paiJobInfoCollector.ts 6.09 KB
Newer Older
liuzhe-lz's avatar
liuzhe-lz committed
1
2
// Copyright (c) Microsoft Corporation.
// Licensed under the MIT license.
3
4
5
6
7
8

'use strict';

import * as request from 'request';
import { Deferred } from 'ts-deferred';
import { NNIError, NNIErrorNames } from '../../common/errors';
9
import { getLogger, Logger } from '../../common/log';
10
import { TrialJobStatus } from '../../common/trainingService';
11
import { PAIClusterConfig, PAITrialJobDetail } from './paiConfig';
12
13
14
15
16

/**
 * Collector PAI jobs info from PAI cluster, and update pai job status locally
 */
export class PAIJobInfoCollector {
chicm-ms's avatar
chicm-ms committed
17
    private readonly trialJobsMap: Map<string, PAITrialJobDetail>;
18
    private readonly log: Logger = getLogger();
chicm-ms's avatar
chicm-ms committed
19
20
    private readonly statusesNeedToCheck: TrialJobStatus[];
    private readonly finalStatuses: TrialJobStatus[];
21
22
23
24

    constructor(jobMap: Map<string, PAITrialJobDetail>) {
        this.trialJobsMap = jobMap;
        this.statusesNeedToCheck = ['RUNNING', 'UNKNOWN', 'WAITING'];
QuanluZhang's avatar
QuanluZhang committed
25
        this.finalStatuses = ['SUCCEEDED', 'FAILED', 'USER_CANCELED', 'SYS_CANCELED', 'EARLY_STOPPED'];
26
27
    }

28
29
    public async retrieveTrialStatus(token? : string, paiBaseClusterConfig?: PAIClusterConfig): Promise<void> {
        if (paiBaseClusterConfig === undefined || token === undefined) {
30
            return Promise.resolve();
31
32
        }

chicm-ms's avatar
chicm-ms committed
33
        const updatePaiTrialJobs: Promise<void>[] = [];
34
35
        for (const [trialJobId, paiTrialJob] of this.trialJobsMap) {
            if (paiTrialJob === undefined) {
36
37
                throw new NNIError(NNIErrorNames.NOT_FOUND, `trial job id ${trialJobId} not found`);
            }
38
            updatePaiTrialJobs.push(this.getSinglePAITrialJobInfo(paiTrialJob, token, paiBaseClusterConfig));
39
40
41
42
43
        }

        await Promise.all(updatePaiTrialJobs);
    }

chicm-ms's avatar
chicm-ms committed
44
45
    private getSinglePAITrialJobInfo(paiTrialJob: PAITrialJobDetail, paiToken: string, paiClusterConfig: PAIClusterConfig): Promise<void> {
        const deferred: Deferred<void> = new Deferred<void>();
46
47
        if (!this.statusesNeedToCheck.includes(paiTrialJob.status)) {
            deferred.resolve();
48

49
50
51
52
53
54
            return deferred.promise;
        }

        // Rest call to get PAI job info and update status
        // Refer https://github.com/Microsoft/pai/blob/master/docs/rest-server/API.md for more detail about PAI Rest API
        const getJobInfoRequest: request.Options = {
SparkSnail's avatar
SparkSnail committed
55
            uri: `${paiClusterConfig.host}/rest-server/api/v1/user/${paiClusterConfig.userName}/jobs/${paiTrialJob.paiJobName}`,
56
57
            method: 'GET',
            json: true,
58
               headers: {
59
60
                'Content-Type': 'application/json',
                Authorization: `Bearer ${paiToken}`
61
62
            }
        };
63

64
        //TODO : pass in request timeout param?
65
        request(getJobInfoRequest, (error: Error, response: request.Response, body: any) => {
66
            if ((error !== undefined && error !== null) || response.statusCode >= 500) {
67
68
                this.log.error(`PAI Training service: get job info for trial ${paiTrialJob.id} from PAI Cluster failed!`);
                // Queried PAI job info failed, set job status to UNKNOWN
69
                if (paiTrialJob.status === 'WAITING' || paiTrialJob.status === 'RUNNING') {
70
71
72
                    paiTrialJob.status = 'UNKNOWN';
                }
            } else {
73
74
                if (response.body.jobStatus && response.body.jobStatus.state) {
                    switch (response.body.jobStatus.state) {
75
                        case 'WAITING':
76
77
78
79
                            paiTrialJob.status = 'WAITING';
                            break;
                        case 'RUNNING':
                            paiTrialJob.status = 'RUNNING';
80
                            if (paiTrialJob.startTime === undefined) {
81
82
                                paiTrialJob.startTime = response.body.jobStatus.appLaunchedTime;
                            }
83
                            if (paiTrialJob.url === undefined) {
84
                                paiTrialJob.url = response.body.jobStatus.appTrackingUrl;
85
86
87
88
89
90
                            }
                            break;
                        case 'SUCCEEDED':
                            paiTrialJob.status = 'SUCCEEDED';
                            break;
                        case 'STOPPED':
91
                            if (paiTrialJob.isEarlyStopped !== undefined) {
92
                                paiTrialJob.status = paiTrialJob.isEarlyStopped === true ?
93
94
                                        'EARLY_STOPPED' : 'USER_CANCELED';
                            } else {
95
96
97
                                /* if paiTrialJob's isEarlyStopped is undefined, that mean we didn't stop it via cancellation,
                                 * mark it as SYS_CANCELLED by PAI
                                 */
98
                                paiTrialJob.status = 'SYS_CANCELED';
QuanluZhang's avatar
QuanluZhang committed
99
                            }
100
101
                            break;
                        case 'FAILED':
102
                            paiTrialJob.status = 'FAILED';
103
104
105
106
107
                            break;
                        default:
                            paiTrialJob.status = 'UNKNOWN';
                    }
                    // For final job statues, update startTime, endTime and url
108
109
                    if (this.finalStatuses.includes(paiTrialJob.status)) {
                        if (paiTrialJob.startTime === undefined) {
110
111
                            paiTrialJob.startTime = response.body.jobStatus.appLaunchedTime;
                        }
112
                        if (paiTrialJob.endTime === undefined) {
113
114
115
                            paiTrialJob.endTime = response.body.jobStatus.completedTime;
                        }
                        // Set pai trial job's url to WebHDFS output path
116
117
118
119
120
121
                        if (paiTrialJob.logPath !== undefined) {
                            if (paiTrialJob.url) {
                                paiTrialJob.url += `,${paiTrialJob.logPath}`;
                            } else {
                                paiTrialJob.url = `${paiTrialJob.logPath}`;
                            }
122
123
124
125
126
127
128
129
130
                        }
                    }
                }
            }
            deferred.resolve();
        });

        return deferred.promise;
    }
QuanluZhang's avatar
QuanluZhang committed
131
}