Merge pull request #65 from microsoft/zhiyuhe/event_api

add event api

src/api/v2/clients/jobClient.ts

@@ -7,7 +7,7 @@ import {
 import { Util } from '@pai/commom/util';
 import * as yaml from 'js-yaml';
 
-import { IJobListQeury } from '../models/job';
+import { IEventListQuery, IJobListQeury } from '../models/job';
 
 import { OpenPAIBaseClient } from './baseClient';
 
@@ -122,4 +122,18 @@ export class JobClient extends OpenPAIBaseClient {
         );
         return await this.httpClient.delete(url, undefined, { data: { value: tag } });
     }
+
+   /**
+    * Get the events of a job.
+    * @param userName The user name.
+    * @param jobName The job name.
+    * @param query filter jobs by event type
+    */
+  public async listEvents(userName: string, jobName: string, query?: IEventListQuery): Promise<any> {
+    const url: string = Util.fixUrl(
+      `${this.cluster.rest_server_uri}/api/v2/jobs/${userName}~${jobName}/events`,
+      this.cluster.https
+    );
+    return await this.httpClient.get(url, undefined, undefined, query);
+  }
 }

src/api/v2/models/job.ts

@@ -53,6 +53,10 @@ export interface IJobListQeury {
     withTotalCount?: boolean;
 }
 
+export interface IEventListQuery {
+    type?: 'Warning' | 'Normal';
+}
+
 export interface IAppExitSpec {
     code: number;
     phrase: string;

src/api/v2/swagger.yaml

@@ -9,10 +9,11 @@ info:
     Version 2.0.4: add default field in get storage list
     Version 2.0.5: add more parameters to job list; add submissionTime
     Version 2.1.0: add add/delete tag api; add tags field in get job detail and get job list; add tags filter in get job list
+    Version 2.1.1: add get event api
   license:
     name: MIT License
     url: "https://github.com/microsoft/pai/blob/master/LICENSE"
-  version: 2.1.0
+  version: 2.1.1
 externalDocs:
   description: Find out more about OpenPAI
   url: "https://github.com/microsoft/pai"
@@ -1397,6 +1398,49 @@ paths:
                   $ref: "#/components/responses/NoTagError/content/application~1json/examples/NoTagError"
         "500":
           $ref: "#/components/responses/UnknownError"
+  "/api/v2/jobs/{user}~{job}/events":
+    get:
+      tags:
+        - job
+      summary: Get events of a job.
+      description: Get events of a job. The events are sorted by "lastTimestamp,DESC".
+      operationId: getEvents
+      security:
+        - bearerAuth: []
+      parameters:
+        - name: type
+          in: query
+          description: filter events with type. Could be "Warning" or "Normal".
+          schema:
+            type: string
+      responses:
+        "200":
+          description: Succeeded
+          content:
+            application/json:
+              schema:
+                $ref: "#/components/schemas/JobEventsWithTotalCount"
+              example:
+                totalCount: 1
+                data:
+                - uid: "b8a87b12-29de-4702-818a-442d5a97623e"
+                  frameworkName: "399af21acb5befd015d4c45f6c7c1eb9"
+                  podUid: "94920b63-f4d6-4c63-b72b-b926b8bc1334"
+                  taskroleName: "taskrole"
+                  taskName: "399af21acb5befd015d4c45f6c7c1eb9-taskrole-11"
+                  taskIndex: 11
+                  type: "Warning"
+                  reason: "FailedScheduling"
+                  message: "0/17 nodes are available: 1 Insufficient memory, 1 node(s) didn't match node selector, 1 node(s) were unschedulable, 14 Insufficient cpu, 16 Insufficient nvidia.com/gpu."
+                  firstTimestamp: "2020-09-29T09:46:42.000Z"
+                  lastTimestamp: "2020-09-29T09:55:53.000Z"
+                  count: 9
+                  sourceComponent: "default-scheduler"
+                  sourceHost: null
+        "404":
+          $ref: "#/components/responses/NoJobError"
+        "500":
+          $ref: "#/components/responses/UnknownError"
   "/api/v2/jobs/{user}~{job}/job-attempts/healthz":
     get:
       tags:
@@ -1831,6 +1875,84 @@ components:
           - createdTime
           - completedTime
           - appExitCode
+    JobEventsWithTotalCount:
+      type: object
+      properties:
+        totalCount:
+          type: number
+          description: total count of events with given filters
+        data:
+          $ref: "#/components/schemas/JobEvents"
+      required:
+        - totalCount
+        - data
+    JobEvents:
+      type: array
+      description: job event list
+      items:
+        type: object
+        properties:
+          uid:
+            type: string
+            description: event uid
+          frameworkName:
+            type: string
+            description: related framework name
+          podUid:
+            type: string
+            description: pod uid of the event
+          taskroleName:
+            type: string
+            description: taskrole name of the event
+          taskName:
+            type: string
+            description: task name of the event
+          taskIndex:
+            type: integer
+            description: task index of the event
+          type:
+            type: string
+            description: type of the event
+            enum:
+              - Normal
+              - Warning
+          reason:
+            type: string
+            description: event reason
+          message:
+            type: string
+            description: event message
+          firstTimestamp:
+            type: string
+            description: firstTimestamp of the event
+          lastTimestamp:
+            type: string
+            description: lastTimestamp of the event
+          count:
+            type: integer
+            description: event count between firstTimestamp and lastTimestamp
+          sourceComponent:
+            type: string
+            description: source component of the event
+          sourceHost:
+            type: string
+            nullable: true
+            description: source host of the event
+        required:
+          - uid
+          - frameworkName
+          - podUid
+          - taskroleName
+          - taskName
+          - taskIndex
+          - type
+          - reason
+          - message
+          - firstTimestamp
+          - lastTimestamp
+          - count
+          - sourceComponent
+          - sourceHost
     JobDetail:
       type: object
       description: job details

tests/common/apiTestCases.ts

@@ -950,6 +950,16 @@ export const ApiDefaultTestCases: {[key: string]: IApiTestCase} = {
         }],
         after: [ updateTestJobExecutionType('STOP') ]
     },
+    'get /api/v2/jobs/{user}~{job}/events': {
+        tests: [
+            {
+                // We cannot predict cluster events in most time.
+                // So skip the test.
+                description: 'Skip',
+                customizedTest: 'skipTest'
+            }
+        ]
+    },
     'get /api/v2/jobs/{user}~{job}/job-attempts': {
         before: [ createTestJob() ],
         tests: [{
@@ -998,6 +1008,10 @@ export const ApiDefaultTestCases: {[key: string]: IApiTestCase} = {
 class CustomizedTestsClass {
     private readonly ajvInstance: Ajv = new ajv({ nullable: true });
 
+    public async skipTest(): Promise<void> {
+        return;
+    }
+
     public async getTokensWithUnauthorizedUser(
         test: IApiTestItem, operationResults?: IOperationResults
     ): Promise<void> {

tests/common/test_data/testEventList.ts

@@ -0,0 +1,27 @@
+// Copyright (c) Microsoft Corporation.
+// Licensed under the MIT License.
+
+/**
+ * Event list test data.
+ */
+export const testEventList: any = {
+    totalCount: 1,
+    data: [
+        {
+            uid: 'b8a87b12-29de-4702-818a-442d5a97623e',
+            frameworkName: '399af21acb5befd015d4c45f6c7c1eb9',
+            podUid: '94920b63-f4d6-4c63-b72b-b926b8bc1334',
+            taskroleName: 'taskrole',
+            taskName: '399af21acb5befd015d4c45f6c7c1eb9-taskrole-11',
+            taskIndex: 11,
+            type: 'Warning',
+            reason: 'FailedScheduling',
+            message: '0/17 nodes are available: 1 Insufficient memory, 1 node(s) didn\'t match node selector, 1 node(s) were unschedulable, 14 Insufficient cpu, 16 Insufficient nvidia.com/gpu.',
+            firstTimestamp: '2020-09-29T09:46:42.000Z',
+            lastTimestamp: '2020-09-29T09:55:53.000Z',
+            count: 9,
+            sourceComponent: 'default-scheduler',
+            sourceHost: null
+        }
+    ]
+};

tests/unit_tests/jobClient.spec.ts

@@ -11,6 +11,7 @@ import dirtyChai from 'dirty-chai';
 import * as yaml from 'js-yaml';
 import nock from 'nock';
 
+import { testEventList } from '../common/test_data/testEventList';
 import { testJobConfig, testJobConfigV1 } from '../common/test_data/testJobConfig';
 import { testJobList } from '../common/test_data/testJobList';
 import { testJobSshInfo } from '../common/test_data/testJobSshInfo';
@@ -186,3 +187,17 @@ describe('Delete a tag', () => {
         expect(result).to.be.eql(response);
     });
 });
+
+describe('List events', () => {
+    const response: any = testEventList;
+    const userName: string = 'core';
+    const jobName: string = 'tensorflow_serving_mnist_2019_6585ba19';
+    const queryString: string = 'type=Warning';
+    before(() => nock(`http://${testUri}`).get(`/api/v2/jobs/${userName}~${jobName}/events?${queryString}`).reply(200, response));
+
+    it('should return events', async () => {
+        const jobClient: JobClient = new JobClient(cluster);
+        const result: any = await jobClient.listEvents(userName, jobName, { type: 'Warning' });
+        expect(result).to.be.eql(response);
+    });
+});