|
| 1 | +/* |
| 2 | +Copyright 2025 The Kubernetes Authors. |
| 3 | +
|
| 4 | +Licensed under the Apache License, Version 2.0 (the "License"); |
| 5 | +you may not use this file except in compliance with the License. |
| 6 | +You may obtain a copy of the License at |
| 7 | +
|
| 8 | + http://www.apache.org/licenses/LICENSE-2.0 |
| 9 | +
|
| 10 | +Unless required by applicable law or agreed to in writing, software |
| 11 | +distributed under the License is distributed on an "AS IS" BASIS, |
| 12 | +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 13 | +See the License for the specific language governing permissions and |
| 14 | +limitations under the License. |
| 15 | +*/ |
| 16 | + |
| 17 | +package v1 |
| 18 | + |
| 19 | +import ( |
| 20 | + metav1 "k8s.io/apimachinery/pkg/apis/meta/v1" |
| 21 | +) |
| 22 | + |
| 23 | +// InferencePool is the Schema for the InferencePools API. |
| 24 | +// |
| 25 | +// +kubebuilder:object:root=true |
| 26 | +// TODO: change the annotation once it gets officially approved |
| 27 | +// +kubebuilder:metadata:annotations="api-approved.kubernetes.io=unapproved, experimental-only" |
| 28 | +// +kubebuilder:subresource:status |
| 29 | +// +kubebuilder:storageversion |
| 30 | +// +genclient |
| 31 | +type InferencePool struct { |
| 32 | + metav1.TypeMeta `json:",inline"` |
| 33 | + metav1.ObjectMeta `json:"metadata,omitempty"` |
| 34 | + |
| 35 | + Spec InferencePoolSpec `json:"spec,omitempty"` |
| 36 | + |
| 37 | + // Status defines the observed state of InferencePool. |
| 38 | + // |
| 39 | + // +kubebuilder:default={parent: {{parentRef: {kind: "Status", name: "default"}, conditions: {{type: "Accepted", status: "Unknown", reason: "Pending", message: "Waiting for controller", lastTransitionTime: "1970-01-01T00:00:00Z"}}}}} |
| 40 | + Status InferencePoolStatus `json:"status,omitempty"` |
| 41 | +} |
| 42 | + |
| 43 | +// InferencePoolList contains a list of InferencePool. |
| 44 | +// |
| 45 | +// +kubebuilder:object:root=true |
| 46 | +type InferencePoolList struct { |
| 47 | + metav1.TypeMeta `json:",inline"` |
| 48 | + metav1.ListMeta `json:"metadata,omitempty"` |
| 49 | + Items []InferencePool `json:"items"` |
| 50 | +} |
| 51 | + |
| 52 | +// InferencePoolSpec defines the desired state of InferencePool |
| 53 | +type InferencePoolSpec struct { |
| 54 | + // Selector defines a map of labels to watch model server Pods |
| 55 | + // that should be included in the InferencePool. |
| 56 | + // In some cases, implementations may translate this field to a Service selector, so this matches the simple |
| 57 | + // map used for Service selectors instead of the full Kubernetes LabelSelector type. |
| 58 | + // If specified, it will be applied to match the model server pods in the same namespace as the InferencePool. |
| 59 | + // Cross namesoace selector is not supported. |
| 60 | + // |
| 61 | + // +kubebuilder:validation:Required |
| 62 | + Selector map[LabelKey]LabelValue `json:"selector"` |
| 63 | + |
| 64 | + // TargetPortNumber defines the port number to access the selected model server Pods. |
| 65 | + // The number must be in the range 1 to 65535. |
| 66 | + // |
| 67 | + // +kubebuilder:validation:Minimum=1 |
| 68 | + // +kubebuilder:validation:Maximum=65535 |
| 69 | + // +kubebuilder:validation:Required |
| 70 | + TargetPortNumber int32 `json:"targetPortNumber"` |
| 71 | + |
| 72 | + // EndpointPickerConfig specifies the configuration needed by the proxy to discover and connect to the endpoint |
| 73 | + // picker service that picks endpoints for the requests routed to this pool. |
| 74 | + EndpointPickerConfig `json:",inline"` |
| 75 | +} |
| 76 | + |
| 77 | +// EndpointPickerConfig specifies the configuration needed by the proxy to discover and connect to the endpoint picker extension. |
| 78 | +// This type is intended to be a union of mutually exclusive configuration options that we may add in the future. |
| 79 | +type EndpointPickerConfig struct { |
| 80 | + // Extension configures an endpoint picker as an extension service. |
| 81 | + // |
| 82 | + // +kubebuilder:validation:Required |
| 83 | + ExtensionRef *Extension `json:"extensionRef,omitempty"` |
| 84 | +} |
| 85 | + |
| 86 | +// Extension specifies how to configure an extension that runs the endpoint picker. |
| 87 | +type Extension struct { |
| 88 | + // Reference is a reference to a service extension. When ExtensionReference is invalid, |
| 89 | + // a 5XX status code MUST be returned for the request that would have otherwise been routed |
| 90 | + // to the invalid backend. |
| 91 | + ExtensionReference `json:",inline"` |
| 92 | + |
| 93 | + // ExtensionConnection configures the connection between the Gateway and the extension. |
| 94 | + ExtensionConnection `json:",inline"` |
| 95 | +} |
| 96 | + |
| 97 | +// ExtensionReference is a reference to the extension. |
| 98 | +// |
| 99 | +// If a reference is invalid, the implementation MUST update the `ResolvedRefs` |
| 100 | +// Condition on the InferencePool's status to `status: False`. A 5XX status code MUST be returned |
| 101 | +// for the request that would have otherwise been routed to the invalid backend. |
| 102 | +type ExtensionReference struct { |
| 103 | + // Group is the group of the referent. |
| 104 | + // The default value is "", representing the Core API group. |
| 105 | + // |
| 106 | + // +optional |
| 107 | + // +kubebuilder:default="" |
| 108 | + Group *Group `json:"group,omitempty"` |
| 109 | + |
| 110 | + // Kind is the Kubernetes resource kind of the referent. |
| 111 | + // |
| 112 | + // Defaults to "Service" when not specified. |
| 113 | + // |
| 114 | + // ExternalName services can refer to CNAME DNS records that may live |
| 115 | + // outside of the cluster and as such are difficult to reason about in |
| 116 | + // terms of conformance. They also may not be safe to forward to (see |
| 117 | + // CVE-2021-25740 for more information). Implementations MUST NOT |
| 118 | + // support ExternalName Services. |
| 119 | + // |
| 120 | + // +optional |
| 121 | + // +kubebuilder:default=Service |
| 122 | + Kind *Kind `json:"kind,omitempty"` |
| 123 | + |
| 124 | + // Name is the name of the referent. |
| 125 | + // |
| 126 | + // +kubebuilder:validation:Required |
| 127 | + Name ObjectName `json:"name"` |
| 128 | + |
| 129 | + // The port number on the service running the extension. When unspecified, |
| 130 | + // implementations SHOULD infer a default value of 9002 when the Kind is |
| 131 | + // Service. |
| 132 | + // |
| 133 | + // +optional |
| 134 | + PortNumber *PortNumber `json:"portNumber,omitempty"` |
| 135 | +} |
| 136 | + |
| 137 | +// ExtensionConnection encapsulates options that configures the connection to the extension. |
| 138 | +type ExtensionConnection struct { |
| 139 | + // Configures how the gateway handles the case when the extension is not responsive. |
| 140 | + // Defaults to failClose. |
| 141 | + // |
| 142 | + // +optional |
| 143 | + // +kubebuilder:default="FailClose" |
| 144 | + FailureMode *ExtensionFailureMode `json:"failureMode"` |
| 145 | +} |
| 146 | + |
| 147 | +// ExtensionFailureMode defines the options for how the gateway handles the case when the extension is not |
| 148 | +// responsive. |
| 149 | +// +kubebuilder:validation:Enum=FailOpen;FailClose |
| 150 | +type ExtensionFailureMode string |
| 151 | + |
| 152 | +const ( |
| 153 | + // FailOpen specifies that the proxy should forward the request to an endpoint of its picking when the Endpoint Picker fails. |
| 154 | + FailOpen ExtensionFailureMode = "FailOpen" |
| 155 | + // FailClose specifies that the proxy should drop the request when the Endpoint Picker fails. |
| 156 | + FailClose ExtensionFailureMode = "FailClose" |
| 157 | +) |
| 158 | + |
| 159 | +// InferencePoolStatus defines the observed state of InferencePool. |
| 160 | +type InferencePoolStatus struct { |
| 161 | + // Parents is a list of parent resources (usually Gateways) that are |
| 162 | + // associated with the InferencePool, and the status of the InferencePool with respect to |
| 163 | + // each parent. |
| 164 | + // |
| 165 | + // A maximum of 32 Gateways will be represented in this list. When the list contains |
| 166 | + // `kind: Status, name: default`, it indicates that the InferencePool is not |
| 167 | + // associated with any Gateway and a controller must perform the following: |
| 168 | + // |
| 169 | + // - Remove the parent when setting the "Accepted" condition. |
| 170 | + // - Add the parent when the controller will no longer manage the InferencePool |
| 171 | + // and no other parents exist. |
| 172 | + // |
| 173 | + // +kubebuilder:validation:MaxItems=32 |
| 174 | + Parents []PoolStatus `json:"parent,omitempty"` |
| 175 | +} |
| 176 | + |
| 177 | +// PoolStatus defines the observed state of InferencePool from a Gateway. |
| 178 | +type PoolStatus struct { |
| 179 | + // GatewayRef indicates the gateway that observed state of InferencePool. |
| 180 | + GatewayRef ParentGatewayReference `json:"parentRef"` |
| 181 | + |
| 182 | + // Conditions track the state of the InferencePool. |
| 183 | + // |
| 184 | + // Known condition types are: |
| 185 | + // |
| 186 | + // * "Accepted" |
| 187 | + // * "ResolvedRefs" |
| 188 | + // |
| 189 | + // +optional |
| 190 | + // +listType=map |
| 191 | + // +listMapKey=type |
| 192 | + // +kubebuilder:validation:MaxItems=8 |
| 193 | + // +kubebuilder:default={{type: "Accepted", status: "Unknown", reason:"Pending", message:"Waiting for controller", lastTransitionTime: "1970-01-01T00:00:00Z"}} |
| 194 | + Conditions []metav1.Condition `json:"conditions,omitempty"` |
| 195 | +} |
| 196 | + |
| 197 | +// InferencePoolConditionType is a type of condition for the InferencePool |
| 198 | +type InferencePoolConditionType string |
| 199 | + |
| 200 | +// InferencePoolReason is the reason for a given InferencePoolConditionType |
| 201 | +type InferencePoolReason string |
| 202 | + |
| 203 | +const ( |
| 204 | + // This condition indicates whether the InferencePool has been accepted or rejected |
| 205 | + // by a Gateway, and why. |
| 206 | + // |
| 207 | + // Possible reasons for this condition to be True are: |
| 208 | + // |
| 209 | + // * "Accepted" |
| 210 | + // |
| 211 | + // Possible reasons for this condition to be False are: |
| 212 | + // |
| 213 | + // * "NotSupportedByGateway" |
| 214 | + // * "HTTPRouteNotAccepted" |
| 215 | + // |
| 216 | + // Possible reasons for this condition to be Unknown are: |
| 217 | + // |
| 218 | + // * "Pending" |
| 219 | + // |
| 220 | + // Controllers MAY raise this condition with other reasons, but should |
| 221 | + // prefer to use the reasons listed above to improve interoperability. |
| 222 | + InferencePoolConditionAccepted InferencePoolConditionType = "Accepted" |
| 223 | + |
| 224 | + // This reason is used with the "Accepted" condition when the InferencePool has been |
| 225 | + // accepted by the Gateway. |
| 226 | + InferencePoolReasonAccepted InferencePoolReason = "Accepted" |
| 227 | + |
| 228 | + // This reason is used with the "Accepted" condition when the InferencePool |
| 229 | + // has not been accepted by a Gateway because the Gateway does not support |
| 230 | + // InferencePool as a backend. |
| 231 | + InferencePoolReasonNotSupportedByGateway InferencePoolReason = "NotSupportedByGateway" |
| 232 | + |
| 233 | + // This reason is used with the "Accepted" condition when the InferencePool is |
| 234 | + // referenced by an HTTPRoute that has been rejected by the Gateway. The user |
| 235 | + // should inspect the status of the referring HTTPRoute for the specific reason. |
| 236 | + InferencePoolReasonHTTPRouteNotAccepted InferencePoolReason = "HTTPRouteNotAccepted" |
| 237 | + |
| 238 | + // This reason is used with the "Accepted" when a controller has not yet |
| 239 | + // reconciled the InferencePool. |
| 240 | + InferencePoolReasonPending InferencePoolReason = "Pending" |
| 241 | +) |
| 242 | + |
| 243 | +const ( |
| 244 | + // This condition indicates whether the controller was able to resolve all |
| 245 | + // the object references for the InferencePool. |
| 246 | + // |
| 247 | + // Possible reasons for this condition to be True are: |
| 248 | + // |
| 249 | + // * "ResolvedRefs" |
| 250 | + // |
| 251 | + // Possible reasons for this condition to be False are: |
| 252 | + // |
| 253 | + // * "InvalidExtensionRef" |
| 254 | + // |
| 255 | + // Controllers MAY raise this condition with other reasons, but should |
| 256 | + // prefer to use the reasons listed above to improve interoperability. |
| 257 | + InferencePoolConditionResolvedRefs InferencePoolConditionType = "ResolvedRefs" |
| 258 | + |
| 259 | + // This reason is used with the "ResolvedRefs" condition when the condition |
| 260 | + // is true. |
| 261 | + InferencePoolReasonResolvedRefs InferencePoolReason = "ResolvedRefs" |
| 262 | + |
| 263 | + // This reason is used with the "ResolvedRefs" condition when the |
| 264 | + // ExtensionRef is invalid in some way. This can include an unsupported kind |
| 265 | + // or API group, or a reference to a resource that can not be found. |
| 266 | + InferencePoolReasonInvalidExtensionRef InferencePoolReason = "InvalidExtensionRef" |
| 267 | +) |
| 268 | + |
| 269 | +// ParentGatewayReference identifies an API object including its namespace, |
| 270 | +// defaulting to Gateway. |
| 271 | +type ParentGatewayReference struct { |
| 272 | + // Group is the group of the referent. |
| 273 | + // |
| 274 | + // +optional |
| 275 | + // +kubebuilder:default="gateway.networking.k8s.io" |
| 276 | + Group *Group `json:"group"` |
| 277 | + |
| 278 | + // Kind is kind of the referent. For example "Gateway". |
| 279 | + // |
| 280 | + // +optional |
| 281 | + // +kubebuilder:default=Gateway |
| 282 | + Kind *Kind `json:"kind"` |
| 283 | + |
| 284 | + // Name is the name of the referent. |
| 285 | + Name ObjectName `json:"name"` |
| 286 | + |
| 287 | + // Namespace is the namespace of the referent. If not present, |
| 288 | + // the namespace of the referent is assumed to be the same as |
| 289 | + // the namespace of the referring object. |
| 290 | + // |
| 291 | + // +optional |
| 292 | + Namespace *Namespace `json:"namespace,omitempty"` |
| 293 | +} |
0 commit comments