Crate k8s_openapi

source ·
Expand description

Bindings for the Kubernetes client API, generated from the OpenAPI spec.

Each supported version of Kubernetes is represented by a feature name (like v1_9). Only one such feature can be enabled at a time.

These docs have been generated with the v1_26 feature enabled. To see docs for one of the other supported versions, please generate the docs locally with cargo doc --features 'v1_<>'

Examples

Resources

This example creates an instance of api::core::v1::PodSpec with no other properties set, and pretty-prints it.

use k8s_openapi::api::core::v1 as api;

fn main() {
    let pod_spec: api::PodSpec = Default::default();
    println!("{pod_spec:#?}");
}

Client API

(This requires the api feature to be enabled. The feature is enabled by default. See “Crate features” below for more details.)

This example executes the api::core::v1::Pod::list API operation to list all pods inside a namespace. It demonstrates the common patterns implemented by all API operation functions in this crate:

  1. The API function has required parameters and optional parameters. All optional parameters are taken as a single struct with optional fields.

    Specifically for the api::core::v1::Pod::list operation, the namespace parameter is required and taken by the function itself, while other optional parameters like field_selector are fields of the ListOptional struct. An instance of this struct is taken as the last parameter of Pod::list. This struct impls Default so that you can just pass in Default::default() if you don’t want to specify values for any of the optional parameters.

    Some API operations have a single common type for optional parameters:

    • All create API take optional parameters using the CreateOptional struct.
    • All delete API take optional parameters using the DeleteOptional struct.
    • All list API take optional parameters using the ListOptional struct.
    • All patch API take optional parameters using the PatchOptional struct.
    • All replace API take optional parameters using the ReplaceOptional struct.
    • All watch API take optional parameters using the WatchOptional struct.
    • All delete-collection API take optional parameters using the DeleteOptional struct for delete options and the ListOptional struct for list options.

    Other API functions have their own Optional structs with fields corresponding to the specific parameters for those functions, such as api::core::v1::ReadPodLogOptional for api::core::v1::Pod::read_log

  2. The function returns an http::Request value with the URL path, query string, and request body filled out according to the parameters given to the function. The function does not execute this request. You can execute this http::Request using any HTTP client library you want to use. It does not matter whether you use a synchronous client like reqwest, or an asynchronous client like hyper, or a mock client that returns bytes read from a test file.

  3. For each API operation function, there is a corresponding response type. For Pod::list this is ListResponse<api::core::v1::Pod>. This is an enum with variants for each of the possible HTTP status codes that the operation can return, and contains the data that the API server would return corresponding to that status code. For example, the list-namespaced-pod operation returns a pod list with HTTP 200 OK, so one of the variants of that type is Ok(List<api::core::v1::Pod>)

  4. The response types impl the Response trait, which contains a single Response::try_from_parts function. This function takes an http::StatusCode and a &u8 byte buffer, and tries to parse the byte buffer as the response type. For example, if you executed the request and received an HTTP 200 OK response with some bytes, you could call <ListResponse<Pod> as Response>::try_from_parts(status_code, buf) and expect to get Ok(ListResponse::<Pod>::Ok(pod_list)) from it.

    Once again, this design ensures that the crate is not tied to a specific HTTP client library or interface. It does not matter how you execute the HTTP request, nor whether your library is synchronous or asynchronous, since every HTTP client library gives you a way to get the HTTP response status code and the bytes of the response body.

  5. The API operation function also returns another value next to the http::Request. This value is a function that takes an http::StatusCode and returns a ResponseBody<ListResponse<Pod>>. As mentioned above, Response::try_from_parts requires you to maintain a byte buffer for the response body. ResponseBody is a helper that maintains such a buffer internally. It provides an append_slice() function to append slices to this internal buffer, and a parse() function to parse the buffer as the expected type (ListResponse<Pod> in this case).

    It is not necessary to use the ResponseBody returned by the API operation function to parse the response. The ResponseBody::parse function is only a wrapper around the underlying Response::try_from_parts function, and handles growing and shrinking its inner buffer as necessary. It also helps ensure that the response body is parsed as the correct type for the operation, ListResponse<Pod> in this case, and not some other type. However, you can instead use your own byte buffer instead of the ResponseBody value and call ListResponse<Pod>::try_from_parts yourself.

  6. The response types are enums with variants corresponding to HTTP status codes. For example, the ListResponse<Pod>::Ok variant corresponds to the HTTP 200 response of the list-namespaced-pod API.

    Each response enum also has an Other variant, that is yielded when the response status code does not match any of the other variants. This variant has a Result<Option<serde_json::Value>, serde_json::Error> value.

    If the response body is empty, this value will be Ok(None).

    If the response body is not empty, this value will be an Ok(Some(value)) or Err(err) from attempting to parse that body as a serde_json::Value. If you expect the response body to be a specific JSON type such as apimachinery::pkg::apis::meta::v1::Status, you can use the serde_json::Value as a serde::Deserializer like let status = <Status as Deserialize>::deserialize(value)?;. On the other hand, if you expect the response body to not be a JSON value, then ignore the Err(err) and parse the raw bytes of the response into the appropriate type.

Also see the get_single_value and get_multiple_values functions in the k8s-openapi-tests directory in the repository for examples of how to use a synchronous client with this style of API.

// Re-export of the http crate since it's used in the public API
use k8s_openapi::http;

use k8s_openapi::api::core::v1 as api;

// Assume `execute` is some function that takes an `http::Request` and
// executes it synchronously or asynchronously to get a response. This is
// provided by your HTTP client library.
//
// Note that the `http::Request` values returned by API operation functions
// only have a URL path, query string and request body filled out. That is,
// they do *not* have a URL host. So the real `execute` implementation
// would first mutate the URL of the request to an absolute URL with
// the API server's authority, add authorization headers, etc before
// actually executing it.
fn execute(req: http::Request<Vec<u8>>) -> Response { unimplemented!(); }

fn main() -> Result<(), Box<dyn std::error::Error>> {
    // Create a `http::Request` to list all the pods in the
    // "kube-system" namespace.
    let (request, response_body) =
        api::Pod::list("kube-system", Default::default())?;

    // Execute the request and get a response.
    // If this is an asynchronous operation, you would await
    // or otherwise yield to the event loop here.
    let response = execute(request);

    // Got a status code from executing the request.
    let status_code: http::StatusCode = response.status_code();

    // Construct the `ResponseBody<ListResponse<Pod>>` using the
    // constructor returned by the API function.
    let mut response_body = response_body(status_code);

    // Buffer used for each read from the HTTP response.
    let mut buf = Box::new([0_u8; 4096]);

    let pod_list = loop {
        // Read some bytes from the HTTP response into the buffer.
        // If this is an asynchronous operation, you would await or
        // yield to the event loop here.
        let read = response.read_into(&mut *buf)?;

        // `buf` now contains some data read from the response. Append it
        // to the `ResponseBody` and try to parse it into
        // the response type.
        response_body.append_slice(&buf[..read]);
        let response = response_body.parse();
        match response {
            // Successful response (HTTP 200 and parsed successfully)
            Ok(k8s_openapi::ListResponse::Ok(pod_list)) =>
                break pod_list,

            // Some unexpected response
            // (not HTTP 200, but still parsed successfully)
            Ok(other) => return Err(format!(
                "expected Ok but got {status_code} {other:?}").into()),

            // Need more response data.
            // Read more bytes from the response into the `ResponseBody`
            Err(k8s_openapi::ResponseError::NeedMoreData) => continue,

            // Some other error, like the response body being
            // malformed JSON or invalid UTF-8.
            Err(err) => return Err(format!(
                "error: {status_code} {err:?}").into()),
        }
    };

    for pod in pod_list.items {
        println!("{pod:#?}",);
    }

    Ok(())
}

Crate features

  • This crate contains several v1_* features. Enabling one of the v1_* features selects which version of the Kubernetes API server this crate should target. For example, enabling the v1_23 feature means the crate will only contain the API exposed by Kubernetes 1.23. It will not expose API that were removed in 1.23 or earlier, nor any API added in 1.24 or later.

  • The crate also contains a feature named api. If this feature is disabled, the library will only contain the resource types like api::core::v1::Pod, and not the associated operation functions like api::core::v1::Pod::read . The Response and Optional types for the operation functions will also not be accessible.

    This feature is enabled by default, but can be disabled if your crate does not need the operation functions to save on compile time and resources.

One and only one of the v1_* features must be enabled at the same time, otherwise the crate will not compile. This ensures that all crates in the crate graph use the same types. If it was possible for one library crate to use api::core::v1::Pod corresponding to v1.50 and another to use the type corresponding to v1.51, an application would not be able to use the same Pod value with both.

Thus, it is recommended that only application crates must enable one of the v1_* features, corresponding to the version of Kubernetes that the application wants to support.

# For application crates

[dependencies]
k8s-openapi = { version = "...", features = ["v1_50"] }

If you’re writing a library crate, your crate must not enable any features of k8s-openapi directly. The choice of which feature to enable must be left to any application crates that use your library. This ensures that all k8s-openapi-using dependencies in that application crate’s dependency graph use the same set of k8s-openapi types and are interoperable.

If your library crate has tests or examples, you should also add a dev-dependency on k8s-openapi in addition to the direct dependency, and enable a version feature only for that dev-dependency.

# For library crates

[dependencies]
k8s-openapi = "..."

[dev-dependencies]
k8s-openapi = { version = "...", features = ["v1_50"] }

However, commands like cargo check and cargo doc do not build dev dependencies, so they will not enable the feature and will fail to build. There are two ways you can resolve this:

  1. Add a feature to your library that enables one of the k8s-openapi v1_* features, and then remember to enable this feature when running such commands.

    [features]
__check = ["k8s-openapi/v1_50"]

    $ cargo check --features __check

  2. Define the K8S_OPENAPI_ENABLED_VERSION env var when running such commands:

    $ K8S_OPENAPI_ENABLED_VERSION=1.50 cargo check

Conditional compilation

As the previous section explained, library crates must not enable any version features in their k8s-openapi dependency. However, your library crate may need to know about which version gets selected eventually.

For example:

  1. Your crate creates a service spec and wants to set the cluster IP. This field is only available in Kubernetes 1.20+, so you want your crate to fail to compile if a lower feature was enabled.

  2. Your crate creates a service spec and wants to set the cluster IP, but you want it to be skipped when compiling for older versions.

There are two ways for your crate to determine which feature of k8s-openapi is enabled:

  1. The k8s-openapi crate exports k8s_if_* macros, which either expand to their contents or don’t. See the docs of the macros for more details.

    With these macros, the two cases above would be solved like this:

    • ⓘ
      // The compile_error!() is only emitted if 1.20 or lower is selected.
k8s_openapi::k8s_if_le_1_20! {
    compile_error!("This crate requires the v1_21 (or higher) feature to be enabled on the k8s-openapi crate.");
}

...

let service_spec = k8s_openapi::api::core::v1::ServiceSpec {
    cluster_ips: ...,
    ...
};
    • ⓘ
      let mut service_spec = k8s_openapi::api::core::v1::ServiceSpec {
    ...
};

k8s_openapi::k8s_if_ge_1_20! {
    service_spec.cluster_ips = ...;
}

  2. The k8s-openapi crate emits the selected version number as metadata that your crate can read in a build script from the DEP_K8S_OPENAPI_*_VERSION env var.

    // Your crate's build.rs

fn main() {
    let k8s_openapi_version: u32 =
        std::env::vars_os()
        .find_map(|(key, value)| {
            let key = key.into_string().ok()?;
            if key.starts_with("DEP_K8S_OPENAPI_") && key.ends_with("_VERSION") {
                let value = value.into_string().ok()?;
                Some(value)
            }
            else {
                None
            }
        }).expect("DEP_K8S_OPENAPI_*_VERSION must have been set by k8s-openapi")
        .parse().expect("DEP_K8S_OPENAPI_*_VERSION is malformed");

    // k8s_openapi_version has the format 0x00_MM_NN_00.
    //
    // - MM is the major version.
    // - NN is the minor version.
    //
    // Thus, if the v1_20 feature was enabled, k8s_openapi_version would be 0x00_01_14_00

    // The build script can now do arbitrary things with the information.
    // For example, it could define custom cfgs:
    if k8s_openapi_version >= 0x00_01_14_00 {
        println!(r#"cargo:rustc-cfg=k8s_service_spec_supports_cluster_ips"#);
    }

    // or emit new source code files under OUT_DIR, or anything else a build script can do.
}

    With these cfgs, the two cases above would be solved like this:

    • ⓘ
      // The compile_error!() is only emitted if 1.19 or lower is selected.
#[cfg(not(k8s_service_spec_supports_cluster_ips))]
compile_error!("This crate requires the v1_20 (or higher) feature to be enabled on the k8s-openapi crate.");

...

let service_spec = k8s_openapi::api::core::v1::ServiceSpec {
    cluster_ips: ...,
    ...
};
    • ⓘ
      let service_spec = k8s_openapi::api::core::v1::ServiceSpec {
    #[cfg(not(k8s_service_spec_supports_cluster_ips))]
    cluster_ips: ...,
    ...
};

Note that both approaches require your crate to have a direct dependency on the k8s-openapi crate. Neither approach is available if your crate only has a transitive dependency on the k8s-openapi crate.

The macros approach is easier to use since it doesn’t require a build script.

The build script method lets you emit arbitrary cfgs, emit arbitrary source code, and generally gives you more options, at the cost of needing a build script. cfg()s can be used in places where macros cannot, such as how the second example above shows it being used on a single field in a struct literal.

Custom resource definitions

The k8s-openapi-derive crate provides a custom derive for generating clientsets for custom resources. See that crate’s docs for more information.

Re-exports

Modules

Macros

  • k8s_if_1_20
    This macro evaluates to its contents if the v1_20 feature is enabled, otherwise it evaluates to nothing.
  • k8s_if_1_21
    This macro evaluates to its contents if the v1_21 feature is enabled, otherwise it evaluates to nothing.
  • k8s_if_1_22
    This macro evaluates to its contents if the v1_22 feature is enabled, otherwise it evaluates to nothing.
  • k8s_if_1_23
    This macro evaluates to its contents if the v1_23 feature is enabled, otherwise it evaluates to nothing.
  • k8s_if_1_24
    This macro evaluates to its contents if the v1_24 feature is enabled, otherwise it evaluates to nothing.
  • k8s_if_1_25
    This macro evaluates to its contents if the v1_25 feature is enabled, otherwise it evaluates to nothing.
  • k8s_if_1_26
    This macro evaluates to its contents if the v1_26 feature is enabled, otherwise it evaluates to nothing.
  • k8s_if_ge_1_20
    This macro evaluates to its contents if the v1_20 or higher feature is enabled, otherwise it evaluates to nothing.
  • k8s_if_ge_1_21
    This macro evaluates to its contents if the v1_21 or higher feature is enabled, otherwise it evaluates to nothing.
  • k8s_if_ge_1_22
    This macro evaluates to its contents if the v1_22 or higher feature is enabled, otherwise it evaluates to nothing.
  • k8s_if_ge_1_23
    This macro evaluates to its contents if the v1_23 or higher feature is enabled, otherwise it evaluates to nothing.
  • k8s_if_ge_1_24
    This macro evaluates to its contents if the v1_24 or higher feature is enabled, otherwise it evaluates to nothing.
  • k8s_if_ge_1_25
    This macro evaluates to its contents if the v1_25 or higher feature is enabled, otherwise it evaluates to nothing.
  • k8s_if_ge_1_26
    This macro evaluates to its contents if the v1_26 or higher feature is enabled, otherwise it evaluates to nothing.
  • k8s_if_le_1_20
    This macro evaluates to its contents if the v1_20 or lower feature is enabled, otherwise it evaluates to nothing.
  • k8s_if_le_1_21
    This macro evaluates to its contents if the v1_21 or lower feature is enabled, otherwise it evaluates to nothing.
  • k8s_if_le_1_22
    This macro evaluates to its contents if the v1_22 or lower feature is enabled, otherwise it evaluates to nothing.
  • k8s_if_le_1_23
    This macro evaluates to its contents if the v1_23 or lower feature is enabled, otherwise it evaluates to nothing.
  • k8s_if_le_1_24
    This macro evaluates to its contents if the v1_24 or lower feature is enabled, otherwise it evaluates to nothing.
  • k8s_if_le_1_25
    This macro evaluates to its contents if the v1_25 or lower feature is enabled, otherwise it evaluates to nothing.
  • k8s_if_le_1_26
    This macro evaluates to its contents if the v1_26 or lower feature is enabled, otherwise it evaluates to nothing.
  • k8s_match
    A macro that emits a match expr with the given test expression and arms. The match arms can be annotated with the other conditional compilation macros in this crate so that they’re only emitted if the predicate is true.

Structs

Enums

Traits

  • DeepMerge
    A trait applies to types that support deep merging.
  • ListableResource
    A trait applied to all Kubernetes resources that can be part of a corresponding list.
  • Metadata
    A trait applied to all Kubernetes resources that have metadata.
  • Resource
    A trait applied to all Kubernetes resources.
  • ResourceScope
    The scope of a Resource.
  • Response
    A trait implemented by all response types corresponding to Kubernetes API functions.

Functions