For example a function may choose not to copy a desired resource to prevent that resource from existing.

Most function SDKs handle copying desired state automatically. {{}}

A function should only add the fields it cares about to the desired state. It should add these fields every time Crossplane calls it. If a function adds a field to the desired state once, but doesn't add it the next time it's called, Crossplane deletes the field. The same is true for composed resources. If a function adds a composed resource to the desired state, but doesn't add it the next time it's called, Crossplane deletes the composed resource.

{{<hint "tip">}} Crossplane uses server side apply to apply the desired state returned by a function pipeline. In server side apply terminology, the desired state is a fully specified intent. {{}}

For example, if all a function wants is to make sure an S3 bucket in region us-east-2 exists, it should add this resource to its desired composed resources.

apiVersion: s3.aws.m.upbound.io/v1beta1
kind: Bucket
spec:
  forProvider:
    region: us-east-2

Even if the Bucket already exists and has other spec fields, or a status, name, labels, etc the function should omit them. The function should only include the fields it has an opinion about. Crossplane takes care of applying the fields the function cares about, merging them with the existing Bucket.

{{<hint "tip">}} Composition functions don't actually use YAML for desired and observed resources. This example uses YAML for illustration purposes only. {{}}

Function input

If a Composition includes {{}}input{{}} Crossplane sends it to the function. Input is a useful way to provide extra configuration to a function. Supporting input is optional. Not all functions support input.

apiVersion: apiextensions.crossplane.io/v1
kind: Composition
metadata:
  name: example-render
spec:
  compositeTypeRef:
    apiVersion: example.crossplane.io/v1
    kind: Bucket
  mode: Pipeline
  pipeline:
  - step: patch-and-transform
    functionRef:
      name: function-patch-and-transform
    input:
      apiVersion: pt.fn.crossplane.io/v1beta1
      kind: Resources
      resources:
      - name: storage-bucket
        base:
          apiVersion: s3.aws.m.upbound.io/v1beta1
          kind: Bucket
        patches:
        - type: FromCompositeFieldPath
          fromFieldPath: spec.bucketRegion
          toFieldPath: spec.forProvider.region

{{<hint "important">}} Crossplane doesn't validate function input. It's a good idea for a function to validate its own input. {{}}

Required resources

{{<hint "note">}} Crossplane v1 called this feature "extra resources." The v2 API uses the name "required resources" and adds support for bootstrap requirements. {{}}

Functions can request access to existing Kubernetes resources to help determine the desired state. Functions use this capability to read configuration from ConfigMaps, select the status of other resources, or make decisions based on existing cluster state.

Functions can receive required resources in two ways:

Bootstrap requirements

You can provide required resources in the Composition pipeline step. This approach performs better than requesting resources during function runtime:

apiVersion: apiextensions.crossplane.io/v1
kind: Composition
metadata:
  name: app-with-config
spec:
  compositeTypeRef:
    apiVersion: example.crossplane.io/v1
    kind: App
  mode: Pipeline
  pipeline:
  - step: create-deployment-from-config
    functionRef:
      name: crossplane-contrib-function-python
    requirements:
      requiredResources:
      - requirementName: app-config
        apiVersion: v1
        kind: ConfigMap
        name: app-configuration
        namespace: default
    input:
      apiVersion: python.fn.crossplane.io/v1beta1
      kind: Script
      script: |
        from crossplane.function import request
        
        def compose(req, rsp):
            observed_xr = req.observed.composite.resource
            
            # Access the required ConfigMap using the helper function
            config_map = request.get_required_resource(req, "app-config")
            
            if not config_map:
                # Fallback image if ConfigMap not found
                image = "nginx:latest"
            else:
                # Read image from ConfigMap data
                image = config_map.get("data", {}).get("image", "nginx:latest")
            
            # Create deployment with the configured image
            rsp.desired.resources["deployment"].resource.update({
                "apiVersion": "apps/v1",
                "kind": "Deployment",
                "metadata": {
                    "labels": {"example.crossplane.io/app": observed_xr["metadata"]["name"]},
                },
                "spec": {
                    "replicas": 2,
                    "selector": {"matchLabels": {"example.crossplane.io/app": observed_xr["metadata"]["name"]}},
                    "template": {
                        "metadata": {
                            "labels": {"example.crossplane.io/app": observed_xr["metadata"]["name"]},
                        },
                        "spec": {
                            "containers": [{
                                "name": "app",
                                "image": image,
                                "ports": [{"containerPort": 80}]
                            }],
                        },
                    },
                },
            })        

Dynamic resource requests

Functions can also request resources during runtime through the RunFunctionResponse. Crossplane calls the function again with the requested resources:

apiVersion: apiextensions.crossplane.io/v1
kind: Composition
metadata:
  name: app-dynamic-config
spec:
  compositeTypeRef:
    apiVersion: example.crossplane.io/v1
    kind: App
  mode: Pipeline
  pipeline:
  - step: create-deployment-from-dynamic-config
    functionRef:
      name: crossplane-contrib-function-python
    input:
      apiVersion: python.fn.crossplane.io/v1beta1
      kind: Script
      script: |
        from crossplane.function import request, response
        
        def compose(req, rsp):
            observed_xr = req.observed.composite.resource
            
            # Always request the ConfigMap to ensure stable requirements
            config_name = observed_xr["spec"].get("configName", "default-config")
            namespace = observed_xr["metadata"].get("namespace", "default")
            
            response.require_resources(
                rsp, 
                name="dynamic-config",
                api_version="v1",
                kind="ConfigMap",
                match_name=config_name,
                namespace=namespace
            )
            
            # Check if we have the required ConfigMap
            config_map = request.get_required_resource(req, "dynamic-config")
            
            if not config_map:
                # ConfigMap not found yet - Crossplane will call us again
                return
            
            # ConfigMap found - use the image data to create deployment
            image = config_map.get("data", {}).get("image", "nginx:latest")
            
            rsp.desired.resources["deployment"].resource.update({
                "apiVersion": "apps/v1",
                "kind": "Deployment",
                "metadata": {
                    "labels": {"example.crossplane.io/app": observed_xr["metadata"]["name"]},
                },
                "spec": {
                    "replicas": 2,
                    "selector": {"matchLabels": {"example.crossplane.io/app": observed_xr["metadata"]["name"]}},
                    "template": {
                        "metadata": {
                            "labels": {"example.crossplane.io/app": observed_xr["metadata"]["name"]},
                        },
                        "spec": {
                            "containers": [{
                                "name": "app",
                                "image": image,
                                "ports": [{"containerPort": 80}]
                            }],
                        },
                    },
                },
            })        

{{<hint "tip">}} Use bootstrap requirements when possible for better performance. Dynamic requests require more function calls and work best when the required resources depend on the observed state or earlier function results. {{}}

Functions can request resources by:

• Name: name: "my-configmap" for a specific resource
• Labels: matchLabels: {"env": "prod"} for multiple resources
• Namespace: Include namespace: "production" for namespaced resources

Crossplane limits dynamic resource requests to 5 iterations to prevent infinite loops. The function signals completion by returning the same resource requirements two iterations in a row.

Function pipeline context

Sometimes two functions in a pipeline want to share information with each other that isn't desired state. Functions can use context for this. Any function can write to the pipeline context. Crossplane passes the context to all following functions. When Crossplane has called all functions it discards the pipeline context.

Function response cache

{{<hint "note" >}} Function response caching is an alpha feature. Enable it by setting the --enable-function-response-cache feature flag. {{< /hint >}}

Crossplane can cache function responses to improve performance by reducing repeated function calls. When enabled, Crossplane caches responses from composition functions that include a time to live (TTL) value.

The cache works by:

• Storing function responses on disk based on a hash of the request
• Only caching responses with a nonzero TTL
• Automatically removing expired cache entries
• Reusing cached responses for identical requests until they expire

This feature helps functions that:

• Perform expensive computations or external API calls
• Return stable results for the same inputs
• Include appropriate TTL values in their responses

Cache configuration

Control the cache behavior with these Crossplane pod arguments:

• --xfn-cache-max-ttl - Maximum cache duration (default: 24 hours)

The cache stores files in the /cache/xfn/ directory in the Crossplane pod. For better performance, consider using an in-memory cache by mounting an emptyDir volume with medium: Memory.