Merge pull request #173 from microsoft/master

merge master

docs/zh_CN/assessors.rst

@@ -15,5 +15,5 @@ Assessor 从 Trial 中接收中间结果，并通过指定的算法决定此 Tri
 ..  toctree::
     :maxdepth: 2
 
-    内置 Assessor<builtinAssessor>
-    自定义 Assessor<Customize_Assessor>
+    内置 Assessor<BuiltinAssessor>
+    自定义 Assessor<CustomizeAssessor>

docs/zh_CN/automl_practice_sharing.rst

+#################
+自动机器学习的经验分享
+#################
+
+..  toctree::
+    :maxdepth: 2
+
+    神经网络架构搜索的对比<CommunitySharings/AutomlPracticeSharing/NasComparison>

docs/zh_CN/builtinTuner.rst

-内置 Tuner
-==================
-
-..  toctree::
-    :maxdepth: 1
-    
-    介绍<Builtin_Tuner>
-    TPE<hyperoptTuner>
-    Random Search<hyperoptTuner>
-    Anneal<hyperoptTuner>
-    Naive Evolution<evolutionTuner>
-    SMAC<smacTuner>
-    Batch Tuner<batchTuner>
-    Grid Search<gridsearchTuner>
-    Hyperband<hyperbandAdvisor>
-    Network Morphism<networkmorphismTuner>
-    Metis Tuner<metisTuner>
-    BOHB<bohbAdvisor>
\ No newline at end of file

docs/zh_CN/builtinAssessor.rst → docs/zh_CN/builtin_assessor.rst

@@ -4,6 +4,6 @@
 ..  toctree::
     :maxdepth: 1
 
-    介绍<Builtin_Assessors>
-    Medianstop<medianstopAssessor>
-    Curvefitting<curvefittingAssessor>
+    介绍<BuiltinAssessors>
+    Medianstop<MedianstopAssessor>
+    Curvefitting<CurvefittingAssessor>
\ No newline at end of file

docs/zh_CN/builtin_tuner.rst

+内置 Tuner
+==================
+
+..  toctree::
+    :maxdepth: 1
+    
+    介绍<BuiltinTuner>
+    TPE<HyperoptTuner>
+    Random Search<HyperoptTuner>
+    Anneal<HyperoptTuner>
+    Naive Evolution<EvolutionTuner>
+    SMAC<SmacTuner>
+    Batch Tuner<BatchTuner>
+    Grid Search<GridsearchTuner>
+    Hyperband<HyperbandAdvisor>
+    Network Morphism<NetworkmorphismTuner>
+    Metis Tuner<MetisTuner>
+    BOHB<BohbAdvisor>
\ No newline at end of file

docs/zh_CN/community_sharings.rst

+######################
+社区分享
+######################
+
+除了官方的教程和示例之外，也支持社区贡献者分享自己的自动机器学习实践经验，特别是使用 NNI 的实践经验。
+
+..  toctree::
+    :maxdepth: 2
+
+    NNI 经验分享<nni_practice_sharing>
+    神经网络结构搜索的对比<CommunitySharings/NasComparison>
+    超参调优算法的对比<CommunitySharings/HpoComparison>

docs/zh_CN/Contribution.rst → docs/zh_CN/contribution.rst

@@ -3,5 +3,5 @@
 ###############################
 
 ..  toctree::
-    设置开发环境<SetupNNIDeveloperEnvironment>
-    贡献指南<CONTRIBUTING>
+    设置开发环境<SetupNniDeveloperEnvironment>
+    贡献指南<Contributing>
\ No newline at end of file

docs/zh_CN/Examples.rst → docs/zh_CN/examples.rst

@@ -5,8 +5,8 @@
 ..  toctree::
     :maxdepth: 2
 
-    MNIST<mnist_examples>
-    Cifar10<cifar10_examples>
-    Scikit-learn<sklearn_examples>
-    EvolutionSQuAD<SQuAD_evolution_examples>
-    GBDT<gbdt_example>
+    MNIST<MnistExamples>
+    Cifar10<Cifar10Examples>
+    Scikit-learn<SklearnExamples>
+    EvolutionSQuAD<SquadEvolutionExamples>
+    GBDT<GbdtExample>

docs/zh_CN/index.rst

@@ -13,10 +13,10 @@ Neural Network Intelligence（NNI）文档
 
     概述<Overview>
     入门<QuickStart>
-    教程<Tutorials>
-    样例<Examples>
-    参考<Reference>
+    教程<tutorials>
+    示例<examples>
+    参考<reference>
     常见问答<FAQ>
-    贡献<Contribution>
-    版本日志<RELEASE>
-    博客<Blog/index>
+    贡献<contribution>
+    更改日志<Release>
+    社区经验分享<community_sharings>

docs/zh_CN/nni_practice_sharing.rst

+#################
+教程
+#################
+
+分享使用 NNI 来调优模型和系统的经验
+
+..  toctree::
+    :maxdepth: 2
+
+    在 NNI 上调优 Recommenders 的 SVD<CommunitySharings/NniPracticeSharing/RecommendersSvd>
\ No newline at end of file

docs/zh_CN/Reference.rst → docs/zh_CN/reference.rst

@@ -4,7 +4,7 @@
 ..  toctree::
     :maxdepth: 3
     
-    命令行<NNICTLDOC>
+    命令行<Nnictl>
     Python API<sdk_reference>
     Annotation<AnnotationSpec>
     配置<ExperimentConfig>

docs/zh_CN/training_services.rst

@@ -4,6 +4,6 @@ NNI 支持的训练平台介绍
 ..  toctree::
     本机<LocalMode>
     远程<RemoteMachineMode>
-    OpenPAI<PAIMode>
+    OpenPAI<PaiMode>
     Kubeflow<KubeflowMode>
     FrameworkController<FrameworkControllerMode>
\ No newline at end of file

docs/zh_CN/tuners.rst

@@ -13,6 +13,6 @@ Tuner 从 Trial 接收指标结果，来评估一组超参或网络结构的性
 ..  toctree::
     :maxdepth: 2
 
-    内置 Tuner<builtinTuner>
-    自定义 Tuner<Customize_Tuner>
-    自定义 Advisor<Customize_Advisor>
+    内置 Tuner<BuiltinTuner>
+    自定义 Tuner<CustomizeTuner>
+    自定义 Advisor<CustomizeAdvisor>
\ No newline at end of file

docs/zh_CN/tutorials.rst

+######################
+教程
+######################
+
+..  toctree::
+    :maxdepth: 2
+
+    安装<Installation>
+    实现 Trial<Trials>
+    Tuner<tuners>
+    Assessor<assessors>
+    Web 界面<WebUI>
+    训练平台<training_services>
+    如何使用 Docker <HowToUseDocker>
+    高级功能<advanced>
+    如何调试<HowToDebug>
\ No newline at end of file

install.ps1

@@ -15,7 +15,7 @@ $yarnUrl = "https://yarnpkg.com/latest.tar.gz"
 $unzipNodeDir = "node-v*"
 $unzipYarnDir = "yarn-v*"
 
-$NNI_DEPENDENCY_FOLDER = "C:\tmp\$env:USERNAME"
+$NNI_DEPENDENCY_FOLDER = [System.IO.Path]::GetTempPath()+$env:USERNAME
 
 $WHICH_PYTHON = where.exe python
 if($WHICH_PYTHON -eq $null){

src/nni_manager/common/utils.ts

@@ -43,11 +43,11 @@ function getExperimentRootDir(): string {
             .getLogDir();
 }
 
-function getLogDir(): string{
+function getLogDir(): string {
     return path.join(getExperimentRootDir(), 'log');
 }
 
-function getLogLevel(): string{
+function getLogLevel(): string {
     return getExperimentStartupInfo()
     .getLogLevel();
 }
@@ -149,7 +149,7 @@ function parseArg(names: string[]): string {
     return '';
 }
 
-function encodeCmdLineArgs(args:any):any{
+function encodeCmdLineArgs(args: any): any {
     if(process.platform === 'win32'){
         return JSON.stringify(args);
     }
@@ -158,7 +158,7 @@ function encodeCmdLineArgs(args:any):any{
     }
 }
 
-function getCmdPy():string{
+function getCmdPy(): string {
     let cmd = 'python3';
     if(process.platform === 'win32'){
         cmd = 'python';
@@ -390,7 +390,7 @@ async function getVersion(): Promise<string> {
 /**
  * run command as ChildProcess
  */
-function getTunerProc(command: string, stdio: StdioOptions, newCwd: string, newEnv: any): ChildProcess{
+function getTunerProc(command: string, stdio: StdioOptions, newCwd: string, newEnv: any): ChildProcess {
     let cmd: string = command;
     let arg: string[] = [];
     let newShell: boolean = true;
@@ -411,7 +411,7 @@ function getTunerProc(command: string, stdio: StdioOptions, newCwd: string, newE
 /**
  * judge whether the process is alive
  */
-async function isAlive(pid:any): Promise<boolean>{
+async function isAlive(pid:any): Promise<boolean> {
     let deferred : Deferred<boolean> = new Deferred<boolean>();
     let alive: boolean = false;
     if(process.platform ==='win32'){
@@ -439,7 +439,7 @@ async function isAlive(pid:any): Promise<boolean>{
 /**
  * kill process 
  */
-async function killPid(pid:any): Promise<void>{
+async function killPid(pid:any): Promise<void> {
     let deferred : Deferred<void> = new Deferred<void>();
     try {
         if (process.platform === "win32") {
@@ -455,7 +455,7 @@ async function killPid(pid:any): Promise<void>{
     return deferred.promise;
 }
 
-function getNewLine(): string{
+function getNewLine(): string {
     if (process.platform === "win32") {
         return "\r\n";
     }

src/nni_manager/core/nnimanager.ts

@@ -58,7 +58,8 @@ class NNIManager implements Manager {
     private status: NNIManagerStatus;
     private waitingTrials: string[];
     private trialJobs: Map<string, TrialJobDetail>;
-
+    private trialJobMetricListener: (metric: TrialJobMetric) => void;
+    
     constructor() {
         this.currSubmittedTrialNum = 0;
         this.trialConcurrencyChange = 0;
@@ -76,6 +77,11 @@ class NNIManager implements Manager {
             status: 'INITIALIZED',
             errors: []
         };
+        this.trialJobMetricListener = (metric: TrialJobMetric) => {
+            this.onTrialJobMetrics(metric).catch((err: Error) => {
+                this.criticalError(NNIError.FromError(err, 'Job metrics error: '));
+            });
+        };
     }
 
     public updateExperimentProfile(experimentProfile: ExperimentProfile, updateType: ProfileUpdateType): Promise<void> {
@@ -342,6 +348,7 @@ class NNIManager implements Manager {
         if (this.dispatcher === undefined) {
             throw new Error('Error: tuner has not been setup');
         }
+        this.trainingService.removeTrialJobMetricListener(this.trialJobMetricListener); 
         this.dispatcher.sendCommand(TERMINATE);
         let tunerAlive: boolean = true;
         // gracefully terminate tuner and assessor here, wait at most 30 seconds.
@@ -589,11 +596,7 @@ class NNIManager implements Manager {
         if (this.dispatcher === undefined) {
             throw new Error('Error: tuner or job maintainer have not been setup');
         }
-        this.trainingService.addTrialJobMetricListener((metric: TrialJobMetric) => {
-            this.onTrialJobMetrics(metric).catch((err: Error) => {
-                this.criticalError(NNIError.FromError(err, 'Job metrics error: '));
-            });
-        });
+        this.trainingService.addTrialJobMetricListener(this.trialJobMetricListener);
 
         this.dispatcher.onCommand((commandType: string, content: string) => {
             this.onTunerCommand(commandType, content).catch((err: Error) => {

src/nni_manager/training_service/common/util.ts

@@ -24,7 +24,10 @@ import { getLogger } from "common/log";
 import { countFilesRecursively } from '../../common/utils'
 import * as cpp from 'child-process-promise';
 import * as cp from 'child_process';
-import { GPU_INFO_COLLECTOR_FORMAT_LINUX, GPU_INFO_COLLECTOR_FORMAT_WINDOWS } from './gpuData'
+import * as os from 'os';
+import * as fs from 'fs';
+import { getNewLine } from '../../common/utils';
+import { GPU_INFO_COLLECTOR_FORMAT_LINUX, GPU_INFO_COLLECTOR_FORMAT_WINDOWS } from './gpuData';
 import * as path from 'path';
 import { String } from 'typescript-string-operations';
 import { file } from "../../node_modules/@types/tmp";
@@ -66,6 +69,20 @@ export async function execMkdir(directory: string): Promise<void> {
     return Promise.resolve();
 }
 
+/**
+ * copy files to the directory
+ * @param source
+ * @param destination
+ */
+export async function execCopydir(source: string, destination: string): Promise<void> {
+    if (process.platform === 'win32') {
+        await cpp.exec(`powershell.exe Copy-Item ${source} -Destination ${destination} -Recurse`);
+    } else {
+        await cpp.exec(`cp -r ${source} ${destination}`);
+    }
+    return Promise.resolve();
+}
+
 /**
  * crete a new file
  * @param filename 
@@ -91,8 +108,6 @@ export function execScript(filePath: string): cp.ChildProcess {
     }
 }
 
-
-
 /**
  * output the last line of a file
  * @param filePath 
@@ -111,9 +126,9 @@ export async function execTail(filePath: string): Promise<cpp.childProcessPromis
  * delete a directory
  * @param directory 
  */
-export async function execRemove(directory: string): Promise<void>{
+export async function execRemove(directory: string): Promise<void> {
     if (process.platform === 'win32') {
-        await cpp.exec(`powershell.exe Remove-Item ${directory}`);
+        await cpp.exec(`powershell.exe Remove-Item ${directory} -Recurse -Force`);
     } else {
         await cpp.exec(`rm -rf ${directory}`);
     }
@@ -124,7 +139,7 @@ export async function execRemove(directory: string): Promise<void>{
  * kill a process
  * @param directory 
  */
-export async function execKill(pid: string): Promise<void>{
+export async function execKill(pid: string): Promise<void> {
     if (process.platform === 'win32') {
         await cpp.exec(`cmd /c taskkill /PID ${pid} /T /F`);
     } else {
@@ -138,7 +153,7 @@ export async function execKill(pid: string): Promise<void>{
  * @param  variable
  * @returns command string  
  */
-export function setEnvironmentVariable(variable: { key: string; value: string }): string{
+export function setEnvironmentVariable(variable: { key: string; value: string }): string {
     if (process.platform === 'win32') {
         return `$env:${variable.key}="${variable.value}"`;
     }
@@ -147,6 +162,32 @@ export function setEnvironmentVariable(variable: { key: string; value: string })
     }
 }
 
+/**
+ * Compress files in directory to tar file
+ * @param  source_path
+ * @param  tar_path
+ */
+export async function tarAdd(tar_path: string, source_path: string): Promise<void> {
+    if (process.platform === 'win32') {
+        tar_path = tar_path.split('\\').join('\\\\');
+        source_path = source_path.split('\\').join('\\\\');
+        let script: string[] = [];
+        script.push(
+            `import os`,
+            `import tarfile`,
+            String.Format(`tar = tarfile.open("{0}","w:gz")\r\nfor root,dir,files in os.walk("{1}"):`, tar_path, source_path),
+            `    for file in files:`,
+            `        fullpath = os.path.join(root,file)`,
+            `        tar.add(fullpath, arcname=file)`,
+            `tar.close()`);
+        await fs.promises.writeFile(path.join(os.tmpdir(), 'tar.py'), script.join(getNewLine()), { encoding: 'utf8', mode: 0o777 });
+        const tarScript: string = path.join(os.tmpdir(), 'tar.py');
+        await cpp.exec(`python ${tarScript}`);
+    } else {
+        await cpp.exec(`tar -czf ${tar_path} -C ${source_path} .`);
+    }
+    return Promise.resolve();
+}
 
 /**
  * generate script file name

src/nni_manager/training_service/remote_machine/remoteMachineTrainingService.ts

@@ -36,7 +36,7 @@ import { ObservableTimer } from '../../common/observableTimer';
 import {
     HostJobApplicationForm, HyperParameters, JobApplicationForm, TrainingService, TrialJobApplicationForm, TrialJobDetail, TrialJobMetric, NNIManagerIpConfig
 } from '../../common/trainingService';
-import { delay, generateParamFileName, getExperimentRootDir, uniqueString, getJobCancelStatus, getRemoteTmpDir,getIPV4Address  } from '../../common/utils';
+import { delay, generateParamFileName, getExperimentRootDir, uniqueString, getJobCancelStatus, getRemoteTmpDir,getIPV4Address, getVersion, unixPathJoin } from '../../common/utils';
 import { GPUSummary } from '../common/gpuData';
 import { TrialConfig } from '../common/trialConfig';
 import { TrialConfigMetadataKey } from '../common/trialConfigMetadataKey';
@@ -48,10 +48,9 @@ import {
 } from './remoteMachineData';
 import { GPU_INFO_COLLECTOR_FORMAT_LINUX } from '../common/gpuData';
 import { SSHClientUtility } from './sshClientUtility';
-import { validateCodeDir } from '../common/util';
+import { validateCodeDir, execRemove, execMkdir, execCopydir } from '../common/util';
 import { RemoteMachineJobRestServer } from './remoteMachineJobRestServer';
 import { CONTAINER_INSTALL_NNI_SHELL_FORMAT } from '../common/containerJobData';
-import { mkDirP, getVersion } from '../../common/utils';
 
 /**
  * Training Service implementation for Remote Machine (Linux)
@@ -234,7 +233,7 @@ class RemoteMachineTrainingService implements TrainingService {
         } else if (form.jobType === 'TRIAL') {
             // Generate trial job id(random)
             const trialJobId: string = uniqueString(5);
-            const trialWorkingFolder: string = path.join(this.remoteExpRootDir, 'trials', trialJobId);
+            const trialWorkingFolder: string = unixPathJoin(this.remoteExpRootDir, 'trials', trialJobId);
 
             const trialJobDetail: RemoteMachineTrialJobDetail = new RemoteMachineTrialJobDetail(
                 trialJobId,
@@ -354,7 +353,7 @@ class RemoteMachineTrainingService implements TrainingService {
             case TrialConfigMetadataKey.MACHINE_LIST:
                 await this.setupConnections(value);
                 //remove local temp files
-                await cpp.exec(`rm -rf ${this.getLocalGpuMetricCollectorDir()}`);
+                await execRemove(this.getLocalGpuMetricCollectorDir());
                 break;
             case TrialConfigMetadataKey.TRIAL_CONFIG:
                 const remoteMachineTrailConfig: TrialConfig = <TrialConfig>JSON.parse(value);
@@ -417,7 +416,7 @@ class RemoteMachineTrainingService implements TrainingService {
     private async cleanupConnections(): Promise<void> {
         try{
             for (const [rmMeta, sshClientManager] of this.machineSSHClientMap.entries()) {
-                let jobpidPath: string = path.join(this.getRemoteScriptsPath(rmMeta.username), 'pid');
+                let jobpidPath: string = unixPathJoin(this.getRemoteScriptsPath(rmMeta.username), 'pid');
                 let client: Client | undefined = sshClientManager.getFirstSSHClient();
                 if(client) {
                     await SSHClientUtility.remoteExeCommand(`pkill -P \`cat ${jobpidPath}\``, client);
@@ -438,7 +437,7 @@ class RemoteMachineTrainingService implements TrainingService {
      */
     private getLocalGpuMetricCollectorDir(): string {
         let userName: string = path.basename(os.homedir()); //get current user name of os
-        return `${os.tmpdir()}/${userName}/nni/scripts/`;
+        return path.join(os.tmpdir(), userName, 'nni', 'scripts');
     }
 
     /**
@@ -447,14 +446,14 @@ class RemoteMachineTrainingService implements TrainingService {
      */
     private async generateGpuMetricsCollectorScript(userName: string): Promise<void> {
         let gpuMetricCollectorScriptFolder : string = this.getLocalGpuMetricCollectorDir();
-        await cpp.exec(`mkdir -p ${path.join(gpuMetricCollectorScriptFolder, userName)}`);
+        await execMkdir(path.join(gpuMetricCollectorScriptFolder, userName));
         //generate gpu_metrics_collector.sh
         let gpuMetricsCollectorScriptPath: string = path.join(gpuMetricCollectorScriptFolder, userName, 'gpu_metrics_collector.sh');
         const remoteGPUScriptsDir: string = this.getRemoteScriptsPath(userName); // This directory is used to store gpu_metrics and pid created by script
         const gpuMetricsCollectorScriptContent: string = String.Format(
             GPU_INFO_COLLECTOR_FORMAT_LINUX, 
             remoteGPUScriptsDir, 
-            path.join(remoteGPUScriptsDir, 'pid'), 
+            unixPathJoin(remoteGPUScriptsDir, 'pid'), 
         );
         await fs.promises.writeFile(gpuMetricsCollectorScriptPath, gpuMetricsCollectorScriptContent, { encoding: 'utf8' });
     }
@@ -481,7 +480,7 @@ class RemoteMachineTrainingService implements TrainingService {
     private async initRemoteMachineOnConnected(rmMeta: RemoteMachineMeta, conn: Client): Promise<void> {
         // Create root working directory after ssh connection is ready
         await this.generateGpuMetricsCollectorScript(rmMeta.username); //generate gpu script in local machine first, will copy to remote machine later
-        const nniRootDir: string = `${os.tmpdir()}/nni`;
+        const nniRootDir: string = unixPathJoin(getRemoteTmpDir(this.remoteOS), 'nni');
         await SSHClientUtility.remoteExeCommand(`mkdir -p ${this.remoteExpRootDir}`, conn);
 
         // Copy NNI scripts to remote expeirment working directory
@@ -490,15 +489,15 @@ class RemoteMachineTrainingService implements TrainingService {
         await SSHClientUtility.remoteExeCommand(`mkdir -p ${remoteGpuScriptCollectorDir}`, conn);
         await SSHClientUtility.remoteExeCommand(`chmod 777 ${nniRootDir} ${nniRootDir}/* ${nniRootDir}/scripts/*`, conn);
         //copy gpu_metrics_collector.sh to remote
-        await SSHClientUtility.copyFileToRemote(path.join(localGpuScriptCollectorDir, rmMeta.username, 'gpu_metrics_collector.sh'), path.join(remoteGpuScriptCollectorDir, 'gpu_metrics_collector.sh'), conn);
+        await SSHClientUtility.copyFileToRemote(path.join(localGpuScriptCollectorDir, rmMeta.username, 'gpu_metrics_collector.sh'), unixPathJoin(remoteGpuScriptCollectorDir, 'gpu_metrics_collector.sh'), conn);
 
         //Begin to execute gpu_metrics_collection scripts
-        SSHClientUtility.remoteExeCommand(`bash ${path.join(remoteGpuScriptCollectorDir, 'gpu_metrics_collector.sh')}`, conn);
+        SSHClientUtility.remoteExeCommand(`bash ${unixPathJoin(remoteGpuScriptCollectorDir, 'gpu_metrics_collector.sh')}`, conn);
 
         this.timer.subscribe(
             async (tick: number) => {
                 const cmdresult: RemoteCommandResult = await SSHClientUtility.remoteExeCommand(
-                    `tail -n 1 ${path.join(remoteGpuScriptCollectorDir, 'gpu_metrics')}`, conn);
+                    `tail -n 1 ${unixPathJoin(remoteGpuScriptCollectorDir, 'gpu_metrics')}`, conn);
                 if (cmdresult && cmdresult.stdout) {
                     rmMeta.gpuSummary = <GPUSummary>JSON.parse(cmdresult.stdout);
                 }
@@ -531,7 +530,7 @@ class RemoteMachineTrainingService implements TrainingService {
         } else if (rmScheduleResult.resultType === ScheduleResultType.SUCCEED
             && rmScheduleResult.scheduleInfo !== undefined) {
             const rmScheduleInfo : RemoteMachineScheduleInfo = rmScheduleResult.scheduleInfo;
-            const trialWorkingFolder: string = path.join(this.remoteExpRootDir, 'trials', trialJobId);
+            const trialWorkingFolder: string = unixPathJoin(this.remoteExpRootDir, 'trials', trialJobId);
 
             trialJobDetail.rmMeta = rmScheduleInfo.rmMeta;
 
@@ -575,7 +574,7 @@ class RemoteMachineTrainingService implements TrainingService {
         const trialLocalTempFolder: string = path.join(this.expRootDir, 'trials-local', trialJobId);
 
         await SSHClientUtility.remoteExeCommand(`mkdir -p ${trialWorkingFolder}`, sshClient);
-        await SSHClientUtility.remoteExeCommand(`mkdir -p ${path.join(trialWorkingFolder, '.nni')}`, sshClient);
+        await SSHClientUtility.remoteExeCommand(`mkdir -p ${unixPathJoin(trialWorkingFolder, '.nni')}`, sshClient);
 
         // RemoteMachineRunShellFormat is the run shell format string,
         // See definition in remoteMachineData.ts
@@ -603,20 +602,20 @@ class RemoteMachineTrainingService implements TrainingService {
             getExperimentId(),
             trialJobDetail.sequenceId.toString(),
             this.isMultiPhase,
-            path.join(trialWorkingFolder, '.nni', 'jobpid'),
+            unixPathJoin(trialWorkingFolder, '.nni', 'jobpid'),
             command,
             nniManagerIp,
             this.remoteRestServerPort,
             version,
             this.logCollection,
-            path.join(trialWorkingFolder, '.nni', 'code')
+            unixPathJoin(trialWorkingFolder, '.nni', 'code')
         )
 
         //create tmp trial working folder locally.
-        await cpp.exec(`mkdir -p ${path.join(trialLocalTempFolder, '.nni')}`);
+        await execMkdir(path.join(trialLocalTempFolder, '.nni'));
 
         //create tmp trial working folder locally.
-        await cpp.exec(`cp -r ${this.trialConfig.codeDir}/* ${trialLocalTempFolder}`);
+        await execCopydir(path.join(this.trialConfig.codeDir, '*'), trialLocalTempFolder);
         const installScriptContent : string = CONTAINER_INSTALL_NNI_SHELL_FORMAT;
         // Write NNI installation file to local tmp files
         await fs.promises.writeFile(path.join(trialLocalTempFolder, 'install_nni.sh'), installScriptContent, { encoding: 'utf8' });
@@ -626,7 +625,7 @@ class RemoteMachineTrainingService implements TrainingService {
         // Copy files in codeDir to remote working directory
         await SSHClientUtility.copyDirectoryToRemote(trialLocalTempFolder, trialWorkingFolder, sshClient, this.remoteOS);
         // Execute command in remote machine
-        SSHClientUtility.remoteExeCommand(`bash ${path.join(trialWorkingFolder, 'run.sh')}`, sshClient);
+        SSHClientUtility.remoteExeCommand(`bash ${unixPathJoin(trialWorkingFolder, 'run.sh')}`, sshClient);
     }
 
     private async runHostJob(form: HostJobApplicationForm): Promise<TrialJobDetail> {
@@ -646,8 +645,8 @@ class RemoteMachineTrainingService implements TrainingService {
         );
         await fs.promises.writeFile(path.join(localDir, 'run.sh'), runScriptContent, { encoding: 'utf8' });
         await SSHClientUtility.copyFileToRemote(
-            path.join(localDir, 'run.sh'), path.join(remoteDir, 'run.sh'), sshClient);
-        SSHClientUtility.remoteExeCommand(`bash ${path.join(remoteDir, 'run.sh')}`, sshClient);
+            path.join(localDir, 'run.sh'), unixPathJoin(remoteDir, 'run.sh'), sshClient);
+        SSHClientUtility.remoteExeCommand(`bash ${unixPathJoin(remoteDir, 'run.sh')}`, sshClient);
 
         const jobDetail: RemoteMachineTrialJobDetail =  new RemoteMachineTrialJobDetail(
             jobId, 'RUNNING', Date.now(), remoteDir, form, this.generateSequenceId()
@@ -672,7 +671,7 @@ class RemoteMachineTrainingService implements TrainingService {
     private async updateTrialJobStatus(trialJob: RemoteMachineTrialJobDetail, sshClient: Client): Promise<TrialJobDetail> {
         const deferred: Deferred<TrialJobDetail> = new Deferred<TrialJobDetail>();
         const jobpidPath: string = this.getJobPidPath(trialJob.id);
-        const trialReturnCodeFilePath: string = path.join(this.remoteExpRootDir, 'trials', trialJob.id, '.nni', 'code');
+        const trialReturnCodeFilePath: string = unixPathJoin(this.remoteExpRootDir, 'trials', trialJob.id, '.nni', 'code');
         try {
             const killResult: number = (await SSHClientUtility.remoteExeCommand(`kill -0 \`cat ${jobpidPath}\``, sshClient)).exitCode;
             // if the process of jobpid is not alive any more
@@ -712,15 +711,15 @@ class RemoteMachineTrainingService implements TrainingService {
     }
 
     private getRemoteScriptsPath(userName: string): string {
-        return path.join(getRemoteTmpDir(this.remoteOS), userName, 'nni', 'scripts');
+        return unixPathJoin(getRemoteTmpDir(this.remoteOS), userName, 'nni', 'scripts');
     }
 
     private getHostJobRemoteDir(jobId: string): string {
-        return path.join(this.remoteExpRootDir, 'hostjobs', jobId);
+        return unixPathJoin(this.remoteExpRootDir, 'hostjobs', jobId);
     }
 
     private getRemoteExperimentRootDir(): string{
-        return path.join(getRemoteTmpDir(this.remoteOS), 'nni', 'experiments', getExperimentId());
+        return unixPathJoin(getRemoteTmpDir(this.remoteOS), 'nni', 'experiments', getExperimentId());
     }
 
     public get MetricsEmitter() : EventEmitter {
@@ -735,9 +734,9 @@ class RemoteMachineTrainingService implements TrainingService {
 
         let jobpidPath: string;
         if (trialJobDetail.form.jobType === 'TRIAL') {
-            jobpidPath = path.join(trialJobDetail.workingDirectory, '.nni', 'jobpid');
+            jobpidPath = unixPathJoin(trialJobDetail.workingDirectory, '.nni', 'jobpid');
         } else if (trialJobDetail.form.jobType === 'HOST') {
-            jobpidPath = path.join(this.getHostJobRemoteDir(jobId), 'jobpid');
+            jobpidPath = unixPathJoin(this.getHostJobRemoteDir(jobId), 'jobpid');
         } else {
             throw new Error(`Job type not supported: ${trialJobDetail.form.jobType}`);
         }
@@ -751,14 +750,14 @@ class RemoteMachineTrainingService implements TrainingService {
             throw new Error('sshClient is undefined.');
         }
 
-        const trialWorkingFolder: string = path.join(this.remoteExpRootDir, 'trials', trialJobId);
+        const trialWorkingFolder: string = unixPathJoin(this.remoteExpRootDir, 'trials', trialJobId);
         const trialLocalTempFolder: string = path.join(this.expRootDir, 'trials-local', trialJobId);
 
         const fileName: string = generateParamFileName(hyperParameters);
         const localFilepath: string = path.join(trialLocalTempFolder, fileName);
         await fs.promises.writeFile(localFilepath, hyperParameters.value, { encoding: 'utf8' });
 
-        await SSHClientUtility.copyFileToRemote(localFilepath, path.join(trialWorkingFolder, fileName), sshClient);
+        await SSHClientUtility.copyFileToRemote(localFilepath, unixPathJoin(trialWorkingFolder, fileName), sshClient);
     }
 
     private generateSequenceId(): number {

src/nni_manager/training_service/remote_machine/sshClientUtility.ts

@@ -28,8 +28,9 @@ import * as stream from 'stream';
 import { Deferred } from 'ts-deferred';
 import { NNIError, NNIErrorNames } from '../../common/errors';
 import { getLogger, Logger } from '../../common/log';
-import { uniqueString, getRemoteTmpDir } from '../../common/utils';
+import { uniqueString, getRemoteTmpDir, unixPathJoin } from '../../common/utils';
 import { RemoteCommandResult } from './remoteMachineData';
+import { execRemove, tarAdd } from '../common/util';
 
 /**
  *
@@ -47,13 +48,13 @@ export namespace SSHClientUtility {
         const deferred: Deferred<void> = new Deferred<void>();
         const tmpTarName: string = `${uniqueString(10)}.tar.gz`;
         const localTarPath: string = path.join(os.tmpdir(), tmpTarName);
-        const remoteTarPath: string = path.join(getRemoteTmpDir(remoteOS), tmpTarName);
+        const remoteTarPath: string = unixPathJoin(getRemoteTmpDir(remoteOS), tmpTarName);
 
         // Compress files in local directory to experiment root directory
-        await cpp.exec(`tar -czf ${localTarPath} -C ${localDirectory} .`);
+        await tarAdd(localTarPath, localDirectory);
         // Copy the compressed file to remoteDirectory and delete it
         await copyFileToRemote(localTarPath, remoteTarPath, sshClient);
-        await cpp.exec(`rm ${localTarPath}`);
+        await execRemove(localTarPath);
         // Decompress the remote compressed file in and delete it
         await remoteExeCommand(`tar -oxzf ${remoteTarPath} -C ${remoteDirectory}`, sshClient);
         await remoteExeCommand(`rm ${remoteTarPath}`, sshClient);