@daocloud-proto/hydra 0.18.0-dev-3 → 0.18.0-dev-5

package/admin-api/maas/v1alpha1/maas.pb.ts

@@ -9,12 +9,29 @@ import * as fm from "../../../fetch.pb"
 import * as GoogleProtobufTimestamp from "../../../google/protobuf/timestamp.pb"
 import * as HydraAdmin_apiModelV1alpha1Model from "../../model/v1alpha1/model.pb"
 
+type Absent<T, K extends keyof T> = { [k in Exclude<keyof T, K>]?: undefined };
+type OneOf<T> =
+  | { [k in keyof T]?: undefined }
+  | (
+    keyof T extends infer K ?
+      (K extends string & keyof T ? { [k in K]: T[K] } & Absent<T, K>
+        : never)
+    : never);
+
 export enum MAASWorkspaceVisibilityScope {
   MAAS_WORKSPACE_VISIBILITY_SCOPE_UNSPECIFIED = "MAAS_WORKSPACE_VISIBILITY_SCOPE_UNSPECIFIED",
   MAAS_WORKSPACE_VISIBILITY_SCOPE_ALL = "MAAS_WORKSPACE_VISIBILITY_SCOPE_ALL",
   MAAS_WORKSPACE_VISIBILITY_SCOPE_SPECIFIED = "MAAS_WORKSPACE_VISIBILITY_SCOPE_SPECIFIED",
 }
 
+export enum MAASModelHealthStatus {
+  MAAS_MODEL_HEALTH_STATUS_UNSPECIFIED = "MAAS_MODEL_HEALTH_STATUS_UNSPECIFIED",
+  MAAS_MODEL_HEALTH_STATUS_UNKNOWN = "MAAS_MODEL_HEALTH_STATUS_UNKNOWN",
+  MAAS_MODEL_HEALTH_STATUS_HEALTHY = "MAAS_MODEL_HEALTH_STATUS_HEALTHY",
+  MAAS_MODEL_HEALTH_STATUS_DEGRADED = "MAAS_MODEL_HEALTH_STATUS_DEGRADED",
+  MAAS_MODEL_HEALTH_STATUS_FAILED = "MAAS_MODEL_HEALTH_STATUS_FAILED",
+}
+
 export enum LoadBalanceStrategy {
   LOAD_BALANCE_STRATEGY_UNSPECIFIED = "LOAD_BALANCE_STRATEGY_UNSPECIFIED",
   LOAD_BALANCE_STRATEGY_ROUND_ROBIN = "LOAD_BALANCE_STRATEGY_ROUND_ROBIN",
@@ -40,6 +57,62 @@ export enum MAASBackendStatus {
   MAAS_BACKEND_STATUS_FAILED = "MAAS_BACKEND_STATUS_FAILED",
 }
 
+export enum RateLimitMatchStrategy {
+  RATE_LIMIT_MATCH_STRATEGY_UNSPECIFIED = "RATE_LIMIT_MATCH_STRATEGY_UNSPECIFIED",
+  RATE_LIMIT_MATCH_STRATEGY_ANY = "RATE_LIMIT_MATCH_STRATEGY_ANY",
+  RATE_LIMIT_MATCH_STRATEGY_ALL = "RATE_LIMIT_MATCH_STRATEGY_ALL",
+}
+
+export enum RateLimitType {
+  RATE_LIMIT_TYPE_UNSPECIFIED = "RATE_LIMIT_TYPE_UNSPECIFIED",
+  RATE_LIMIT_TYPE_REQUEST = "RATE_LIMIT_TYPE_REQUEST",
+  RATE_LIMIT_TYPE_TOKEN = "RATE_LIMIT_TYPE_TOKEN",
+  RATE_LIMIT_TYPE_IMAGE = "RATE_LIMIT_TYPE_IMAGE",
+  RATE_LIMIT_TYPE_AUDIO = "RATE_LIMIT_TYPE_AUDIO",
+  RATE_LIMIT_TYPE_VIDEO = "RATE_LIMIT_TYPE_VIDEO",
+  RATE_LIMIT_TYPE_EXEMPTION = "RATE_LIMIT_TYPE_EXEMPTION",
+}
+
+export enum RateLimitDimension {
+  RATE_LIMIT_DIMENSION_UNSPECIFIED = "RATE_LIMIT_DIMENSION_UNSPECIFIED",
+  RATE_LIMIT_DIMENSION_WORKSPACE = "RATE_LIMIT_DIMENSION_WORKSPACE",
+  RATE_LIMIT_DIMENSION_USER = "RATE_LIMIT_DIMENSION_USER",
+  RATE_LIMIT_DIMENSION_API_KEY = "RATE_LIMIT_DIMENSION_API_KEY",
+  RATE_LIMIT_DIMENSION_ACCOUNT = "RATE_LIMIT_DIMENSION_ACCOUNT",
+  RATE_LIMIT_DIMENSION_SUB_ACCOUNT = "RATE_LIMIT_DIMENSION_SUB_ACCOUNT",
+}
+
+export enum RateLimitWindow {
+  RATE_LIMIT_WINDOW_UNSPECIFIED = "RATE_LIMIT_WINDOW_UNSPECIFIED",
+  RATE_LIMIT_WINDOW_SECOND = "RATE_LIMIT_WINDOW_SECOND",
+  RATE_LIMIT_WINDOW_MINUTE = "RATE_LIMIT_WINDOW_MINUTE",
+  RATE_LIMIT_WINDOW_HOUR = "RATE_LIMIT_WINDOW_HOUR",
+  RATE_LIMIT_WINDOW_DAY = "RATE_LIMIT_WINDOW_DAY",
+}
+
+export enum TokenRateLimitMode {
+  TOKEN_RATE_LIMIT_MODE_UNSPECIFIED = "TOKEN_RATE_LIMIT_MODE_UNSPECIFIED",
+  TOKEN_RATE_LIMIT_MODE_TOTAL = "TOKEN_RATE_LIMIT_MODE_TOTAL",
+  TOKEN_RATE_LIMIT_MODE_INPUT_OUTPUT = "TOKEN_RATE_LIMIT_MODE_INPUT_OUTPUT",
+}
+
+export enum ImageRateLimitMetric {
+  IMAGE_RATE_LIMIT_METRIC_UNSPECIFIED = "IMAGE_RATE_LIMIT_METRIC_UNSPECIFIED",
+  IMAGE_RATE_LIMIT_METRIC_OUTPUT_IMAGES = "IMAGE_RATE_LIMIT_METRIC_OUTPUT_IMAGES",
+}
+
+export enum AudioRateLimitMetric {
+  AUDIO_RATE_LIMIT_METRIC_UNSPECIFIED = "AUDIO_RATE_LIMIT_METRIC_UNSPECIFIED",
+  AUDIO_RATE_LIMIT_METRIC_INPUT_DURATION_MILLISECONDS = "AUDIO_RATE_LIMIT_METRIC_INPUT_DURATION_MILLISECONDS",
+  AUDIO_RATE_LIMIT_METRIC_OUTPUT_DURATION_MILLISECONDS = "AUDIO_RATE_LIMIT_METRIC_OUTPUT_DURATION_MILLISECONDS",
+}
+
+export enum VideoRateLimitMetric {
+  VIDEO_RATE_LIMIT_METRIC_UNSPECIFIED = "VIDEO_RATE_LIMIT_METRIC_UNSPECIFIED",
+  VIDEO_RATE_LIMIT_METRIC_INPUT_DURATION_MILLISECONDS = "VIDEO_RATE_LIMIT_METRIC_INPUT_DURATION_MILLISECONDS",
+  VIDEO_RATE_LIMIT_METRIC_OUTPUT_DURATION_MILLISECONDS = "VIDEO_RATE_LIMIT_METRIC_OUTPUT_DURATION_MILLISECONDS",
+}
+
 export enum MAASModelGatewayStatus {
   GATEWAY_STATUS_UNSPECIFIED = "GATEWAY_STATUS_UNSPECIFIED",
   GATEWAY_STATUS_HEALTHY = "GATEWAY_STATUS_HEALTHY",
@@ -80,6 +153,7 @@ export type MAASModel = {
   tokenWeight?: string
   workspaceVisibility?: MAASWorkspaceVisibility
   capabilityRoutes?: MAASCapabilityRoute[]
+  healthStatus?: MAASModelHealthStatus
 }
 
 export type LoadBalanceConfig = {
@@ -122,6 +196,8 @@ export type MAASCapabilityUpstream = {
   apiKey?: string
   status?: MAASBackendStatus
   capabilityPaths?: {[key: string]: string}
+  healthCheck?: HealthCheck
+  healthMessage?: string
 }
 
 export type UpstreamModel = {
@@ -129,6 +205,16 @@ export type UpstreamModel = {
   apiKey?: string
   endpoint?: string
   status?: UpstreamModelStatus
+  healthCheck?: HealthCheck
+  healthMessage?: string
+}
+
+export type HealthCheck = {
+  path?: string
+  interval?: number
+  timeout?: number
+  healthyThreshold?: number
+  unhealthyThreshold?: number
 }
 
 export type ListMAASModelRequest = {
@@ -159,6 +245,135 @@ export type GetMAASModelRequest = {
   modelId?: string
 }
 
+export type ModelRateLimitPolicy = {
+  name?: string
+  enabled?: boolean
+  modelName?: string[]
+  paths?: string[]
+  matchStrategy?: RateLimitMatchStrategy
+  rules?: ModelRateLimitPolicyRule[]
+  creationTimestamp?: GoogleProtobufTimestamp.Timestamp
+  updateTimestamp?: GoogleProtobufTimestamp.Timestamp
+  generation?: string
+}
+
+export type RateLimitTarget = {
+  selectors?: RateLimitSelector[]
+  dimension?: RateLimitDimension
+  values?: string[]
+}
+
+export type RateLimitSelector = {
+  dimension?: RateLimitDimension
+  values?: string[]
+}
+
+export type RequestRateLimit = {
+  limit?: string
+  window?: RateLimitWindow
+}
+
+
+type BaseTokenRateLimit = {
+  mode?: TokenRateLimitMode
+  window?: RateLimitWindow
+}
+
+export type TokenRateLimit = BaseTokenRateLimit
+  & OneOf<{ totalLimit: string }>
+  & OneOf<{ inputLimit: string }>
+  & OneOf<{ outputLimit: string }>
+
+export type ImageRateLimit = {
+  metric?: ImageRateLimitMetric
+  limit?: string
+  window?: RateLimitWindow
+}
+
+export type AudioRateLimit = {
+  metric?: AudioRateLimitMetric
+  limit?: string
+  window?: RateLimitWindow
+}
+
+export type VideoRateLimit = {
+  metric?: VideoRateLimitMetric
+  limit?: string
+  window?: RateLimitWindow
+}
+
+export type RateLimitExemption = {
+}
+
+export type ModelRateLimitPolicyRule = {
+  id?: string
+  type?: RateLimitType
+  target?: RateLimitTarget
+  independentQuota?: boolean
+  resolvedReplaces?: string[]
+  request?: RequestRateLimit
+  token?: TokenRateLimit
+  image?: ImageRateLimit
+  audio?: AudioRateLimit
+  video?: VideoRateLimit
+  exemption?: RateLimitExemption
+}
+
+export type ModelRateLimitPolicyRuleCount = {
+  type?: RateLimitType
+  count?: number
+}
+
+export type ModelRateLimitPolicyListItem = {
+  name?: string
+  modelName?: string[]
+  paths?: string[]
+  matchStrategy?: RateLimitMatchStrategy
+  ruleCounts?: ModelRateLimitPolicyRuleCount[]
+  enabled?: boolean
+  updateTimestamp?: GoogleProtobufTimestamp.Timestamp
+}
+
+export type CreateModelRateLimitPolicyRequest = {
+  name?: string
+  enabled?: boolean
+  modelName?: string[]
+  paths?: string[]
+  matchStrategy?: RateLimitMatchStrategy
+  rules?: ModelRateLimitPolicyRule[]
+}
+
+export type GetModelRateLimitPolicyRequest = {
+  name?: string
+}
+
+export type ListModelRateLimitPoliciesRequest = {
+  page?: HydraCommonCommon.Pagination
+}
+
+export type ListModelRateLimitPoliciesResponse = {
+  items?: ModelRateLimitPolicyListItem[]
+  page?: HydraCommonCommon.Pagination
+}
+
+export type UpdateModelRateLimitPolicyRequest = {
+  name?: string
+  enabled?: boolean
+  modelName?: string[]
+  paths?: string[]
+  matchStrategy?: RateLimitMatchStrategy
+  rules?: ModelRateLimitPolicyRule[]
+}
+
+export type UpdateModelRateLimitPolicyStatusRequest = {
+  name?: string
+  enable?: boolean
+}
+
+export type DeleteModelRateLimitPolicyRequest = {
+  name?: string
+}
+
 export class MAASService {
   static GetMAASModel(req: GetMAASModelRequest, initReq?: fm.InitReq): Promise<MAASModel> {
     return fm.fetchReq<GetMAASModelRequest, MAASModel>(`/apis/admin.hydra.io/v1alpha1/maas-models/${req["modelId"]}?${fm.renderURLSearchParams(req, ["modelId"])}`, {...initReq, method: "GET"})
@@ -175,4 +390,22 @@ export class MAASService {
   static UpdateMAASModelStatus(req: UpdateMAASModelStatusRequest, initReq?: fm.InitReq): Promise<MAASModel> {
     return fm.fetchReq<UpdateMAASModelStatusRequest, MAASModel>(`/apis/admin.hydra.io/v1alpha1/maas-models/${req["modelId"]}/status`, {...initReq, method: "PUT", body: JSON.stringify(req, fm.replacer)})
   }
+  static CreateModelRateLimitPolicy(req: CreateModelRateLimitPolicyRequest, initReq?: fm.InitReq): Promise<ModelRateLimitPolicy> {
+    return fm.fetchReq<CreateModelRateLimitPolicyRequest, ModelRateLimitPolicy>(`/apis/admin.hydra.io/v1alpha1/model-rate-limit-policies`, {...initReq, method: "POST", body: JSON.stringify(req, fm.replacer)})
+  }
+  static GetModelRateLimitPolicy(req: GetModelRateLimitPolicyRequest, initReq?: fm.InitReq): Promise<ModelRateLimitPolicy> {
+    return fm.fetchReq<GetModelRateLimitPolicyRequest, ModelRateLimitPolicy>(`/apis/admin.hydra.io/v1alpha1/model-rate-limit-policies/${req["name"]}?${fm.renderURLSearchParams(req, ["name"])}`, {...initReq, method: "GET"})
+  }
+  static ListModelRateLimitPolicies(req: ListModelRateLimitPoliciesRequest, initReq?: fm.InitReq): Promise<ListModelRateLimitPoliciesResponse> {
+    return fm.fetchReq<ListModelRateLimitPoliciesRequest, ListModelRateLimitPoliciesResponse>(`/apis/admin.hydra.io/v1alpha1/model-rate-limit-policies?${fm.renderURLSearchParams(req, [])}`, {...initReq, method: "GET"})
+  }
+  static UpdateModelRateLimitPolicy(req: UpdateModelRateLimitPolicyRequest, initReq?: fm.InitReq): Promise<ModelRateLimitPolicy> {
+    return fm.fetchReq<UpdateModelRateLimitPolicyRequest, ModelRateLimitPolicy>(`/apis/admin.hydra.io/v1alpha1/model-rate-limit-policies/${req["name"]}`, {...initReq, method: "PUT", body: JSON.stringify(req, fm.replacer)})
+  }
+  static UpdateModelRateLimitPolicyStatus(req: UpdateModelRateLimitPolicyStatusRequest, initReq?: fm.InitReq): Promise<ModelRateLimitPolicy> {
+    return fm.fetchReq<UpdateModelRateLimitPolicyStatusRequest, ModelRateLimitPolicy>(`/apis/admin.hydra.io/v1alpha1/model-rate-limit-policies/${req["name"]}/status`, {...initReq, method: "PUT", body: JSON.stringify(req, fm.replacer)})
+  }
+  static DeleteModelRateLimitPolicy(req: DeleteModelRateLimitPolicyRequest, initReq?: fm.InitReq): Promise<ModelRateLimitPolicy> {
+    return fm.fetchReq<DeleteModelRateLimitPolicyRequest, ModelRateLimitPolicy>(`/apis/admin.hydra.io/v1alpha1/model-rate-limit-policies/${req["name"]}`, {...initReq, method: "DELETE"})
+  }
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@daocloud-proto/hydra",
-  "version": "v0.18.0-dev-3",
+  "version": "v0.18.0-dev-5",
   "description": "",
   "author": "",
   "license": "ISC"