Merge pull request 'GPU智算网络' (#2386) from grampus-specialpool into V20220630

Reviewed-on: https://git.openi.org.cn/OpenI/aiforge/pulls/2386
Reviewed-by: ychao_1983 <ychao_1983@sina.com>

modules/grampus/grampus.go

@@ -19,12 +19,14 @@ const (
 	ProcessorTypeNPU = "npu.huawei.com/NPU"
 	ProcessorTypeGPU = "nvidia.com/gpu"
 
-	CommandPrepareScript = "pwd;cd /cache;mkdir -p output;mkdir -p code;mkdir -p dataset;echo \"start loading script\";wget -q https://git.openi.org.cn/OpenIOSSG/script_for_grampus/archive/master.zip;" +
+	GpuWorkDir = "/tmp/"
+	NpuWorkDir = "/cache/"
+
+	CommandPrepareScript = ";mkdir -p output;mkdir -p code;mkdir -p dataset;echo \"start loading script\";wget -q https://git.openi.org.cn/OpenIOSSG/script_for_grampus/archive/master.zip;" +
 		"echo \"finish loading script\";unzip -q master.zip;cd script_for_grampus;chmod 777 downloader_for_obs uploader_for_obs downloader_for_minio uploader_for_minio;"
+	//CommandPrepareScript = "pwd;cd /cache;mkdir -p output;mkdir -p code;mkdir -p dataset;echo \"start loading script\";wget -q https://git.openi.org.cn/OpenIOSSG/script_for_grampus/archive/master.zip;" +
+	//	"echo \"finish loading script\";unzip -q master.zip;cd script_for_grampus;chmod 777 downloader_for_obs uploader_for_obs downloader_for_minio uploader_for_minio;"
 
-	//CommandPrepareScript = "bash;pwd;apt-get -y update;apt-get -y upgrade;apt-get -y install wget;apt-get -y install unzip;" +
-	//	"cd /tmp;mkdir -p output;mkdir -p code;mkdir -p dataset;wget -q https://git.openi.org.cn/OpenIOSSG/script_for_grampus/archive/master.zip;" +
-	//	"unzip -q master.zip;cd script_for_grampus;chmod 777 downloader_for_obs uploader_for_obs downloader_for_minio uploader_for_minio;"
 	CodeArchiveName = "master.zip"
 )
 

public/home/home.js

@@ -156,7 +156,7 @@ document.onreadystatechange = function () {
                     html += recordPrefix + actionName;
                     html += " <a href=\"" + getRepoLink(record) + "\" rel=\"nofollow\">" + getRepotext(record) + "</a>"
                 } 
-                else if(record.OpType == "24"  ||  record.OpType == "26" || record.OpType == "27" || record.OpType == "28"  || record.OpType == "30" || record.OpType == "31"){
+                else if(record.OpType == "24"  ||  record.OpType == "26" || record.OpType == "27" || record.OpType == "28"  || record.OpType == "30" || record.OpType == "31" || record.OpType == "32" || record.OpType == "33"){
                     html += recordPrefix + actionName;
                     html += " <a href=\"" + getTaskLink(record) + "\" rel=\"nofollow\">" + record.RefName + "</a>"
                 }
@@ -201,6 +201,8 @@ function getTaskLink(record){
         re = re + "/modelmanage/show_model_info?name=" + record.RefName;
     }else if(record.OpType == 31){
         re = re + "/cloudbrain/train-job/" + record.Content;
+    }else if(record.OpType == 32 || record.OpType == 33){
+        re = re + "/grampus/train-job/" + record.Content;
     }
     re = encodeURI(re);
     return re;
@@ -374,7 +376,9 @@ var actionNameZH={
     "28":"创建了推理任务",
     "29":"创建了评测任务",
     "30":"导入了新模型",
-    "31":"创建了CPU/GPU类型训练任务"
+    "31":"创建了CPU/GPU类型训练任务",
+    "32":"创建了NPU类型训练任务",
+    "33":"创建了CPU/GPU类型训练任务"
 };
 
 var actionNameEN={
@@ -401,6 +405,8 @@ var actionNameEN={
     "29":" created profiling task",
     "30":" created new model",
     "31":" created CPU/GPU type training task",
+    "32":" created NPU type training task",
+    "33":" created CPU/GPU type training task"
 };
 
 var repoAndOrgZH={

routers/admin/cloudbrains.go

@@ -43,12 +43,6 @@ func CloudBrains(ctx *context.Context) {
 	if page <= 0 {
 		page = 1
 	}
-	debugType := models.TypeCloudBrainAll
-	if listType == models.GPUResource {
-		debugType = models.TypeCloudBrainOne
-	} else if listType == models.NPUResource {
-		debugType = models.TypeCloudBrainTwo
-	}
 
 	var jobTypes []string
 	jobTypeNot := false
@@ -77,13 +71,14 @@ func CloudBrains(ctx *context.Context) {
 			PageSize: setting.UI.IssuePagingNum,
 		},
 		Keyword:         keyword,
-		Type:            debugType,
 		JobTypeNot:      jobTypeNot,
 		JobStatusNot:    jobStatusNot,
 		JobStatus:       jobStatuses,
 		JobTypes:        jobTypes,
 		NeedRepoInfo:    true,
 		IsLatestVersion: modelarts.IsLatestVersion,
+		ComputeResource: listType,
+		Type:            models.TypeCloudBrainAll,
 	})
 	if err != nil {
 		ctx.ServerError("Get job failed:", err)

routers/repo/grampus.go

@@ -664,7 +664,12 @@ func GrampusGetLog(ctx *context.Context) {
 func generateCommand(repoName, processorType, codeRemotePath, dataRemotePath, bootFile, paramSrc, outputRemotePath, datasetName string) (string, error) {
 	var command string
 
-	command += grampus.CommandPrepareScript
+	workDir := grampus.NpuWorkDir
+	if processorType == grampus.ProcessorTypeGPU {
+		workDir = grampus.GpuWorkDir
+	}
+
+	command += "pwd;cd " + workDir + grampus.CommandPrepareScript
 	//download code & dataset
 	if processorType == grampus.ProcessorTypeNPU {
 		commandDownload := "./downloader_for_obs " + setting.Bucket + " " + codeRemotePath + " " + grampus.CodeArchiveName + " " + dataRemotePath + " " + datasetName + ";"
@@ -683,7 +688,7 @@ func generateCommand(repoName, processorType, codeRemotePath, dataRemotePath, bo
 	if strings.HasSuffix(datasetName, ".tar.gz") {
 		toolUnzip = "tar -zxvf "
 	}
-	commandUnzip := "cd /cache/code;unzip -q master.zip;echo \"start to unzip dataset\";cd /cache/dataset;" + toolUnzip + datasetName + ";"
+	commandUnzip := "cd " + workDir + "code;unzip -q master.zip;echo \"start to unzip dataset\";cd " + workDir + "dataset;" + toolUnzip + datasetName + ";"
 	command += commandUnzip
 
 	//check unzip result
@@ -712,7 +717,7 @@ func generateCommand(repoName, processorType, codeRemotePath, dataRemotePath, bo
 		}
 	}
 
-	commandCode := "cd /cache/code/" + strings.ToLower(repoName) + ";python " + bootFile + paramCode + ";"
+	commandCode := "cd " + workDir + "code/" + strings.ToLower(repoName) + ";python " + bootFile + paramCode + ";"
 	command += commandCode
 
 	//get exec result
@@ -721,10 +726,10 @@ func generateCommand(repoName, processorType, codeRemotePath, dataRemotePath, bo
 
 	//upload models
 	if processorType == grampus.ProcessorTypeNPU {
-		commandUpload := "cd /cache/script_for_grampus/;./uploader_for_obs " + setting.Bucket + " " + outputRemotePath + " " + "/cache/output/;"
+		commandUpload := "cd " + workDir + "script_for_grampus/;./uploader_for_obs " + setting.Bucket + " " + outputRemotePath + " " + workDir + "output/;"
 		command += commandUpload
 	} else if processorType == grampus.ProcessorTypeGPU {
-		commandUpload := "cd /cache/script_for_grampus/;./uploader_for_minio " + setting.Grampus.Env + " " + outputRemotePath + " " + "/cache/output/;"
+		commandUpload := "cd " + workDir + "script_for_grampus/;./uploader_for_minio " + setting.Grampus.Env + " " + outputRemotePath + " " + workDir + "output/;"
 		command += commandUpload
 	}
 

services/socketwrap/clientManager.go

@@ -10,7 +10,7 @@ import (
 	"github.com/elliotchance/orderedmap"
 )
 
-var opTypes = []int{1, 2, 5, 6, 7, 9, 10, 11, 12, 13, 14, 15, 17, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31}
+var opTypes = []int{1, 2, 5, 6, 7, 9, 10, 11, 12, 13, 14, 15, 17, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33}
 
 type ClientsManager struct {
 	Clients    *orderedmap.OrderedMap

templates/admin/cloudbrain/list.tmpl

@@ -102,7 +102,7 @@
                                     </a>
                                     {{else if eq .JobType "TRAIN"}}
                                     <a class="title"
-                                        href="{{AppSubUrl}}/{{.Repo.OwnerName}}/{{.Repo.Name}}{{if eq .ComputeResource "CPU/GPU"}}/cloudbrain{{else}}/modelarts{{end}}/train-job/{{$JobID}}"
+                                        href="{{AppSubUrl}}/{{.Repo.OwnerName}}/{{.Repo.Name}}{{if eq .Cloudbrain.Type 0}}/cloudbrain{{else if eq .Cloudbrain.Type 1}}/modelarts{{else if eq .Cloudbrain.Type 2}}/grampus{{end}}/train-job/{{$JobID}}"
                                         title="{{.DisplayJobName}}" style="font-size: 14px;">
                                         <span class="fitted"
                                             style="width: 90%;vertical-align: middle;">{{.DisplayJobName}}</span>
@@ -204,7 +204,7 @@
                                         {{else}}
                                         <a style="padding: 0.5rem 1rem;" id="ai-stop-{{$JobID}}"
                                             class="ui basic ai_stop_version {{if eq .Status "KILLED" "FAILED" "START_FAILED" "KILLING" "COMPLETED" "SUCCEEDED" "STOPPED"}}disabled {{else}} blue {{end}}button"
-                                            data-repopath="{{.Repo.OwnerName}}/{{.Repo.Name}}{{if eq .ComputeResource "CPU/GPU"}}/cloudbrain/train-job{{else}}/modelarts/{{if eq .JobType "INFERENCE"}}inference-job{{else}}train-job{{end}}{{end}}"
+                                            data-repopath="{{.Repo.OwnerName}}/{{.Repo.Name}}{{if eq .Cloudbrain.Type 0}}/cloudbrain/train-job{{else if eq .Cloudbrain.Type 2}}/grampus/train-job{{else}}/modelarts/{{if eq .JobType "INFERENCE"}}inference-job{{else}}train-job{{end}}{{end}}"
                                             data-jobid="{{$JobID}}" data-version="{{.VersionName}}">
                                             {{$.i18n.Tr "repo.stop"}}
                                         </a>
@@ -212,7 +212,7 @@
                                     </div>
                                     <!-- 删除任务 -->
                                     <form class="ui compact buttons" id="delForm-{{$JobID}}"
-                                        action='{{AppSubUrl}}/{{.Repo.OwnerName}}/{{.Repo.Name}}{{if eq .JobType "BENCHMARK"}}/cloudbrain/benchmark{{else if or (eq .JobType "SNN4IMAGENET") (eq .JobType "BRAINSCORE")}}/cloudbrain{{else if eq .JobType "DEBUG"}}{{if eq .ComputeResource "NPU"}}/modelarts/notebook{{else}}/cloudbrain{{end}}{{else if eq .JobType "TRAIN"}}{{if eq .ComputeResource "NPU"}}/modelarts/notebook{{else}}/cloudbrain{{end}}/train-job{{end}}/{{$JobID}}/del?isadminpage=true'
+                                        action='{{AppSubUrl}}/{{.Repo.OwnerName}}/{{.Repo.Name}}{{if eq .JobType "BENCHMARK"}}/cloudbrain/benchmark{{else if or (eq .JobType "SNN4IMAGENET") (eq .JobType "BRAINSCORE")}}/cloudbrain{{else if eq .JobType "DEBUG"}}{{if eq .ComputeResource "NPU"}}/modelarts/notebook{{else}}/cloudbrain{{end}}{{else if eq .JobType "TRAIN"}}{{if eq .Cloudbrain.Type 1}}/modelarts/train-job{{else if eq .Cloudbrain.Type 0}}/cloudbrain/train-job{{else if eq .Cloudbrain.Type 2}}/grampus/train-job{{end}}{{end}}/{{$JobID}}/del?isadminpage=true'
                                         method="post">
                                         {{$.CsrfTokenHtml}}
                                         <a style="padding: 0.5rem 1rem;margin-left:0.2rem" id="ai-delete-{{$JobID}}"