Merge edf91f9 into 6903b18

microsoft · Oct 15, 2020 · 87457cf · 87457cf
2 parents 6903b18 + edf91f9
commit 87457cf
Show file tree

Hide file tree

Showing 8 changed files with 781 additions and 98 deletions.
diff --git a/docs/rest-api.md b/docs/rest-api.md
@@ -194,6 +194,12 @@ const openPAIClient = new PAIV2.OpenPAIClient(cluster);
     job = await openPAIClient.job.getJob(username, jobname);
     ```
 
+- [x] Get job status with specific attempt(GET /api/v2/jobs/{username}~{jobname}/attempts/{jobAttemptId})
+
+    ```ts
+    job = await openPAIClient.job.getJob(username, jobname, jobAttemptId);
+    ```
+
 - [x] Get job config (GET /api/v2/jobs/{username}~{jobname}/config)
 
     ```ts
@@ -218,6 +224,11 @@ const openPAIClient = new PAIV2.OpenPAIClient(cluster);
     ```ts
     await openPAIClient.job.deleteTag(username, jobname, tag);
     ```
+- [x] Get task status (GET /api/v2/jobs/{username}~{jobname}/attempts/{jobAttemptId}/taskRoles/{taskRoleName}/taskIndex/{taskIndex}/attempts)
+
+    ```ts
+    await openPAIClient.job.getTask(username, jobname, jobAttemptId, taskRoleName, taskIndex);
+    ```
 
 ## job history
 

diff --git a/src/api/v2/clients/jobClient.ts b/src/api/v2/clients/jobClient.ts
@@ -2,7 +2,7 @@
 // Licensed under the MIT License.
 
 import {
-    IJobConfig, IJobInfo, IJobSshInfo, IJobStatus, IPAICluster, IPAIResponse
+    IJobConfig, IJobInfo, IJobStatus, IPAICluster, IPAIResponse
 } from '@api/v2';
 import { Util } from '@pai/commom/util';
 import * as yaml from 'js-yaml';
@@ -55,12 +55,21 @@ export class JobClient extends OpenPAIBaseClient {
      * Get job status.
      * @param userName The user name.
      * @param jobName The job name.
+     * @param jobAttemptId The job attempt id
      */
-    public async getJob(userName: string, jobName: string): Promise<IJobStatus> {
-        const url: string = Util.fixUrl(
-            `${this.cluster.rest_server_uri}/api/v2/jobs/${userName}~${jobName}`,
-            this.cluster.https
-        );
+    public async getJob(userName: string, jobName: string, jobAttemptId?: number): Promise<IJobStatus> {
+        let url: string;
+        if (jobAttemptId !== undefined) {
+            url = Util.fixUrl(
+                `${this.cluster.rest_server_uri}/api/v2/jobs/${userName}~${jobName}/attempts/${jobAttemptId}`,
+                this.cluster.https
+            );
+        } else {
+            url = Util.fixUrl(
+                `${this.cluster.rest_server_uri}/api/v2/jobs/${userName}~${jobName}`,
+                this.cluster.https
+            );
+        }
         return await this.httpClient.get(url);
     }
 
@@ -123,6 +132,22 @@ export class JobClient extends OpenPAIBaseClient {
         return await this.httpClient.delete(url, undefined, { data: { value: tag } });
     }
 
+    /**
+     * Get task status.
+     * @param userName The user name.
+     * @param jobName The job name.
+     * @param jobAttemptId The job attempt id.
+     * @param taskRoleName The task role name.
+     * @param taskIndex The task index.
+     */
+    public async getTask(userName: string, jobName: string, jobAttemptId: number, taskRoleName: string, taskIndex: number)
+        : Promise<IJobStatus> {
+        const url: string = Util.fixUrl(
+            `${this.cluster.rest_server_uri}/api/v2/jobs/${userName}~${jobName}/attempts/${jobAttemptId}/taskRoles/${taskRoleName}/taskIndex/${taskIndex}/attempts`,
+            this.cluster.https
+        );
+        return await this.httpClient.get(url);
+    }
    /**
     * Get the events of a job.
     * @param userName The user name.

diff --git a/src/api/v2/index.ts b/src/api/v2/index.ts
@@ -20,7 +20,7 @@ import { StorageNodeV2 as StorageNode } from './clients/storageClient';
 import { IAuthnInfo, ILoginInfo } from './models/authn';
 import { IPAICluster, IPAIClusterInfo } from './models/cluster';
 import { IGroup } from './models/group';
-import { IJobAttempt, IJobFrameworkInfo, IJobInfo, IJobListQeury, IJobSshInfo, IJobStatus } from './models/job';
+import { IJobAttempt, IJobFrameworkInfo, IJobInfo, IJobListQeury, IJobSshInfo, IJobStatus, ITaskDetail } from './models/job';
 import { IPAIResponse } from './models/paiResponse';
 import { IMountInfo, IStorageConfig, IStorageDetail, IStorageServer, IStorageSummary } from './models/storage';
 import { IToken, ITokenList } from './models/token';
@@ -59,6 +59,7 @@ export {
     IStorageDetail,
     IMountInfo,
     IJobStatus,
+    ITaskDetail,
     IJobListQeury,
     StorageNode,
     IGroup,

diff --git a/src/api/v2/models/job.ts b/src/api/v2/models/job.ts
@@ -135,6 +135,56 @@ export interface IJobStatus {
     };
 }
 
+interface ITaskAttempt {
+    attemptId: number;
+    attemptState: string;
+    currentAttemptLaunchedTime: number | null;
+    currentAttemptCompletedTime: number | null;
+    containerId: string | null;
+    containerIp: string | null;
+    containerNodeName: string | null;
+    containerPorts: object;
+    containerGpus?: any;
+    containerLog: string;
+    containerExitCode: number | null;
+    containerExitSpec: {
+        code: number;
+        phrase: string;
+        issuer: string;
+        causer: string;
+        type: string;
+        stage: string;
+        behavior: string;
+        reaction: string;
+        reason: string;
+        repro: string[];
+        solution?: string[];
+    } ;
+    containerExitDiagnostics: string | null;
+    hived?: {
+        affinityGroupName?: any;
+        lazyPreempted?: any;
+        lazyPreemptionStatus?: any;
+    };
+}
+
+/**
+ * OpenPAI Task Detail.
+ */
+export interface ITaskDetail {
+    taskIndex: number;
+    taskUid: string;
+    taskState: string;
+    retries: number;
+    accountableRetries: number;
+    createdTime: number | null;
+    completedTime: number | null;
+    frameworkName: string;
+    jobAttemptId: number;
+    taskRoleName: string;
+    attempts: ITaskAttempt[];
+}
+
 /**
  * OpenPAI Job Framework Infomation.
  */