fix(portal): 作业提交页面修复并优化模板时间应用 (#1302)

![image](https://github.com/PKUHPC/SCOW/assets/78541912/8b9ac8ed-6020-4a6c-8526-47b8d1296073) ### 要修复问题 1.从模板值中获取到的时间无法填入 2.因为新增了时间单位，模板文件中保存的只有分钟所以无法换算回分钟以外的单位 <img width="302" alt="1d0a79840fb94fee37610e85dc107bc" src="https://github.com/PKUHPC/SCOW/assets/78541912/0f47fec2-a763-4c91-81a1-dbbe1d67a4ce"> 3.最长运行时间为空时，提示不随中英文切换变动 ### 修复方法 1.增加最长运行时间的时间单位 ![image](https://github.com/PKUHPC/SCOW/assets/78541912/c180c218-ddcb-4f9a-8fa9-d138e077f0f2) 2.最长运行时间为空时，提示随中英文切换变动 ![image](https://github.com/PKUHPC/SCOW/assets/78541912/5e740f49-51d8-4ee5-ad18-59d6fe55a33c) --------- Co-authored-by: Chen Junda <[email protected]>
PKUHPC · Jun 20, 2024 · 3558bd4 · 3558bd4
1 parent a0e9199
commit 3558bd4
Show file tree

Hide file tree

Showing 13 changed files with 99 additions and 54 deletions.
diff --git a/.changeset/olive-rings-heal.md b/.changeset/olive-rings-heal.md
@@ -0,0 +1,6 @@
+---
+"@scow/portal-server": patch
+"@scow/portal-web": patch
+---
+
+提交作业保存作业模板时最长运行时间的单位也保存入模板中
diff --git a/.changeset/spotty-carpets-smile.md b/.changeset/spotty-carpets-smile.md
@@ -0,0 +1,5 @@
+---
+"@scow/grpc-api": minor
+---
+
+JobTemplate 与 ListAllJobsResponse 增加 max_time_unit 可选字段
diff --git a/apps/portal-server/src/clusterops/api/job.ts b/apps/portal-server/src/clusterops/api/job.ts
@@ -10,6 +10,7 @@
  * See the Mulan PSL v2 for more details.
  */
 
+import { TimeUnit } from "@scow/protos/build/portal/job";
 import { Logger } from "ts-log";
 
 
@@ -30,6 +31,7 @@ export interface JobTemplate {
   memory?: string;
   comment?: string | undefined;
   scriptOutput?: string | undefined;
+  maxTimeUnit?: TimeUnit | undefined;
 }
 
 export interface ListJobTemplatesRequest {

diff --git a/apps/portal-server/src/services/job.ts b/apps/portal-server/src/services/job.ts
@@ -17,7 +17,7 @@ import { Status } from "@grpc/grpc-js/build/src/constants";
 import { jobInfoToPortalJobInfo, jobInfoToRunningjob } from "@scow/lib-scheduler-adapter";
 import { checkSchedulerApiVersion } from "@scow/lib-server";
 import { createDirectoriesRecursively, sftpReadFile, sftpStat, sftpWriteFile } from "@scow/lib-ssh";
-import { AccountStatusFilter, JobServiceServer, JobServiceService } from "@scow/protos/build/portal/job";
+import { AccountStatusFilter, JobServiceServer, JobServiceService, TimeUnit } from "@scow/protos/build/portal/job";
 import { parseErrorDetails } from "@scow/rich-error-model";
 import { ApiVersion } from "@scow/utils/build/version";
 import path, { join } from "path";
@@ -228,8 +228,8 @@ export const jobServiceServer = plugin((server) => {
     },
 
     submitJob: async ({ request, logger }) => {
-      const { cluster, command, jobName, coreCount, gpuCount, maxTime, saveAsTemplate, userId,
-        nodeCount, partition, qos, account, comment, workingDirectory, output
+      const { cluster, command, jobName, coreCount, gpuCount, maxTime, maxTimeUnit = TimeUnit.MINUTES,
+        saveAsTemplate, userId, nodeCount, partition, qos, account, comment, workingDirectory, output
         , errorOutput, memory, scriptOutput } = request;
       await checkActivatedClusters({ clusterIds: cluster });
 
@@ -240,13 +240,18 @@ export const jobServiceServer = plugin((server) => {
         const sftp = await ssh.requestSFTP();
         await createDirectoriesRecursively(sftp, workingDirectory);
       });
-
+      const timeUnitConversion = {
+        [TimeUnit.MINUTES]: 1,
+        [TimeUnit.HOURS]: 60,
+        [TimeUnit.DAYS]: 60 * 24,
+      };
+      const maxTimeConversion = maxTime * (timeUnitConversion[maxTimeUnit]);
       const reply = await callOnOne(
         cluster,
         logger,
         async (client) => await asyncClientCall(client.job, "submitJob", {
           userId, jobName, account, partition: partition!, qos, nodeCount, gpuCount: gpuCount || 0,
-          memoryMb: Number(memory?.split("M")[0]), coreCount, timeLimitMinutes: maxTime,
+          memoryMb: Number(memory?.split("M")[0]), coreCount, timeLimitMinutes: maxTimeConversion,
           script: command, workingDirectory, stdout: output, stderr: errorOutput, extraOptions: [],
         }).catch((e) => {
           const ex = e as ServiceError;
@@ -289,6 +294,7 @@ export const jobServiceServer = plugin((server) => {
           errorOutput,
           memory,
           scriptOutput,
+          maxTimeUnit,
         };
 
 

diff --git a/apps/portal-web/src/apis/api.mock.ts b/apps/portal-web/src/apis/api.mock.ts
@@ -15,6 +15,7 @@ import { ClusterActivationStatus } from "@scow/config/build/type";
 import type { RunningJob } from "@scow/protos/build/common/job";
 import { JobInfo } from "@scow/protos/build/portal/job";
 import { api } from "src/apis/api";
+import { TimeUnit } from "src/models/job";
 export type MockApi<TApi extends Record<
   string,
  (...args: any[]) => JsonFetchResultPromiseLike<any>>
@@ -198,6 +199,7 @@ export const mockApi: MockApi<typeof api> = {
       output: "job.%j.out",
       errorOutput: "job.%j.err",
       workingDirectory: "/nfs/jobs/123",
+      maxTimeUnit:  TimeUnit.MINUTES,
     },
   }),
 

diff --git a/apps/portal-web/src/i18n/en.ts b/apps/portal-web/src/i18n/en.ts
@@ -181,6 +181,7 @@ export default {
         gpuCount: "Number of GPU Cards per Node",
         coreCount: "Number of CPU Cores per Node",
         maxTime: "Maximum Running Time",
+        requireMaxTime: "Please enter the maximum runtime",
         minute: "Minutes",
         hours:"Hours",
         days: "Days",

diff --git a/apps/portal-web/src/i18n/zh_cn.ts b/apps/portal-web/src/i18n/zh_cn.ts
@@ -181,6 +181,7 @@ export default {
         gpuCount: "单节点GPU卡数",
         coreCount: "单节点核心数",
         maxTime: "最长运行时间",
+        requireMaxTime:"请输入最长运行时间",
         minute: "分钟",
         hours:"小时",
         days: "天",

diff --git a/apps/portal-web/src/models/job.ts b/apps/portal-web/src/models/job.ts
@@ -86,3 +86,8 @@ export enum AccountStatusFilter {
   UNBLOCKED_ONLY = 2,
 };
 
+export enum TimeUnit {
+  MINUTES = 0,
+  HOURS = 1,
+  DAYS = 2,
+}
diff --git a/apps/portal-web/src/pageComponents/job/SubmitJobForm.tsx b/apps/portal-web/src/pageComponents/job/SubmitJobForm.tsx
@@ -23,7 +23,7 @@ import { SingleClusterSelector } from "src/components/ClusterSelector";
 import { CodeEditor } from "src/components/CodeEditor";
 import { ClusterNotAvailablePage } from "src/components/errorPages/ClusterNotAvailablePage";
 import { prefix, useI18nTranslateToString } from "src/i18n";
-import { AccountStatusFilter } from "src/models/job";
+import { AccountStatusFilter, TimeUnit } from "src/models/job";
 import { FileSelectModal } from "src/pageComponents/job/FileSelectModal";
 import { Partition } from "src/pages/api/cluster";
 import { ClusterInfoStore } from "src/stores/ClusterInfoStore";
@@ -42,8 +42,8 @@ interface JobForm {
   command: string;
   jobName: string;
   qos: string | undefined;
-  maxTimeValue: number;
-  maxTimeUnit: "minutes" | "hours";
+  maxTime: number;
+  maxTimeUnit: TimeUnit | undefined;
   account: string;
   comment: string;
   workingDirectory: string;
@@ -70,13 +70,13 @@ const initialValues = {
   nodeCount: 1,
   coreCount: 1,
   gpuCount: 1,
-  maxTimeValue: 30,
-  maxTimeUnit:"minutes",
+  maxTime: 30,
+  maxTimeUnit: TimeUnit.MINUTES,
   output: "job.%j.out",
-  scriptOutput:"job.%j.sh",
+  scriptOutput: "job.%j.sh",
   errorOutput: "job.%j.err",
   save: false,
-  showScriptOutput:true,
+  showScriptOutput: true,
 } as Partial<JobForm>;
 
 interface Props {
@@ -96,24 +96,22 @@ export const SubmitJobForm: React.FC<Props> = ({ initial = initialValues, submit
 
 
   const cluster = Form.useWatch("cluster", form) as Cluster | undefined;
-  const timeUnitConversion = {
-    minutes: 1,
-    hours: 60,
-    days: 60 * 24,
-  };
   const submit = async () => {
     const formValues = await form.validateFields();
     const { cluster, command, jobName, coreCount, gpuCount, workingDirectory, output, errorOutput, save,
-      maxTimeValue, maxTimeUnit, nodeCount, partition, qos, account, comment, showScriptOutput } = formValues;
+      maxTime, maxTimeUnit, nodeCount, partition, qos, account, comment, showScriptOutput } = formValues;
     const scriptOutput = showScriptOutput ? formValues.scriptOutput : "";
-    const maxTime = maxTimeValue * (timeUnitConversion[maxTimeUnit] || 1);
-    await api.submitJob({ body: {
-      cluster: cluster.id, command, jobName, account,
-      coreCount: gpuCount ? gpuCount * Math.floor(currentPartitionInfo!.cores / currentPartitionInfo!.gpus) : coreCount,
-      gpuCount,
-      maxTime, nodeCount, partition, qos, comment,
-      workingDirectory, save, memory, output, errorOutput, scriptOutput,
-    } })
+
+    await api.submitJob({
+      body: {
+        cluster: cluster.id, command, jobName, account,
+        coreCount: gpuCount ? gpuCount * Math.floor(currentPartitionInfo!.cores
+          / currentPartitionInfo!.gpus) : coreCount,
+        gpuCount,
+        maxTime, maxTimeUnit, nodeCount, partition, qos, comment,
+        workingDirectory, save, memory, output, errorOutput, scriptOutput,
+      },
+    })
       .httpError(500, (e) => {
         if (e.code === "SCHEDULER_FAILED") {
           modal.error({
@@ -162,7 +160,7 @@ export const SubmitJobForm: React.FC<Props> = ({ initial = initialValues, submit
   // 获取集群信息
   const clusterInfoQuery = useAsync({
     promiseFn: useCallback(async () => cluster
-      ? api.getClusterInfo({ query: { cluster:  cluster?.id } })
+      ? api.getClusterInfo({ query: { cluster: cluster?.id } })
       : undefined, [cluster]),
     onResolve: () => {
       const jobInitialName = genJobName();
@@ -247,12 +245,14 @@ export const SubmitJobForm: React.FC<Props> = ({ initial = initialValues, submit
       setAccountPartitionsCacheMap({});
       handlePartitionCacheMap({});
 
-      return cluster ? await api.getAccounts({ query: {
-        cluster: cluster.id,
-        statusFilter: AccountStatusFilter.UNBLOCKED_ONLY,
-      } })
+      return cluster ? await api.getAccounts({
+        query: {
+          cluster: cluster.id,
+          statusFilter: AccountStatusFilter.UNBLOCKED_ONLY,
+        },
+      })
         .httpError(404, (error) => { message.error(error.message); })
-        : { accounts: [] as string [] };
+        : { accounts: [] as string[] };
     }, [cluster, accountsReloadTrigger]),
     onResolve: (data) => {
 
@@ -279,10 +279,12 @@ export const SubmitJobForm: React.FC<Props> = ({ initial = initialValues, submit
       const account = form.getFieldValue("account");
       if (cluster && account && selectableAccounts.includes(account) && !accountPartitionsCacheMap[account]) {
         const newPartitionsMap = { ...accountPartitionsCacheMap };
-        return await api.getAvailablePartitionsForCluster({ query: {
-          cluster: cluster?.id,
-          accountName: account,
-        } })
+        return await api.getAvailablePartitionsForCluster({
+          query: {
+            cluster: cluster?.id,
+            accountName: account,
+          },
+        })
           .then((data) => {
             newPartitionsMap[account] = data.partitions;
             // 如果第一次请求时模板值中分区存在，则填入模板值的分区及qos
@@ -360,8 +362,8 @@ export const SubmitJobForm: React.FC<Props> = ({ initial = initialValues, submit
 
   const memorySize = (currentPartitionInfo ?
     currentPartitionInfo.gpus ? nodeCount * gpuCount
-    * Math.floor(currentPartitionInfo.cores / currentPartitionInfo.gpus)
-    * Math.floor(currentPartitionInfo.memMb / currentPartitionInfo.cores) :
+      * Math.floor(currentPartitionInfo.cores / currentPartitionInfo.gpus)
+      * Math.floor(currentPartitionInfo.memMb / currentPartitionInfo.cores) :
       nodeCount * coreCount * Math.floor(currentPartitionInfo.memMb / currentPartitionInfo.cores) : 0);
   const memory = memorySize + "MB";
   const memoryDisplay = formatSize(memorySize, ["MB", "GB", "TB"]);
@@ -410,10 +412,10 @@ export const SubmitJobForm: React.FC<Props> = ({ initial = initialValues, submit
             dependencies={["cluster"]}
           >
             {/* 加载完集群后再加载账户，保证initial值能被赋值成功 */}
-            { cluster?.id && unblockedAccountsQuery?.data?.accounts &&
+            {cluster?.id && unblockedAccountsQuery?.data?.accounts &&
               (
                 <AccountListSelector
-                  selectableAccounts={ selectableAccounts ?? []}
+                  selectableAccounts={selectableAccounts ?? []}
                   isLoading={unblockedAccountsQuery.isLoading}
                   onReload={handleAccountsReload}
                   onChange={handleAccountChange}
@@ -510,7 +512,11 @@ export const SubmitJobForm: React.FC<Props> = ({ initial = initialValues, submit
         <Col span={24} sm={6}>
           <Form.Item label={t(p("maxTime"))} required>
             <Input.Group compact style={{ display: "flex", minWidth: "120px" }}>
-              <Form.Item name="maxTimeValue" rules={[{ required: true }]} noStyle>
+              <Form.Item
+                name="maxTime"
+                rules={[{ required: true, message: `${t(p("requireMaxTime"))}` }]}
+                noStyle
+              >
                 <InputNumber
                   min={1}
                   step={1}
@@ -523,9 +529,9 @@ export const SubmitJobForm: React.FC<Props> = ({ initial = initialValues, submit
                   popupMatchSelectWidth={false}
                   style={{ flex: "0 1 auto" }}
                 >
-                  <Select.Option value="minutes">{t(p("minute"))}</Select.Option>
-                  <Select.Option value="hours">{t(p("hours"))}</Select.Option>
-                  <Select.Option value="days">{t(p("days"))}</Select.Option>
+                  <Select.Option value={TimeUnit.MINUTES}>{t(p("minute"))}</Select.Option>
+                  <Select.Option value={TimeUnit.HOURS}>{t(p("hours"))}</Select.Option>
+                  <Select.Option value={TimeUnit.DAYS}>{t(p("days"))}</Select.Option>
                 </Select>
               </Form.Item>
             </Input.Group>

diff --git a/apps/portal-web/src/pages/api/job/getJobTemplate.ts b/apps/portal-web/src/pages/api/job/getJobTemplate.ts
@@ -13,7 +13,7 @@
 import { typeboxRouteSchema } from "@ddadaal/next-typed-api-routes-runtime";
 import { asyncUnaryCall } from "@ddadaal/tsgrpc-client";
 import { status } from "@grpc/grpc-js";
-import { JobServiceClient } from "@scow/protos/build/portal/job";
+import { JobServiceClient, TimeUnit } from "@scow/protos/build/portal/job";
 import { Static, Type } from "@sinclair/typebox";
 import { authenticate } from "src/auth/server";
 import { getClient } from "src/utils/client";
@@ -29,14 +29,14 @@ export const JobTemplate = Type.Object({
   nodeCount: Type.Number(),
   coreCount: Type.Number(),
   gpuCount: Type.Optional(Type.Number()),
-  /** in minutes */
-  maxTime: Type.Number(),
+  maxTime: Type.Number(), // 最长运行时间
   command: Type.String(),
   workingDirectory: Type.String(),
   output: Type.Optional(Type.String()),
   errorOutput: Type.Optional(Type.String()),
   comment: Type.Optional(Type.String()),
   scriptOutput:Type.Optional(Type.String()),
+  maxTimeUnit: Type.Optional(Type.Enum(TimeUnit)), // 最长运行时间单位
 });
 export type JobTemplate = Static<typeof JobTemplate>;
 

diff --git a/apps/portal-web/src/pages/api/job/submitJob.ts b/apps/portal-web/src/pages/api/job/submitJob.ts
@@ -13,7 +13,7 @@
 import { typeboxRouteSchema } from "@ddadaal/next-typed-api-routes-runtime";
 import { asyncUnaryCall } from "@ddadaal/tsgrpc-client";
 import { status } from "@grpc/grpc-js";
-import { JobServiceClient } from "@scow/protos/build/portal/job";
+import { JobServiceClient, TimeUnit } from "@scow/protos/build/portal/job";
 import { Static, Type } from "@sinclair/typebox";
 import { authenticate } from "src/auth/server";
 import { OperationResult, OperationType } from "src/models/operationLog";
@@ -40,6 +40,7 @@ export const SubmitJobInfo = Type.Object({
   comment: Type.Optional(Type.String()),
   save: Type.Boolean(),
   scriptOutput:Type.Optional(Type.String()),
+  maxTimeUnit:Type.Optional(Type.Enum(TimeUnit)),
 });
 
 export type SubmitJobInfo = Static<typeof SubmitJobInfo>;
@@ -78,7 +79,7 @@ export default route(SubmitJobSchema, async (req, res) => {
 
   if (!info) { return; }
 
-  const { cluster, command, jobName, coreCount, gpuCount, maxTime, save,
+  const { cluster, command, jobName, coreCount, gpuCount, maxTime, maxTimeUnit, save,
     nodeCount, partition, qos, account, comment
     , workingDirectory, output, errorOutput, scriptOutput, memory } = req.body;
 
@@ -99,6 +100,7 @@ export default route(SubmitJobSchema, async (req, res) => {
     coreCount,
     gpuCount,
     maxTime,
+    maxTimeUnit,
     nodeCount,
     partition,
     qos,

diff --git a/apps/portal-web/src/pages/jobs/submit.tsx b/apps/portal-web/src/pages/jobs/submit.tsx
@@ -21,6 +21,7 @@ import { api } from "src/apis";
 import { requireAuth } from "src/auth/requireAuth";
 import { PageTitle } from "src/components/PageTitle";
 import { useI18nTranslateToString } from "src/i18n";
+import { TimeUnit } from "src/models/job";
 import { SubmitJobForm } from "src/pageComponents/job/SubmitJobForm";
 import { ClusterInfoStore } from "src/stores/ClusterInfoStore";
 import { getServerI18nConfigText, publicConfig } from "src/utils/config";
@@ -62,13 +63,13 @@ export const SubmitJobPage: NextPage<Props> = requireAuth(() => true)(
               output: template.output,
               errorOutput: template.errorOutput,
               save: false,
-              scriptOutput:template.scriptOutput,
+              scriptOutput: template.scriptOutput,
+              maxTimeUnit: template.maxTimeUnit || TimeUnit.MINUTES,
             }));
         } else {
           return undefined;
         }
-      },
-      [cluster, jobTemplateId]),
+      }, [cluster, jobTemplateId]),
     });
 
     const t = useI18nTranslateToString();