Derive Macro kube::CustomResource

source ·
    // Attributes available to this derive:
Available on crate feature derive only.
Expand description

Re-exports from kube-derive A custom derive for kubernetes custom resource definitions.

This will generate a root object containing your spec and metadata. This root object will implement the kube::Resource trait so it can be used with kube::Api.

The generated type will also implement kube’s kube::CustomResourceExt trait to generate the crd and generate kube::core::ApiResource information for use with the dynamic api.


use serde::{Serialize, Deserialize};
use kube::core::{Resource, CustomResourceExt};
use kube_derive::CustomResource;
use schemars::JsonSchema;

#[derive(CustomResource, Clone, Debug, Deserialize, Serialize, JsonSchema)]
#[kube(group = "", version = "v1", kind = "Foo", namespaced)]
struct FooSpec {
    info: String,

println!("kind = {}", Foo::kind(&())); // impl kube::Resource
let f = Foo::new("foo-1", FooSpec {
    info: "informative info".into(),
println!("foo: {:?}", f); // debug print on root type
println!("crd: {}", serde_yaml::to_string(&Foo::crd()).unwrap()); // crd yaml

This example generates a struct Foo containing metadata, the spec, and optionally status. The root struct Foo can be used with the kube crate as an Api<Foo> object (FooSpec can not be used with Api).

 let foos: Api<Foo> = Api::default_namespaced(client.clone());
 let crds: Api<CustomResourceDefinition> = Api::all(client.clone());
 crds.patch("", &PatchParams::apply("myapp"), &Patch::Apply(Foo::crd())).await;

This example posts the generated ::crd to the CustomResourceDefinition API. After this has been accepted (few secs max), you can start using foos as a normal kube Api object. See the crd_ prefixed examples for details on this.

Required properties

#[kube(group = "mygroup.tld")]

Your cr api group. The part before the slash in the top level apiVersion key.

#[kube(version = "v1")]

Your cr api version. The part after the slash in the top level apiVersion key.

#[kube(kind = "Kind")]

Name of your kind, and implied default for your generated root type.

Optional #[kube] attributes

#[kube(singular = "nonstandard-singular")]

To specify the singular name. Defaults to lowercased .kind value.

#[kube(plural = "nonstandard-plural")]

To specify the plural name. Defaults to inferring from singular.


To specify that this is a namespaced resource rather than cluster level.

#[kube(root = "StructName")]

Customize the name of the generated root struct (defaults to .kind value).

#[kube(crates(kube_core = "::kube::core"))]

Customize the crate name the generated code will reach into (defaults to ::kube::core). Should be one of kube::core, kube_client::core or kube_core.

#[kube(crates(k8s_openapi = "::k8s_openapi"))]

Customize the crate name the generated code will use for k8s_openapi (defaults to ::k8s_openapi).

#[kube(crates(schemars = "::schemars"))]

Customize the crate name the generated code will use for schemars (defaults to ::schemars).

#[kube(crates(serde = "::serde"))]

Customize the crate name the generated code will use for serde (defaults to ::serde).

#[kube(crates(serde_json = "::serde_json"))]

Customize the crate name the generated code will use for serde_json (defaults to ::serde_json).

#[kube(status = "StatusStructName")]

Adds a status struct to the top level generated type and enables the status subresource in your crd.

#[kube(derive = "Trait")]

Adding #[kube(derive = "PartialEq")] is required if you want your generated top level type to be able to #[derive(PartialEq)]

#[kube(schema = "mode")]

Defines whether the JsonSchema of the top level generated type should be used when generating a CustomResourceDefinition.

Legal values:

  • "derived": A JsonSchema implementation is automatically derived
  • "manual": JsonSchema is not derived, but used when creating the CustomResourceDefinition object
  • "disabled": No JsonSchema is used

This can be used to provide a completely custom schema, or to interact with third-party custom resources where you are not responsible for installing the CustomResourceDefinition.

Defaults to "derived".

NOTE: CustomResourceDefinitions require a schema. If schema = "disabled" then Self::crd() will not be installable into the cluster as-is.

#[kube(scale = r#"json"#)]

Allow customizing the scale struct for the scale subresource.

#[kube(printcolumn = r#"json"#)]

Allows adding straight json to printcolumns.

#[kube(shortname = "sn")]

Add a single shortname to the generated crd.

#[kube(category = "apps")]

Add a single category to crd.spec.names.categories.

Example with all properties

use serde::{Serialize, Deserialize};
use kube_derive::CustomResource;
use schemars::JsonSchema;

#[derive(CustomResource, Serialize, Deserialize, Debug, PartialEq, Clone, JsonSchema)]
    group = "",
    version = "v1",
    kind = "Foo",
    root = "FooCrd",
    status = "FooStatus",
    derive = "PartialEq",
    singular = "foot",
    plural = "feetz",
    shortname = "f",
    scale = r#"{"specReplicasPath":".spec.replicas", "statusReplicasPath":".status.replicas"}"#,
    printcolumn = r#"{"name":"Spec", "type":"string", "description":"name of foo", "jsonPath":""}"#
#[serde(rename_all = "camelCase")]
struct FooSpec {
    #[schemars(length(min = 3))]
    data: String,
    replicas_count: i32

#[derive(Serialize, Deserialize, Debug, PartialEq, Clone, JsonSchema)]
struct FooStatus {
    replicas: i32


Kubernetes requires that the generated schema is “structural”. This means that the structure of the schema must not depend on the particular values. For enums this imposes a few limitations:

  • Only externally tagged enums are supported
  • Unit variants may not be mixed with struct or tuple variants (enum Foo { Bar, Baz {}, Qux() } is invalid, for example)

If these restrictions are not followed then YourCrd::crd() may panic, or the Kubernetes API may reject the CRD definition.

Generated code

The example above will roughly generate:

#[derive(Serialize, Deserialize, Debug, PartialEq, Clone, JsonSchema)]
#[serde(rename_all = "camelCase")]
pub struct FooCrd {
    api_version: String,
    kind: String,
    metadata: ObjectMeta,
    spec: FooSpec,
    status: Option<FooStatus>,
impl kube::Resource for FooCrd { .. }

impl FooCrd {
    pub fn new(name: &str, spec: FooSpec) -> Self { .. }
    pub fn crd() -> CustomResourceDefinition { .. }

Customizing Schemas

Should you need to customize the schemas, you can use:

You might need to override parts of the schemas (for fields in question) when you are:

See kubernetes openapi validation for the format of the OpenAPI v3 schemas.

If you have to override a lot, you can opt-out of schema-generation entirely

Advanced Features

  • embedding k8s-openapi types can be done by enabling the schemars feature of k8s-openapi from 0.13.0
  • adding validation via validator crate is supported from schemars >= 0.8.5
  • generating rust code from schemas can be done via kopium and is supported on stable crds (> 1.16 kubernetes)

Schema Validation

There are two main ways of doing validation; server-side (embedding validation attributes into the schema for the apiserver to respect), and client-side (provides validate() methods in your code).

Client side validation of structs can be achieved by hooking up #[garde] attributes in your struct and is a replacement of the now unmaintained validator crate. Server-side validation require mutation of your generated schema, and can in the basic cases be achieved through the use of schemars’s validation attributes. For complete control, parts of the schema can be overridden to support more advanced Kubernetes specific validation rules.

When using garde directly, you must add it to your dependencies (with the derive feature).

Validation Caveats

Make sure your validation rules are static and handled by schemars:

  • validations from #[garde(custom(my_func))] will not show up in the schema.
  • similarly; nested / must_match / credit_card were unhandled by schemars at time of writing
  • encoding validations specified through garde (i.e. #[garde(ascii)]), are currently not supported by schemars
  • to validate required attributes client-side, garde requires a custom validation function (#[garde(custom(my_required_check))])
  • when using garde, fields that should not be validated need to be explictly skipped through the #[garde(skip)] attr

For sanity, you should review the generated schema before sending it to kubernetes.


Note that any changes to your struct / validation rules / serialization attributes will require you to re-apply the generated schema to kubernetes, so that the apiserver can validate against the right version of your structs.

Backwards compatibility between schema versions is recommended unless you are in a controlled environment where you can migrate manually. I.e. if you add new properties behind options, and simply mark old fields as deprecated, then you can safely roll schema out changes without bumping the version.

If you need multiple versions, then you need:

  • one module for each version of your types (e.g. v1::MyCrd and v2::MyCrd)
  • use the merge_crds fn to combine crds
  • roll out new schemas utilizing conversion webhooks / manual conversions / or allow kubectl to do its best

See the crd_derive_multi example to see how this upgrade flow works without special logic.

The upgrade flow with breaking changes involves:

  1. upgrade version marked as storage (from v1 to v2)
  2. read instances from the older Api<v1::MyCrd>
  3. perform conversion in memory and write them to the new Api<v2::MyCrd>.
  4. remove support for old version

If you need to maintain support for the old version for some time, then you have to repeat or continuously run steps 2 and 3. I.e. you probably need a conversion webhook.

NB: kube does currently not implement conversion webhooks yet.


Try cargo-expand to see your own macro expansion.


Enable the derive feature on the kube crate:

kube = { version = "...", features = ["derive"] }

Runtime dependencies

Due to rust-lang/rust#54363, we cannot be resilient against crate renames within our generated code. It’s therefore required that you have the following crates in scope, not renamed:

  • serde_json
  • k8s_openapi
  • schemars (by default, unless schema feature disabled)

You are ultimately responsible for maintaining the versions and feature flags of these libraries.