Refactor close experiment implementation

src/nni_manager/common/log.ts

@@ -70,17 +70,23 @@ class Logger {
     private DEFAULT_LOGFILE: string = path.join(getLogDir(), 'nnimanager.log');
     private level: number = DEBUG;
     private bufferSerialEmitter: BufferSerialEmitter;
+    private writble: Writable;
 
     constructor(fileName?: string) {
         let logFile: string | undefined = fileName;
         if (logFile === undefined) {
             logFile = this.DEFAULT_LOGFILE;
         }
-        this.bufferSerialEmitter = new BufferSerialEmitter(fs.createWriteStream(logFile, {
+        this.writble = fs.createWriteStream(logFile, {
             flags: 'a+',
             encoding: 'utf8',
             autoClose: true
-        }));
+        });
+        this.bufferSerialEmitter = new BufferSerialEmitter(this.writble);
+    }
+
+    public close() {
+        this.writble.destroy();
     }
 
     public debug(...param: any[]): void {

src/nni_manager/core/nnimanager.ts

@@ -35,7 +35,7 @@ import {
 import {
     TrainingService, TrialJobApplicationForm, TrialJobDetail, TrialJobMetric, TrialJobStatus
 } from '../common/trainingService';
-import { delay , getLogDir, getMsgDispatcherCommand} from '../common/utils';
+import { delay, getLogDir, getMsgDispatcherCommand } from '../common/utils';
 import {
     ADD_CUSTOMIZED_TRIAL_JOB, KILL_TRIAL_JOB, NEW_TRIAL_JOB, NO_MORE_TRIAL_JOBS, REPORT_METRIC_DATA,
     REQUEST_TRIAL_JOBS, SEND_TRIAL_JOB_PARAMETER, TERMINATE, TRIAL_END, UPDATE_SEARCH_SPACE
@@ -123,7 +123,7 @@ class NNIManager implements Manager {
         this.log.debug('Setup tuner...');
 
         // Set up multiphase config
-        if(expParams.multiPhase && this.trainingService.isMultiPhaseJobSupported) {
+        if (expParams.multiPhase && this.trainingService.isMultiPhaseJobSupported) {
             this.trainingService.setClusterMetadata('multiPhase', expParams.multiPhase.toString());
         }
 
@@ -217,10 +217,9 @@ class NNIManager implements Manager {
         return this.dataStore.getTrialJobStatistics();
     }
 
-    public stopExperiment(): Promise<void> {
+    public async stopExperiment(): Promise<void> {
         this.status.status = 'STOPPING';
-
-        return Promise.resolve();
+        await this.experimentDoneCleanUp();
     }
 
     public async getMetricData(trialJobId?: string, metricType?: MetricType): Promise<MetricDataRecord[]> {
@@ -342,7 +341,7 @@ class NNIManager implements Manager {
 
     private async periodicallyUpdateExecDuration(): Promise<void> {
         let count: number = 1;
-        for (; ;) {
+        while (this.status.status !== 'STOPPING') {
             await delay(1000 * 1); // 1 seconds
             if (this.status.status === 'EXPERIMENT_RUNNING') {
                 this.experimentProfile.execDuration += 1;
@@ -396,10 +395,7 @@ class NNIManager implements Manager {
             throw new Error('Error: tuner has not been setup');
         }
         let allFinishedTrialJobNum: number = 0;
-        for (; ;) {
-            if (this.status.status === 'STOPPING') {
-                break;
-            }
+        while (this.status.status !== 'STOPPING') {
             const finishedTrialJobNum: number = await this.requestTrialJobsStatus();
 
             allFinishedTrialJobNum += finishedTrialJobNum;

src/nni_manager/main.ts

@@ -50,7 +50,7 @@ async function initContainer(platformMode: string): Promise<void> {
         Container.bind(TrainingService).to(LocalTrainingServiceForGPU).scope(Scope.Singleton);
     } else if (platformMode === 'remote') {
         Container.bind(TrainingService).to(RemoteMachineTrainingService).scope(Scope.Singleton);
-    } else if (platformMode === 'pai'){
+    } else if (platformMode === 'pai') {
         Container.bind(TrainingService).to(PAITrainingService).scope(Scope.Singleton);
     } else {
         throw new Error(`Error: unsupported mode: ${mode}`);
@@ -108,3 +108,12 @@ mkDirP(getLogDir()).then(async () => {
 }).catch((err: Error) => {
     console.error(`Failed to create log dir: ${err.stack}`);
 });
+
+process.on('SIGTERM', async () => {
+    const ds: DataStore = component.get(DataStore);
+    await ds.close();
+    const restServer: NNIRestServer = component.get(NNIRestServer);
+    await restServer.stop();
+    const log: Logger = getLogger();
+    log.close();
+})
\ No newline at end of file

src/nni_manager/rest_server/restHandler.ts

@@ -164,8 +164,6 @@ class NNIRestHandler {
                 await this.tb.cleanUp();
                 await this.nniManager.stopExperiment();
                 res.send();
-                this.log.debug('Stopping rest server');
-                await this.restServer.stop();
             } catch (err) {
                 this.handle_error(err, res);
             }

src/nni_manager/training_service/local/localTrainingService.ts

@@ -26,7 +26,7 @@ import { EventEmitter } from 'events';
 import * as fs from 'fs';
 import * as path from 'path';
 import * as ts from 'tail-stream';
-import { MethodNotImplementedError, NNIError, NNIErrorNames } from '../../common/errors';
+import { NNIError, NNIErrorNames } from '../../common/errors';
 import { getLogger, Logger } from '../../common/log';
 import { TrialConfig } from '../common/trialConfig';
 import { TrialConfigMetadataKey } from '../common/trialConfigMetadataKey';
@@ -103,6 +103,7 @@ class LocalTrainingService implements TrainingService {
     protected log: Logger;
     protected localTrailConfig?: TrialConfig;
     private isMultiPhase: boolean = false;
+    private streams: Array<ts.Stream>;
 
     constructor() {
         this.eventEmitter = new EventEmitter();
@@ -112,6 +113,7 @@ class LocalTrainingService implements TrainingService {
         this.stopping = false;
         this.log = getLogger();
         this.trialSequenceId = -1;
+        this.streams = new Array<ts.Stream>();
     }
 
     public async run(): Promise<void> {
@@ -295,7 +297,9 @@ class LocalTrainingService implements TrainingService {
 
     public cleanUp(): Promise<void> {
         this.stopping = true;
-
+        for(const stream of this.streams) {
+            stream.destroy();
+        }
         return Promise.resolve();
     }
 
@@ -382,6 +386,7 @@ class LocalTrainingService implements TrainingService {
                 buffer = remain;
             }
         });
+        this.streams.push(stream);
     }
 
     private async runHostJob(form: HostJobApplicationForm): Promise<TrialJobDetail> {

src/nni_manager/training_service/pai/hdfsClientUtility.ts

@@ -136,7 +136,7 @@ export namespace HDFSClientUtility {
         let timeoutId : NodeJS.Timer
         const delayTimeout : Promise<boolean> = new Promise<boolean>((resolve : Function, reject : Function) : void => {
             // Set timeout and reject the promise once reach timeout (5 seconds)        
-            setTimeout(() => deferred.reject(`Check HDFS path ${hdfsPath} exists timeout`), 5000);
+            timeoutId = setTimeout(() => deferred.reject(`Check HDFS path ${hdfsPath} exists timeout`), 5000);
         });
 
         return Promise.race([deferred.promise, delayTimeout]).finally(() => clearTimeout(timeoutId));

src/nni_manager/types/tail-stream/index.d.ts

 declare module 'tail-stream' {
     export interface Stream {
         on(type: 'data', callback: (data: Buffer) => void): void;
+        destroy(): void;
     }
     export function createReadStream(path: string): Stream;
 }
\ No newline at end of file

tools/nnicmd/nnictl_utils.py

@@ -190,7 +190,7 @@ def stop_experiment(args):
             time.sleep(3)
             rest_pid = nni_config.get_config('restServerPid')
             if rest_pid:
-                stop_rest_cmds = ['pkill', '-P', str(rest_pid)]
+                stop_rest_cmds = ['kill', str(rest_pid)]
                 call(stop_rest_cmds)
                 tensorboard_pid_list = nni_config.get_config('tensorboardPidList')
                 if tensorboard_pid_list: