[Draft] Inference Extension support

gateway-api/src/apis/experimental/extension/inference/mod.rs

@@ -0,0 +1 @@
+// WARNING! generated file do not edit

gateway-api/src/apis/standard/extension/inference/constants.rs

@@ -0,0 +1,13 @@
+// WARNING: generated file - manual changes will be overriden
+
+#[derive(Debug, PartialEq, Eq)]
+pub enum InferencePoolEndpointPickerRefFailureMode {
+    FailOpen,
+    FailClose,
+}
+
+impl std::fmt::Display for InferencePoolEndpointPickerRefFailureMode {
+    fn fmt(&self, f: &mut std::fmt::Formatter) -> std::fmt::Result {
+        write!(f, "{:?}", self)
+    }
+}

gateway-api/src/apis/standard/extension/inference/enum_defaults.rs

@@ -0,0 +1,8 @@
+use super::common::*;
+// WARNING: generated file - manual changes will be overriden
+
+impl Default for InferencePoolEndpointPickerRefFailureMode {
+    fn default() -> Self {
+        InferencePoolEndpointPickerRefFailureMode::FailOpen
+    }
+}

gateway-api/src/apis/standard/extension/inference/inferencepools.rs

@@ -0,0 +1,184 @@
+// WARNING: generated by kopium - manual changes will be overwritten
+// kopium command: kopium --schema=derived --derive=JsonSchema --derive=Default --derive=PartialEq --docs -f -
+// kopium version: 0.22.5
+
+#[allow(unused_imports)]
+mod prelude {
+    pub use kube_derive::CustomResource;
+    pub use schemars::JsonSchema;
+    pub use serde::{Serialize, Deserialize};
+    pub use std::collections::BTreeMap;
+    pub use k8s_openapi::apimachinery::pkg::apis::meta::v1::Condition;
+}
+use self::prelude::*;
+
+/// Spec defines the desired state of the InferencePool.
+#[derive(CustomResource, Serialize, Deserialize, Clone, Debug, JsonSchema, Default, PartialEq)]
+#[kube(group = "inference.networking.k8s.io", version = "v1", kind = "InferencePool", plural = "inferencepools")]
+#[kube(namespaced)]
+#[kube(status = "InferencePoolStatus")]
+#[kube(derive="Default")]
+#[kube(derive="PartialEq")]
+pub struct InferencePoolSpec {
+    /// EndpointPickerRef is a reference to the Endpoint Picker extension and its
+    /// associated configuration.
+    #[serde(rename = "endpointPickerRef")]
+    pub endpoint_picker_ref: InferencePoolEndpointPickerRef,
+    /// Selector determines which Pods are members of this inference pool.
+    /// It matches Pods by their labels only within the same namespace; cross-namespace
+    /// selection is not supported.
+    /// 
+    /// The structure of this LabelSelector is intentionally simple to be compatible
+    /// with Kubernetes Service selectors, as some implementations may translate
+    /// this configuration into a Service resource.
+    pub selector: InferencePoolSelector,
+    /// TargetPorts defines a list of ports that are exposed by this InferencePool.
+    /// Currently, the list may only include a single port definition.
+    #[serde(rename = "targetPorts")]
+    pub target_ports: Vec<InferencePoolTargetPorts>,
+}
+
+/// EndpointPickerRef is a reference to the Endpoint Picker extension and its
+/// associated configuration.
+#[derive(Serialize, Deserialize, Clone, Debug, JsonSchema, Default, PartialEq)]
+pub struct InferencePoolEndpointPickerRef {
+    /// FailureMode configures how the parent handles the case when the Endpoint Picker extension
+    /// is non-responsive. When unspecified, defaults to "FailClose".
+    #[serde(default, skip_serializing_if = "Option::is_none", rename = "failureMode")]
+    pub failure_mode: Option<InferencePoolEndpointPickerRefFailureMode>,
+    /// Group is the group of the referent API object. When unspecified, the default value
+    /// is "", representing the Core API group.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub group: Option<String>,
+    /// Kind is the Kubernetes resource kind of the referent.
+    /// 
+    /// Required if the referent is ambiguous, e.g. service with multiple ports.
+    /// 
+    /// Defaults to "Service" when not specified.
+    /// 
+    /// ExternalName services can refer to CNAME DNS records that may live
+    /// outside of the cluster and as such are difficult to reason about in
+    /// terms of conformance. They also may not be safe to forward to (see
+    /// CVE-2021-25740 for more information). Implementations MUST NOT
+    /// support ExternalName Services.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub kind: Option<String>,
+    /// Name is the name of the referent API object.
+    pub name: String,
+    /// Port is the port of the Endpoint Picker extension service.
+    /// 
+    /// Port is required when the referent is a Kubernetes Service. In this
+    /// case, the port number is the service port number, not the target port.
+    /// For other resources, destination port might be derived from the referent
+    /// resource or this field.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub port: Option<InferencePoolEndpointPickerRefPort>,
+}
+
+/// EndpointPickerRef is a reference to the Endpoint Picker extension and its
+/// associated configuration.
+#[derive(Serialize, Deserialize, Clone, Debug, JsonSchema, PartialEq)]
+pub enum InferencePoolEndpointPickerRefFailureMode {
+    FailOpen,
+    FailClose,
+}
+
+/// Port is the port of the Endpoint Picker extension service.
+/// 
+/// Port is required when the referent is a Kubernetes Service. In this
+/// case, the port number is the service port number, not the target port.
+/// For other resources, destination port might be derived from the referent
+/// resource or this field.
+#[derive(Serialize, Deserialize, Clone, Debug, JsonSchema, Default, PartialEq)]
+pub struct InferencePoolEndpointPickerRefPort {
+    /// Number defines the port number to access the selected model server Pods.
+    /// The number must be in the range 1 to 65535.
+    pub number: i32,
+}
+
+/// Selector determines which Pods are members of this inference pool.
+/// It matches Pods by their labels only within the same namespace; cross-namespace
+/// selection is not supported.
+/// 
+/// The structure of this LabelSelector is intentionally simple to be compatible
+/// with Kubernetes Service selectors, as some implementations may translate
+/// this configuration into a Service resource.
+#[derive(Serialize, Deserialize, Clone, Debug, JsonSchema, Default, PartialEq)]
+pub struct InferencePoolSelector {
+    /// MatchLabels contains a set of required {key,value} pairs.
+    /// An object must match every label in this map to be selected.
+    /// The matching logic is an AND operation on all entries.
+    #[serde(rename = "matchLabels")]
+    pub match_labels: BTreeMap<String, String>,
+}
+
+/// Port defines the network port that will be exposed by this InferencePool.
+#[derive(Serialize, Deserialize, Clone, Debug, JsonSchema, Default, PartialEq)]
+pub struct InferencePoolTargetPorts {
+    /// Number defines the port number to access the selected model server Pods.
+    /// The number must be in the range 1 to 65535.
+    pub number: i32,
+}
+
+/// Status defines the observed state of the InferencePool.
+#[derive(Serialize, Deserialize, Clone, Debug, JsonSchema, Default, PartialEq)]
+pub struct InferencePoolStatus {
+    /// Parents is a list of parent resources, typically Gateways, that are associated with
+    /// the InferencePool, and the status of the InferencePool with respect to each parent.
+    /// 
+    /// A controller that manages the InferencePool, must add an entry for each parent it manages
+    /// and remove the parent entry when the controller no longer considers the InferencePool to
+    /// be associated with that parent.
+    /// 
+    /// A maximum of 32 parents will be represented in this list. When the list is empty,
+    /// it indicates that the InferencePool is not associated with any parents.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub parents: Option<Vec<InferencePoolStatusParents>>,
+}
+
+/// ParentStatus defines the observed state of InferencePool from a Parent, i.e. Gateway.
+#[derive(Serialize, Deserialize, Clone, Debug, JsonSchema, Default, PartialEq)]
+pub struct InferencePoolStatusParents {
+    /// Conditions is a list of status conditions that provide information about the observed
+    /// state of the InferencePool. This field is required to be set by the controller that
+    /// manages the InferencePool.
+    /// 
+    /// Supported condition types are:
+    /// 
+    /// * "Accepted"
+    /// * "ResolvedRefs"
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub conditions: Option<Vec<Condition>>,
+    /// ParentRef is used to identify the parent resource that this status
+    /// is associated with. It is used to match the InferencePool with the parent
+    /// resource, such as a Gateway.
+    #[serde(rename = "parentRef")]
+    pub parent_ref: InferencePoolStatusParentsParentRef,
+}
+
+/// ParentRef is used to identify the parent resource that this status
+/// is associated with. It is used to match the InferencePool with the parent
+/// resource, such as a Gateway.
+#[derive(Serialize, Deserialize, Clone, Debug, JsonSchema, Default, PartialEq)]
+pub struct InferencePoolStatusParentsParentRef {
+    /// Group is the group of the referent API object. When unspecified, the referent is assumed
+    /// to be in the "gateway.networking.k8s.io" API group.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub group: Option<String>,
+    /// Kind is the kind of the referent API object. When unspecified, the referent is assumed
+    /// to be a "Gateway" kind.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub kind: Option<String>,
+    /// Name is the name of the referent API object.
+    pub name: String,
+    /// Namespace is the namespace of the referenced object. When unspecified, the local
+    /// namespace is inferred.
+    /// 
+    /// Note that when a namespace different than the local namespace is specified,
+    /// a ReferenceGrant object is required in the referent namespace to allow that
+    /// namespace's owner to accept the reference. See the ReferenceGrant
+    /// documentation for details: <https://gateway-api.sigs.k8s.io/api-types/referencegrant/>
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub namespace: Option<String>,
+}
+

gateway-api/src/apis/standard/extension/inference/mod.rs

@@ -0,0 +1,4 @@
+// WARNING! generated file do not edit
+pub mod inferencepools;
+mod enum_defaults;
+pub mod constants;

type-reducer/extension/inference/standard_customized_mapped_names.txt

@@ -0,0 +1,3 @@
+#### Pass 1
+InferencePoolEndpointPickerRefPort->EndPointPort
+InferencePoolTargetPorts->EndPointPort

type-reducer/extension/inference/standard_reduced_types_pass_0.txt

@@ -0,0 +1 @@
+Condition

type-reducer/extension/inference/standard_reduced_types_pass_1.txt

@@ -0,0 +1,2 @@
+Condition
+EndPointPort

type-reducer/extension/inference/standard_rename_only_mapped_names.txt

@@ -0,0 +1,3 @@
+### Rename only
+InferencePoolStatusParents->InferencePoolStatusParent
+InferencePoolStatusParentsParentRef->ParentRef

update.sh

@@ -96,8 +96,8 @@ ENUMS=(
 ENUMS_WITH_DEFAULTS=$(printf ",%s" "${ENUMS[@]}")
 ENUMS_WITH_DEFAULTS=${ENUMS_WITH_DEFAULTS:1}
 
-# The task searches for $GATEWAY_API_ENUMS in the environment to get the enum names and their default variants.
-GATEWAY_API_ENUMS=${ENUMS_WITH_DEFAULTS} cargo xtask gen_enum_defaults >> $APIS_DIR/standard/enum_defaults.rs
+# The task searches for $API_ENUMS in the environment to get the enum names and their default variants.
+API_ENUMS=${ENUMS_WITH_DEFAULTS} cargo xtask gen_enum_defaults >> $APIS_DIR/standard/enum_defaults.rs
 echo "mod enum_defaults;" >> $APIS_DIR/standard/mod.rs
 
 GATEWAY_CLASS_CONDITION_CONSTANTS="GatewayClassConditionType=Accepted"
@@ -141,7 +141,7 @@ ENUMS=(
 
 ENUMS_WITH_DEFAULTS=$(printf ",%s" "${ENUMS[@]}")
 ENUMS_WITH_DEFAULTS=${ENUMS_WITH_DEFAULTS:1}
-GATEWAY_API_ENUMS=${ENUMS_WITH_DEFAULTS} cargo xtask gen_enum_defaults >> $APIS_DIR/experimental/enum_defaults.rs
+API_ENUMS=${ENUMS_WITH_DEFAULTS} cargo xtask gen_enum_defaults >> $APIS_DIR/experimental/enum_defaults.rs
 echo "mod enum_defaults;" >> $APIS_DIR/experimental/mod.rs
 
 # GatewayClass conditions vary between standard and experimental
@@ -189,7 +189,7 @@ ENUMS=(
 
 ENUMS_WITH_DEFAULTS=$(printf ",%s" "${ENUMS[@]}")
 ENUMS_WITH_DEFAULTS=${ENUMS_WITH_DEFAULTS:1}
-GATEWAY_API_ENUMS=${ENUMS_WITH_DEFAULTS} cargo xtask gen_enum_defaults > $APIS_DIR/standard/enum_defaults.rs
+API_ENUMS=${ENUMS_WITH_DEFAULTS} cargo xtask gen_enum_defaults > $APIS_DIR/standard/enum_defaults.rs
 
 sed -i '/#\[kube(status = "GrpcRouteStatus")\]/c\#\[kube(status = "RouteStatus")\]' $APIS_DIR/standard/grpcroutes.rs
 sed -i '/#\[kube(status = "HttpRouteStatus")\]/c\#\[kube(status = "RouteStatus")\]' $APIS_DIR/standard/httproutes.rs
@@ -223,7 +223,7 @@ ENUMS=(
 
 ENUMS_WITH_DEFAULTS=$(printf ",%s" "${ENUMS[@]}")
 ENUMS_WITH_DEFAULTS=${ENUMS_WITH_DEFAULTS:1}
-GATEWAY_API_ENUMS=${ENUMS_WITH_DEFAULTS} cargo xtask gen_enum_defaults > $APIS_DIR/experimental/enum_defaults.rs
+API_ENUMS=${ENUMS_WITH_DEFAULTS} cargo xtask gen_enum_defaults > $APIS_DIR/experimental/enum_defaults.rs
 
 sed -i '/#\[kube(status = "GrpcRouteStatus")\]/c\#\[kube(status = "RouteStatus")\]' $APIS_DIR/experimental/grpcroutes.rs
 sed -i '/#\[kube(status = "HttpRouteStatus")\]/c\#\[kube(status = "RouteStatus")\]' $APIS_DIR/experimental/httproutes.rs
@@ -232,9 +232,9 @@ sed -i '/#\[kube(status = "UdpRouteStatus")\]/c\#\[kube(status = "RouteStatus")\
 sed -i '/#\[kube(status = "TcpRouteStatus")\]/c\#\[kube(status = "RouteStatus")\]' $APIS_DIR/experimental/tcproutes.rs
 
 cargo fmt
-echo "API Generation complete"
+echo "Gateway API Generation complete"
 
-echo "Cleaning up temporary files"
+echo "Gateway API Cleaning up temporary files"
 set -x
 rm -f standard_mapped_names_phase_*.txt
 rm -f standard_mapped_types_to_names_phase_*.txt
@@ -243,4 +243,93 @@ rm -f experimental_mapped_types_to_names_phase_*.txt
 rm -f mapped_names.txt
 rm -f mapped_types_to_names.txt
 set +x
-echo "Cleanup complete"
+echo "Gateway API Cleanup complete"
+
+
+# ------------------------------------------------------------------------------
+# Inference Extension
+# ------------------------------------------------------------------------------
+
+echo " **** Inference Extension Processing Starts **** "
+
+INFERENCE_EXT_VERSION="v1.0.2"
+echo "Using Inference Extension version ${INFERENCE_EXT_VERSION}"
+
+INFERENCE_EXT_STANDARD_APIS=(
+    inferencepools
+)
+
+INFERENCE_EXT_EXPERIMENTAL_APIS=(
+    inferencepools
+    inferenceobjectives
+)
+
+mkdir -p $APIS_DIR/standard/extension/inference
+mkdir -p $APIS_DIR/experimental/extension/inference
+
+
+echo "// WARNING! generated file do not edit" > $APIS_DIR/standard/extension/inference/mod.rs
+
+for API in "${INFERENCE_EXT_STANDARD_APIS[@]}"
+do
+    echo "generating inference extension standard api ${API}"
+    curl -sSL "https://raw.githubusercontent.com/kubernetes-sigs/gateway-api-inference-extension/${INFERENCE_EXT_VERSION}/config/crd/bases/inference.networking.k8s.io_${API}.yaml" | kopium --schema=derived --derive=JsonSchema --derive=Default --derive=PartialEq --docs -f - > $APIS_DIR/standard/extension/inference/${API}.rs
+    sed -i 's/pub use kube::CustomResource;/pub use kube_derive::CustomResource;/g' $APIS_DIR/standard/extension/inference/${API}.rs
+    echo "pub mod ${API};" >> $APIS_DIR/standard/extension/inference/mod.rs
+done
+
+# Standard API enums that need a Default trait impl along with their respective default variant.
+ENUMS=(
+    InferencePoolEndpointPickerRefFailureMode=FailOpen
+)
+
+# Create a comma separated string out of $ENUMS.
+ENUMS_WITH_DEFAULTS=$(printf ",%s" "${ENUMS[@]}")
+ENUMS_WITH_DEFAULTS=${ENUMS_WITH_DEFAULTS:1}
+
+# The task searches for $API_ENUMS in the environment to get the enum names and their default variants.
+API_ENUMS=${ENUMS_WITH_DEFAULTS} cargo xtask gen_enum_defaults >> $APIS_DIR/standard/extension/inference/enum_defaults.rs
+echo "mod enum_defaults;" >> $APIS_DIR/standard/extension/inference/mod.rs
+
+EXT_INFERENCE_FAILURE_MODE_CONSTANTS="InferencePoolEndpointPickerRefFailureMode=FailOpen,FailClose"
+
+EXT_INFERENCE_FAILURE_MODE_CONSTANTS=${EXT_INFERENCE_FAILURE_MODE_CONSTANTS} \
+    cargo xtask gen_condition_constants inference >> $APIS_DIR/standard/extension/inference/constants.rs
+echo "pub mod constants;" >> $APIS_DIR/standard/extension/inference/mod.rs
+
+echo "// WARNING! generated file do not edit" > $APIS_DIR/experimental/extension/inference/mod.rs
+
+for API in "${INFERENCE_EXT_EXPERIMENTAL_APIS[@]}"
+do
+    echo "generating inference extension experimental api ${API}"
+    curl -sSL "https://raw.githubusercontent.com/kubernetes-sigs/gateway-api-inference-extension/${INFERENCE_EXT_VERSION}/config/crd/bases/inference.networking.x.k8s.io_${API}.yaml" | kopium --schema=derived --derive=JsonSchema --derive=Default --derive=PartialEq --docs -f - > $APIS_DIR/experimental/extension/inference/${API}.rs
+    sed -i 's/pub use kube::CustomResource;/pub use kube_derive::CustomResource;/g' $APIS_DIR/experimental/extension/inference/${API}.rs
+    echo "pub mod ${API};" >> $APIS_DIR/experimental/extension/inference/mod.rs
+done
+
+
+export RUST_LOG=info
+
+mkdir -p extension/inference/
+
+echo " **** Starting Type Reducer - Collapsing Duplicative Types **** "
+echo " **** Type Reducer - PHASE 1 - First Pass ***** "
+cargo run --manifest-path type-reducer/Cargo.toml -- --apis-dir $APIS_DIR/standard/extension/inference --out-dir $APIS_DIR/standard/extension/inference reduce --previous-pass-derived-type-names ./type-reducer/extension/inference/standard_reduced_types_pass_0.txt --current-pass-substitute-names ./type-reducer/extension/inference/standard_customized_mapped_names.txt
+mv mapped_names.txt extension/inference/standard_extension_inference_mapped_names_phase_1.txt
+mv mapped_types_to_names.txt extension/inference/standard_extension_inference_mapped_types_to_names_phase_1.txt
+echo " **** PHASE 2 ***** "
+cargo run --manifest-path type-reducer/Cargo.toml -- --apis-dir $APIS_DIR/standard/extension/inference --out-dir $APIS_DIR/standard/extension/inference reduce --previous-pass-derived-type-names ./type-reducer/extension/inference/standard_reduced_types_pass_1.txt --current-pass-substitute-names ./type-reducer/extension/inference/standard_customized_mapped_names.txt
+mv mapped_names.txt extension/inference/standard_extension_inference_mapped_names_phase_2.txt
+mv mapped_types_to_names.txt extension/inference/standard_extension_inference_mapped_types_to_names_phase_2.txt
+
+echo " **** RENAMING PHASE ***** "
+cargo run --manifest-path type-reducer/Cargo.toml -- --apis-dir $APIS_DIR/standard/extension/inference --out-dir $APIS_DIR/standard/extension/inference rename --rename-only-substitute-names ./type-reducer/extension/inference/standard_rename_only_mapped_names.txt
+
+cargo fmt
+echo "Inference Extension API Generation complete"
+
+echo "Inference Extension API Cleaning up temporary files"
+set -x
+rm -rf extension/inference/
+set +x
+echo "Inference Extension API Cleanup complete"