Generate rust bindings from Kubernetes openapi

[sehz]

Kubernetes uses OpenAPI spec: https://github.com/kubernetes/kubernetes/tree/master/api/openapi-spec to define their api.

Generate rust k8-api bindings from the spec which implements Spec trait as in the example:

const SERVICE_API: Crd = Crd {
    group: "core",
    version: "v1",
    names: CrdNames {
        kind: "Service",
        plural: "services",
        singular: "service",
    },
};

#[derive(Deserialize, Serialize, Debug, PartialEq, Default, Clone)]
#[serde(rename_all = "camelCase",default)]
pub struct ServiceSpec {
    #[serde(rename = "clusterIP")]
    pub cluster_ip: String,
    #[serde(rename = "externalIPs")]
    pub external_ips: Vec<String>,
    #[serde(rename = "loadBalancerIP")]
    pub load_balancer_ip: Option<String>,
    pub r#type: Option<LoadBalancerType>,
    pub external_name: Option<String>,
    pub external_traffic_policy: Option<ExternalTrafficPolicy>,
    pub ports: Vec<ServicePort>,
    pub selector: Option<HashMap<String, String>>,
}

impl Spec for ServiceSpec {

    type Status = ServiceStatus;
    type Header = DefaultHeader;

    fn metadata() -> &'static Crd {
        &SERVICE_API
    }

[sehz]

@qrilka Interested in working on this?

[qrilka]

I need to take a better look to understand the details but could it help with #92 and #112 ? Is the idea to generate code with a macro or maybe somehow else?

[qrilka]

@vijaylaxmid did you have any progress with this? Maybe some ideas?

[sehz]

Yes, using procedure macro to generate would make sense. Something like:

#[kubeapi(group = "core",version ="1", status = ServiceStatus)]
pub struct ServiceSpec {
    #[serde(rename = "clusterIP")]
    pub cluster_ip: String,
    #[serde(rename = "externalIPs")]
    pub external_ips: Vec<String>,
    #[serde(rename = "loadBalancerIP")]
    pub load_balancer_ip: Option<String>,
    pub r#type: Option<LoadBalancerType>,
    pub external_name: Option<String>,
    pub external_traffic_policy: Option<ExternalTrafficPolicy>,
    pub ports: Vec<ServicePort>,
    pub selector: Option<HashMap<String, String>>,
}

in the kube-rs, similar derive: https://docs.rs/kube-derive/0.63.2/kube_derive/derive.CustomResource.html

[qrilka]

I don't have much experience with macros yes but the task looks interesting and looks helpful for fluvio-community/fluvio#1131 that I was originally looking into when starting with k8-api/fluvio
Not sure how much time this will take though :)

[qrilka]

@sehz how could proc macro approach use Open API spec? The only sensible option I see is to list derived specs and test their conformance to the spec (but that could require some extra Spec details to be available at least when tests are enabled)

[sehz]

procedure macro just generates boiler plate once trait is defined. You can take look at how to parse as in here: https://github.com/infinyon/fluvio/tree/master/crates/fluvio-protocol-derive

[qrilka]

Sure but I guess the mapping from OpenAPI spec to Spec in Rust code base is a manual thing, right?

[qrilka]

@sehz I was looking a bit into other things and read a bit more about proc macros and as for this ticket - what actually should a macro derive here?
When I look into https://github.com/infinyon/k8-api/blob/master/src/k8-types/src/core/service.rs then the only things like that seem to be:

• impl Spec for ServiceSpec (with a bit unclear logic in make_same to empty cluster_ip)
• default_store_spec!(ServiceSpec, ServiceStatus, "Service") which is already a macro

Do I miss something here?

[sehz]

default_store_spec implements StoreSpec which not part of scope. This is just about reducing boiling plate:

impl Spec for ServiceSpec {
    type Status = ServiceStatus;
    type Header = DefaultHeader;

    fn metadata() -> &'static Crd {
        &SERVICE_API
    }

    fn make_same(&mut self, other: &Self) {
        if other.cluster_ip.is_empty() {
            self.cluster_ip = "".to_owned();
        }
    }
}

[qrilka]

This doesn't seem to be a lot boilerplate but OK.
The only remaining question then is about this special handling for cluster_ip - is it needed only for string fields or maybe there's some more complicated logic?

[sehz]

Can implement field attributes like #[serde(default = "path")], so this can be something like:

#[kubeapi(group = "core",version ="1", status = ServiceStatus,makesame="same_service")]
...

[qrilka]

And what does "same_service" mean your example?
As for field attributes, those work OK as (parameterized) markers but it doesn't look like it could allow generic handling, e.g. in StatefulSetSpec I see a different definition:

    // statefulset doesnt' like to change volume claim template
    fn make_same(&mut self, other: &Self) {
        self.volume_claim_templates = other.volume_claim_templates.clone();
    }

So it looks like make_same isn't something that needs to be derived

[sehz]

impl Spec for ServiceSpec {
    type Status = ServiceStatus;
    type Header = DefaultHeader;

    fn metadata() -> &'static Crd {
        &SERVICE_API
    }

    fn make_same(&mut self, other: &Self) {
       make_same2(self,other)
    }
}

fn make_same2(sv1: &mut ServiceSpec,sv2: &mut ServiceSpec2) { 
    if sv2.cluster_ip.is_empty() {
            sv1.cluster_ip = "".to_owned();
        }
}