@daocloud-proto/baize 0.127.0 → 0.127.2

package/management-api/job/v1alpha1/job.pb.ts

@@ -85,7 +85,6 @@ export type Job = {
   imagePullPolicy?: ImagePullPolicy
   suspendReason?: SuspendReason
   uid?: string
-  apiVersion?: BaizeCommonCommon.APIVersion
 }
 
 export type ListJobsRequest = {

package/management-api/job/v1alpha2/common.pb.ts

@@ -0,0 +1,82 @@
+/* eslint-disable */
+// @ts-nocheck
+/*
+* This file is a generated Typescript file for GRPC Gateway, DO NOT MODIFY
+*/
+
+import * as BaizeCommonK8s from "../../../common/k8s.pb"
+import * as BaizeManagement_apiImageV1alpha1Image from "../../image/v1alpha1/image.pb"
+
+export enum QueueType {
+  QUEUE_TYPE_UNSPECIFIED = "QUEUE_TYPE_UNSPECIFIED",
+  KUEUE = "KUEUE",
+}
+
+export enum PlacementPolicy {
+  PLACEMENT_POLICY_UNSPECIFIED = "PLACEMENT_POLICY_UNSPECIFIED",
+  BALANCED = "BALANCED",
+  GROUPED = "GROUPED",
+}
+
+export enum TrainingMode {
+  TRAINING_MODE_UNSPECIFIED = "TRAINING_MODE_UNSPECIFIED",
+  SINGLE = "SINGLE",
+  DISTRIBUTED = "DISTRIBUTED",
+}
+
+export enum JobPhase {
+  JOB_PHASE_UNSPECIFIED = "JOB_PHASE_UNSPECIFIED",
+  CREATED = "CREATED",
+  RUNNING = "RUNNING",
+  FAILED = "FAILED",
+  SUCCEEDED = "SUCCEEDED",
+  SUSPENDED = "SUSPENDED",
+}
+
+export enum JobRole {
+  JOB_ROLE_UNSPECIFIED = "JOB_ROLE_UNSPECIFIED",
+  MASTER = "MASTER",
+  WORKER = "WORKER",
+  TF_CHIEF = "TF_CHIEF",
+  TF_PS = "TF_PS",
+  TF_WORKER = "TF_WORKER",
+  TF_EVALUATOR = "TF_EVALUATOR",
+  PD_MASTER = "PD_MASTER",
+  PD_WORKER = "PD_WORKER",
+  MX_SCHEDULER = "MX_SCHEDULER",
+  MX_SERVER = "MX_SERVER",
+  MX_WORKER = "MX_WORKER",
+  MPI_LAUNCHER = "MPI_LAUNCHER",
+  MPI_WORKER = "MPI_WORKER",
+}
+
+export type CommonConfig = {
+  queueType?: QueueType
+  taskQueue?: string
+  description?: string
+}
+
+export type JobCreationBaseConfig = {
+  image?: string
+  imagePullSecret?: string
+  command?: string[]
+  args?: string[]
+  podConfig?: BaizeCommonK8s.PodConfig
+  labels?: {[key: string]: string}
+  annotations?: {[key: string]: string}
+  workingDir?: string
+  shmSize?: number
+  noOverrideEnvPath?: boolean
+  imageConfig?: BaizeManagement_apiImageV1alpha1Image.ImageConfig
+  preflight?: boolean
+  placement?: PlacementPolicy
+}
+
+export type JobRoleDifferenceConfig = {
+  replicas?: number
+  resources?: BaizeCommonK8s.Resources
+}
+
+export type RecoveryConfig = {
+  enable?: boolean
+}

package/management-api/job/v1alpha2/job.pb.ts

@@ -5,97 +5,208 @@
 */
 
 import * as BaizeCommonCommon from "../../../common/common.pb"
+import * as BaizeCommonK8s from "../../../common/k8s.pb"
 import * as fm from "../../../fetch.pb"
-import * as BaizeManagement_apiJobV1alpha1Common from "../v1alpha1/common.pb"
-import * as BaizeManagement_apiJobV1alpha1Job from "../v1alpha1/job.pb"
-export type Trainer = {
-  command?: string[]
-  args?: string[]
-  labels?: {[key: string]: string}
-  annotations?: {[key: string]: string}
-  queue?: string
-  trainingMode?: BaizeManagement_apiJobV1alpha1Common.TrainingMode
-  imagePullPolicy?: BaizeManagement_apiJobV1alpha1Job.ImagePullPolicy
-  checkpointConfig?: BaizeManagement_apiJobV1alpha1Job.CleanCheckpointConfig
-}
-
-export type RuntimeRef = {
-  name?: string
-  apiGroup?: string
-  kind?: string
+import * as GoogleProtobufStruct from "../../../google/protobuf/struct.pb"
+import * as GoogleProtobufTimestamp from "../../../google/protobuf/timestamp.pb"
+import * as BaizeManagement_apiAnalysisV1alpha1Analysis from "../../analysis/v1alpha1/analysis.pb"
+import * as BaizeManagement_apiEventV1alpha1Event from "../../event/v1alpha1/event.pb"
+import * as BaizeManagement_apiJobV1alpha2Common from "./common.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 JobType {
+  JOB_TYPE_UNSPECIFIED = "JOB_TYPE_UNSPECIFIED",
+  PYTORCH = "PYTORCH",
+  TENSORFLOW = "TENSORFLOW",
+  PADDLE = "PADDLE",
+  MPI = "MPI",
+  MXNET = "MXNET",
 }
 
-export type DatasetInitializer = {
+export enum RestartPolicy {
+  RESTART_POLICY_UNSPECIFIED = "RESTART_POLICY_UNSPECIFIED",
+  RESTART_POLICY_NEVER = "RESTART_POLICY_NEVER",
+  RESTART_POLICY_ON_FAILURE = "RESTART_POLICY_ON_FAILURE",
 }
 
-export type ModelInitializer = {
+export enum SuspendReason {
+  SUSPEND_REASON_UNSPECIFIED = "SUSPEND_REASON_UNSPECIFIED",
+  SUSPEND_REASON_USER_ACTION = "SUSPEND_REASON_USER_ACTION",
 }
 
-export type Initializer = {
-  model?: ModelInitializer
-  dataset?: DatasetInitializer
+export enum ImagePullPolicy {
+  IMAGE_PULL_POLICY_UNSPECIFIED = "IMAGE_PULL_POLICY_UNSPECIFIED",
+  IMAGE_PULL_POLICY_ALWAYS = "IMAGE_PULL_POLICY_ALWAYS",
+  IMAGE_PULL_POLICY_NEVER = "IMAGE_PULL_POLICY_NEVER",
+  IMAGE_PULL_POLICY_IF_NOT_PRESENT = "IMAGE_PULL_POLICY_IF_NOT_PRESENT",
 }
 
-export type CreateJobRequest = {
-  workspace?: number
+export enum JobActionRequestAction {
+  JOB_ACTION_UNSPECIFIED = "JOB_ACTION_UNSPECIFIED",
+  RESTART = "RESTART",
+  CHANGE_PRIORITY = "CHANGE_PRIORITY",
+  SUSPEND = "SUSPEND",
+  RESUME = "RESUME",
+}
+
+export enum JobSchedulersResponseSchedulerFeature {
+  FEATURE_UNSPECIFIED = "FEATURE_UNSPECIFIED",
+  BINPACK = "BINPACK",
+  GANG = "GANG",
+  FAIR = "FAIR",
+}
+
+export type Job = {
+  type?: JobType
   cluster?: string
   namespace?: string
   name?: string
-  runtimeRef?: RuntimeRef
-  initializer?: Initializer
-  trainer?: Trainer
-  suspend?: boolean
+  trainingMode?: BaizeManagement_apiJobV1alpha2Common.TrainingMode
+  creationTimestamp?: GoogleProtobufTimestamp.Timestamp
+  jobSpec?: GoogleProtobufStruct.Struct
   description?: string
+  baseConfig?: BaizeManagement_apiJobV1alpha2Common.JobCreationBaseConfig
+  roleConfig?: {[key: string]: BaizeManagement_apiJobV1alpha2Common.JobRoleDifferenceConfig}
+  phase?: BaizeManagement_apiJobV1alpha2Common.JobPhase
+  runningDuration?: number
+  totalResources?: BaizeCommonK8s.Resources
+  analysis?: BaizeManagement_apiAnalysisV1alpha1Analysis.AnalysisConfig
+  trainingConfig?: TrainingConfig
+  recoveryConfig?: BaizeManagement_apiJobV1alpha2Common.RecoveryConfig
+  cleanCheckpointConfig?: CleanCheckpointConfig
+  imagePullPolicy?: ImagePullPolicy
+  suspendReason?: SuspendReason
+  uid?: string
 }
 
-export type Job = {
+export type ListJobsRequest = {
+  workspace?: number
+  type?: JobType
+  cluster?: string
+  namespace?: string
+  page?: BaizeCommonCommon.Pagination
+}
+
+export type ListJobsResponse = {
+  items?: Job[]
+  page?: BaizeCommonCommon.Pagination
+}
+
+export type ListRdmasRequest = {
   workspace?: number
   cluster?: string
   namespace?: string
+  page?: BaizeCommonCommon.Pagination
+}
+
+export type Rdma = {
   name?: string
-  runtimeRef?: RuntimeRef
-  initializer?: Initializer
-  trainer?: Trainer
-  suspend?: boolean
-  description?: string
 }
 
-export type GetJobRequest = {
+export type ListRdmasResponse = {
+  items?: Rdma[]
+  page?: BaizeCommonCommon.Pagination
+}
+
+
+/* baize modified */ export type BaseSingleJobRequest = {
   workspace?: number
+  type?: JobType
   cluster?: string
   namespace?: string
   name?: string
 }
 
-export type DeleteJobRequest = {
+export type SingleJobRequest = BaseSingleJobRequest
+  & OneOf<{ deleteAnalysis: boolean }>
+
+export type CreateJobRequest = {
   workspace?: number
+  type?: JobType
   cluster?: string
   namespace?: string
   name?: string
+  description?: string
+  baseConfig?: BaizeManagement_apiJobV1alpha2Common.JobCreationBaseConfig
+  trainingMode?: BaizeManagement_apiJobV1alpha2Common.TrainingMode
+  roleConfig?: {[key: string]: BaizeManagement_apiJobV1alpha2Common.JobRoleDifferenceConfig}
+  analysis?: BaizeManagement_apiAnalysisV1alpha1Analysis.AnalysisConfig
+  trainingConfig?: TrainingConfig
+  recoveryConfig?: BaizeManagement_apiJobV1alpha2Common.RecoveryConfig
+  cleanCheckpointConfig?: CleanCheckpointConfig
+  imagePullPolicy?: ImagePullPolicy
 }
 
-export type ListRuntimesRequest = {
+export type CleanCheckpointConfig = {
+  cronExpr?: string
+  dirs?: string[]
+  retained?: number
+  timezone?: string
+  whitelist?: string
+}
+
+export type TrainingConfig = {
+  restartPolicy?: RestartPolicy
+  maxRetries?: number
+  maxTrainingDuration?: string
+  rdmaEnabled?: boolean
+}
+
+
+/* baize modified */ export type BaseJobActionRequestParams = {
+}
+
+export type JobActionRequestParams = BaseJobActionRequestParams
+  & OneOf<{ priorityClass: string }>
+
+export type JobActionRequest = {
   workspace?: number
+  type?: JobType
   cluster?: string
   namespace?: string
-  page?: BaizeCommonCommon.Pagination
+  name?: string
+  action?: JobActionRequestAction
+  params?: JobActionRequestParams
 }
 
-export type ListClusterRuntimesRequest = {
+export type JobSchedulersResponseScheduler = {
+  default?: boolean
+  name?: string
+  alias?: string
+  enabledFeatures?: JobSchedulersResponseSchedulerFeature[]
+}
+
+export type JobSchedulersResponse = {
+  items?: JobSchedulersResponseScheduler[]
+}
+
+export type ListJobEventsRequest = {
   workspace?: number
+  type?: JobType
   cluster?: string
+  namespace?: string
+  name?: string
   page?: BaizeCommonCommon.Pagination
 }
 
-export type ListJobsRequest = {
+export type ListRuntimesRequest = {
   workspace?: number
   cluster?: string
   namespace?: string
   page?: BaizeCommonCommon.Pagination
 }
 
-export type ListJobsResponse = {
-  items?: Job[]
+export type ListClusterRuntimesRequest = {
+  workspace?: number
+  cluster?: string
   page?: BaizeCommonCommon.Pagination
 }
 
@@ -153,17 +264,17 @@ export type CreateClusterRuntimeRequest = {
 }
 
 export class JobService {
-  static CreateJob(req: BaizeManagement_apiJobV1alpha1Job.CreateJobRequest, initReq?: fm.InitReq): Promise<BaizeManagement_apiJobV1alpha1Job.Job> {
-    return fm.fetchReq<BaizeManagement_apiJobV1alpha1Job.CreateJobRequest, BaizeManagement_apiJobV1alpha1Job.Job>(`/apis/baize.io/v1alpha2/workspaces/${req["workspace"]}/clusters/${req["cluster"]}/namespaces/${req["namespace"]}/jobs`, {...initReq, method: "POST", body: JSON.stringify(req, fm.replacer)})
+  static CreateJob(req: CreateJobRequest, initReq?: fm.InitReq): Promise<Job> {
+    return fm.fetchReq<CreateJobRequest, Job>(`/apis/baize.io/v1alpha2/workspaces/${req["workspace"]}/clusters/${req["cluster"]}/namespaces/${req["namespace"]}/jobs`, {...initReq, method: "POST", body: JSON.stringify(req, fm.replacer)})
   }
-  static GetJob(req: BaizeManagement_apiJobV1alpha1Job.SingleJobRequest, initReq?: fm.InitReq): Promise<BaizeManagement_apiJobV1alpha1Job.Job> {
-    return fm.fetchReq<BaizeManagement_apiJobV1alpha1Job.SingleJobRequest, BaizeManagement_apiJobV1alpha1Job.Job>(`/apis/baize.io/v1alpha2/workspaces/${req["workspace"]}/clusters/${req["cluster"]}/namespaces/${req["namespace"]}/jobs/${req["name"]}?${fm.renderURLSearchParams(req, ["workspace", "cluster", "namespace", "name"])}`, {...initReq, method: "GET"})
+  static GetJob(req: SingleJobRequest, initReq?: fm.InitReq): Promise<Job> {
+    return fm.fetchReq<SingleJobRequest, Job>(`/apis/baize.io/v1alpha2/workspaces/${req["workspace"]}/clusters/${req["cluster"]}/namespaces/${req["namespace"]}/jobs/${req["name"]}?${fm.renderURLSearchParams(req, ["workspace", "cluster", "namespace", "name"])}`, {...initReq, method: "GET"})
   }
-  static DeleteJob(req: BaizeManagement_apiJobV1alpha1Job.SingleJobRequest, initReq?: fm.InitReq): Promise<BaizeManagement_apiJobV1alpha1Job.Job> {
-    return fm.fetchReq<BaizeManagement_apiJobV1alpha1Job.SingleJobRequest, BaizeManagement_apiJobV1alpha1Job.Job>(`/apis/baize.io/v1alpha2/workspaces/${req["workspace"]}/clusters/${req["cluster"]}/namespaces/${req["namespace"]}/jobs/${req["name"]}`, {...initReq, method: "DELETE", body: JSON.stringify(req, fm.replacer)})
+  static DeleteJob(req: SingleJobRequest, initReq?: fm.InitReq): Promise<Job> {
+    return fm.fetchReq<SingleJobRequest, Job>(`/apis/baize.io/v1alpha2/workspaces/${req["workspace"]}/clusters/${req["cluster"]}/namespaces/${req["namespace"]}/jobs/${req["name"]}`, {...initReq, method: "DELETE", body: JSON.stringify(req, fm.replacer)})
   }
-  static ListJobs(req: BaizeManagement_apiJobV1alpha1Job.ListJobsRequest, initReq?: fm.InitReq): Promise<BaizeManagement_apiJobV1alpha1Job.ListJobsResponse> {
-    return fm.fetchReq<BaizeManagement_apiJobV1alpha1Job.ListJobsRequest, BaizeManagement_apiJobV1alpha1Job.ListJobsResponse>(`/apis/baize.io/v1alpha2/workspaces/${req["workspace"]}/clusters/${req["cluster"]}/namespaces/${req["namespace"]}/jobs?${fm.renderURLSearchParams(req, ["workspace", "cluster", "namespace"])}`, {...initReq, method: "GET"})
+  static ListJobs(req: ListJobsRequest, initReq?: fm.InitReq): Promise<ListJobsResponse> {
+    return fm.fetchReq<ListJobsRequest, ListJobsResponse>(`/apis/baize.io/v1alpha2/workspaces/${req["workspace"]}/jobs?${fm.renderURLSearchParams(req, ["workspace"])}`, {...initReq, method: "GET"})
   }
   static CreateRuntime(req: CreateRuntimeRequest, initReq?: fm.InitReq): Promise<Runtime> {
     return fm.fetchReq<CreateRuntimeRequest, Runtime>(`/apis/baize.io/v1alpha2/workspaces/${req["workspace"]}/clusters/${req["cluster"]}/namespaces/${req["namespace"]}/runtimes`, {...initReq, method: "POST", body: JSON.stringify(req, fm.replacer)})
@@ -189,7 +300,16 @@ export class JobService {
   static DeleteClusterRuntime(req: DeleteClusterRuntimeRequest, initReq?: fm.InitReq): Promise<Runtime> {
     return fm.fetchReq<DeleteClusterRuntimeRequest, Runtime>(`/apis/baize.io/v1alpha2/workspaces/${req["workspace"]}/clusters/${req["cluster"]}/clusterruntimes/${req["name"]}`, {...initReq, method: "DELETE", body: JSON.stringify(req, fm.replacer)})
   }
-  static DoJobAction(req: BaizeManagement_apiJobV1alpha1Job.JobActionRequest, initReq?: fm.InitReq): Promise<BaizeManagement_apiJobV1alpha1Job.Job> {
-    return fm.fetchReq<BaizeManagement_apiJobV1alpha1Job.JobActionRequest, BaizeManagement_apiJobV1alpha1Job.Job>(`/apis/baize.io/v1alpha2/workspaces/${req["workspace"]}/clusters/${req["cluster"]}/namespaces/${req["namespace"]}/jobs/${req["name"]}/actions`, {...initReq, method: "POST", body: JSON.stringify(req, fm.replacer)})
+  static DoJobAction(req: JobActionRequest, initReq?: fm.InitReq): Promise<Job> {
+    return fm.fetchReq<JobActionRequest, Job>(`/apis/baize.io/v1alpha2/workspaces/${req["workspace"]}/clusters/${req["cluster"]}/namespaces/${req["namespace"]}/jobs/${req["name"]}/actions`, {...initReq, method: "POST", body: JSON.stringify(req, fm.replacer)})
+  }
+  static ListSchedulers(req: CreateJobRequest, initReq?: fm.InitReq): Promise<JobSchedulersResponse> {
+    return fm.fetchReq<CreateJobRequest, JobSchedulersResponse>(`/apis/baize.io/v1alpha2/workspaces/${req["workspace"]}/clusters/${req["cluster"]}/namespaces/${req["namespace"]}/jobs/${req["name"]}/schedulers`, {...initReq, method: "PUT", body: JSON.stringify(req, fm.replacer)})
+  }
+  static ListRdmas(req: ListRdmasRequest, initReq?: fm.InitReq): Promise<ListRdmasResponse> {
+    return fm.fetchReq<ListRdmasRequest, ListRdmasResponse>(`/apis/baize.io/v1alpha2/workspaces/${req["workspace"]}/clusters/${req["cluster"]}/namespaces/${req["namespace"]}/rdmas?${fm.renderURLSearchParams(req, ["workspace", "cluster", "namespace"])}`, {...initReq, method: "GET"})
+  }
+  static ListJobEvents(req: ListJobEventsRequest, initReq?: fm.InitReq): Promise<BaizeManagement_apiEventV1alpha1Event.ListEventsResponse> {
+    return fm.fetchReq<ListJobEventsRequest, BaizeManagement_apiEventV1alpha1Event.ListEventsResponse>(`/apis/baize.io/v1alpha2/workspaces/${req["workspace"]}/clusters/${req["cluster"]}/namespaces/${req["namespace"]}/jobs/${req["name"]}/events?${fm.renderURLSearchParams(req, ["workspace", "cluster", "namespace", "name"])}`, {...initReq, method: "GET"})
   }
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@daocloud-proto/baize",
-  "version": "v0.127.0",
+  "version": "v0.127.2",
   "description": "",
   "author": "",
   "license": "ISC"