main.ts 7.82 KB
Newer Older
liuzhe-lz's avatar
liuzhe-lz committed
1
2
// Copyright (c) Microsoft Corporation.
// Licensed under the MIT license.
Deshui Yu's avatar
Deshui Yu committed
3
4
5
6
7

'use strict';

import { Container, Scope } from 'typescript-ioc';

8
import * as fs from 'fs';
9
import * as path from 'path';
10
import * as component from './common/component';
Deshui Yu's avatar
Deshui Yu committed
11
12
import { Database, DataStore } from './common/datastore';
import { setExperimentStartupInfo } from './common/experimentStartupInfo';
chicm-ms's avatar
chicm-ms committed
13
import { getLogger, Logger, logLevelNameMap } from './common/log';
SparkSnail's avatar
SparkSnail committed
14
import { Manager, ExperimentStartUpMode } from './common/manager';
15
import { ExperimentManager } from './common/experimentManager';
Deshui Yu's avatar
Deshui Yu committed
16
import { TrainingService } from './common/trainingService';
17
import { getLogDir, mkDirP, parseArg } from './common/utils';
Deshui Yu's avatar
Deshui Yu committed
18
19
20
import { NNIDataStore } from './core/nniDataStore';
import { NNIManager } from './core/nnimanager';
import { SqlDB } from './core/sqlDatabase';
21
import { NNIExperimentsManager } from './core/nniExperimentsManager';
22
import { NNIRestServer } from './rest_server/nniRestServer';
23
import { FrameworkControllerTrainingService } from './training_service/kubernetes/frameworkcontroller/frameworkcontrollerTrainingService';
24
import { AdlTrainingService } from './training_service/kubernetes/adl/adlTrainingService';
25
26
import { KubeflowTrainingService } from './training_service/kubernetes/kubeflow/kubeflowTrainingService';
import { LocalTrainingService } from './training_service/local/localTrainingService';
27
import { RouterTrainingService } from './training_service/reusable/routerTrainingService';
28
import { PAIYarnTrainingService } from './training_service/pai/paiYarn/paiYarnTrainingService';
George Cheng's avatar
George Cheng committed
29
import { DLTSTrainingService } from './training_service/dlts/dltsTrainingService';
Deshui Yu's avatar
Deshui Yu committed
30

31

32
function initStartupInfo(
33
    startExpMode: string, experimentId: string, basePort: number, platform: string,
34
    logDirectory: string, experimentLogLevel: string, readonly: boolean, dispatcherPipe: string): void {
SparkSnail's avatar
SparkSnail committed
35
    const createNew: boolean = (startExpMode === ExperimentStartUpMode.NEW);
36
    setExperimentStartupInfo(createNew, experimentId, basePort, platform, logDirectory, experimentLogLevel, readonly, dispatcherPipe);
Deshui Yu's avatar
Deshui Yu committed
37
38
}

39
async function initContainer(foreground: boolean, platformMode: string, logFileName?: string): Promise<void> {
40
41
    const routerPlatformMode = ['remote', 'pai', 'aml', 'heterogeneous'];
    if (routerPlatformMode.includes(platformMode)) {
42
        Container.bind(TrainingService)
43
            .to(RouterTrainingService)
44
45
            .scope(Scope.Singleton);
    } else if (platformMode === 'local') {
46
47
48
        Container.bind(TrainingService)
            .to(LocalTrainingService)
            .scope(Scope.Singleton);
49
    } else if (platformMode === 'paiYarn') {
50
        Container.bind(TrainingService)
51
            .to(PAIYarnTrainingService)
52
            .scope(Scope.Singleton);
53
    } else if (platformMode === 'kubeflow') {
54
55
56
        Container.bind(TrainingService)
            .to(KubeflowTrainingService)
            .scope(Scope.Singleton);
57
    } else if (platformMode === 'frameworkcontroller') {
58
59
60
        Container.bind(TrainingService)
            .to(FrameworkControllerTrainingService)
            .scope(Scope.Singleton);
George Cheng's avatar
George Cheng committed
61
62
63
64
    } else if (platformMode === 'dlts') {
        Container.bind(TrainingService)
            .to(DLTSTrainingService)
            .scope(Scope.Singleton);
65
    } else if (platformMode === 'adl') {
SparkSnail's avatar
SparkSnail committed
66
        Container.bind(TrainingService)
67
            .to(AdlTrainingService)
SparkSnail's avatar
SparkSnail committed
68
            .scope(Scope.Singleton);
69
    } else {
chicm-ms's avatar
chicm-ms committed
70
        throw new Error(`Error: unsupported mode: ${platformMode}`);
Deshui Yu's avatar
Deshui Yu committed
71
    }
72
73
74
75
76
77
78
79
80
    Container.bind(Manager)
        .to(NNIManager)
        .scope(Scope.Singleton);
    Container.bind(Database)
        .to(SqlDB)
        .scope(Scope.Singleton);
    Container.bind(DataStore)
        .to(NNIDataStore)
        .scope(Scope.Singleton);
81
82
83
    Container.bind(ExperimentManager)
        .to(NNIExperimentsManager)
        .scope(Scope.Singleton);
84
85
86
87
88
89
    const DEFAULT_LOGFILE: string = path.join(getLogDir(), 'nnimanager.log');
    if (foreground) {
        logFileName = undefined;
    } else if (logFileName === undefined) {
        logFileName = DEFAULT_LOGFILE;
    }
SparkSnail's avatar
SparkSnail committed
90
91
92
    Container.bind(Logger).provider({
        get: (): Logger => new Logger(logFileName)
    });
Deshui Yu's avatar
Deshui Yu committed
93
94
95
96
97
98
    const ds: DataStore = component.get(DataStore);

    await ds.init();
}

function usage(): void {
99
    console.info('usage: node main.js --port <port> --mode \
100
    <local/remote/pai/kubeflow/frameworkcontroller/paiYarn/aml/adl/heterogeneous> --start_mode <new/resume> --experiment_id <id> --foreground <true/false>');
Deshui Yu's avatar
Deshui Yu committed
101
102
103
}

const strPort: string = parseArg(['--port', '-p']);
goooxu's avatar
goooxu committed
104
105
106
if (!strPort || strPort.length === 0) {
    usage();
    process.exit(1);
Deshui Yu's avatar
Deshui Yu committed
107
108
}

109
110
111
112
113
114
115
116
const foregroundArg: string = parseArg(['--foreground', '-f']);
if (!('true' || 'false').includes(foregroundArg.toLowerCase())) {
    console.log(`FATAL: foreground property should only be true or false`);
    usage();
    process.exit(1);
}
const foreground: boolean = foregroundArg.toLowerCase() === 'true' ? true : false;

goooxu's avatar
goooxu committed
117
118
const port: number = parseInt(strPort, 10);

Deshui Yu's avatar
Deshui Yu committed
119
const mode: string = parseArg(['--mode', '-m']);
120
if (!['local', 'remote', 'pai', 'kubeflow', 'frameworkcontroller', 'paiYarn', 'dlts', 'aml', 'adl', 'heterogeneous'].includes(mode)) {
121
    console.log(`FATAL: unknown mode: ${mode}`);
Deshui Yu's avatar
Deshui Yu committed
122
123
124
125
126
    usage();
    process.exit(1);
}

const startMode: string = parseArg(['--start_mode', '-s']);
SparkSnail's avatar
SparkSnail committed
127
if (![ExperimentStartUpMode.NEW, ExperimentStartUpMode.RESUME].includes(startMode)) {
128
    console.log(`FATAL: unknown start_mode: ${startMode}`);
Deshui Yu's avatar
Deshui Yu committed
129
130
131
132
133
    usage();
    process.exit(1);
}

const experimentId: string = parseArg(['--experiment_id', '-id']);
134
if (experimentId.trim().length < 1) {
SparkSnail's avatar
SparkSnail committed
135
    console.log(`FATAL: cannot resume the experiment, invalid experiment_id: ${experimentId}`);
Deshui Yu's avatar
Deshui Yu committed
136
137
138
139
    usage();
    process.exit(1);
}

140
141
142
143
144
145
146
147
const logDir: string = parseArg(['--log_dir', '-ld']);
if (logDir.length > 0) {
    if (!fs.existsSync(logDir)) {
        console.log(`FATAL: log_dir ${logDir} does not exist`);
    }
}

const logLevel: string = parseArg(['--log_level', '-ll']);
chicm-ms's avatar
chicm-ms committed
148
if (logLevel.length > 0 && !logLevelNameMap.has(logLevel)) {
149
150
151
    console.log(`FATAL: invalid log_level: ${logLevel}`);
}

SparkSnail's avatar
SparkSnail committed
152
153
154
155
156
157
158
159
const readonlyArg: string = parseArg(['--readonly', '-r']);
if (!('true' || 'false').includes(readonlyArg.toLowerCase())) {
    console.log(`FATAL: readonly property should only be true or false`);
    usage();
    process.exit(1);
}
const readonly = readonlyArg.toLowerCase() == 'true' ? true : false;

160
161
162
const dispatcherPipe: string = parseArg(['--dispatcher_pipe']);

initStartupInfo(startMode, experimentId, port, mode, logDir, logLevel, readonly, dispatcherPipe);
Deshui Yu's avatar
Deshui Yu committed
163

164
165
mkDirP(getLogDir())
    .then(async () => {
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
        try {
            await initContainer(foreground, mode);
            const restServer: NNIRestServer = component.get(NNIRestServer);
            await restServer.start();
            const log: Logger = getLogger();
            log.info(`Rest server listening on: ${restServer.endPoint}`);
        } catch (err) {
            const log: Logger = getLogger();
            log.error(`${err.stack}`);
            throw err;
        }
    })
    .catch((err: Error) => {
        console.error(`Failed to create log dir: ${err.stack}`);
    });
181

182
async function cleanUp(): Promise<void> {
183
    const log: Logger = getLogger();
SparkSnail's avatar
SparkSnail committed
184
    let hasError: boolean = false;
185
    try {
SparkSnail's avatar
SparkSnail committed
186
187
        const nniManager: Manager = component.get(Manager);
        await nniManager.stopExperiment();
188
189
        const experimentManager: ExperimentManager = component.get(ExperimentManager);
        await experimentManager.stop();
SparkSnail's avatar
SparkSnail committed
190
191
192
193
        const ds: DataStore = component.get(DataStore);
        await ds.close();
        const restServer: NNIRestServer = component.get(NNIRestServer);
        await restServer.stop();
194
    } catch (err) {
SparkSnail's avatar
SparkSnail committed
195
196
        hasError = true;
        log.error(`${err.stack}`);
197
    } finally {
198
        log.close();
199
        process.exit(hasError ? 1 : 0);
SparkSnail's avatar
SparkSnail committed
200
    }
201
202
203
204
205
}

process.on('SIGTERM', cleanUp);
process.on('SIGBREAK', cleanUp);
process.on('SIGINT', cleanUp);