feat(ai): ai 增加vnc， shell 到容器， ai提交作业优化 (#1202)

# ai 提交作业优化 1. 支持多个挂载点 2. 镜像支持手动输入 3. 已分享的镜像，数据集，以及模型，选取私有时，取privatePath而不是分享后的路径 4. 将提交作业时ai相关的参数都放入extra_options里 # ai 新增以shell 的方式进入容器的功能 ai 新增进入训练中的作业的容器并执行 shell 操作的功能。该功能依赖于 k8s 的 api server，所以需要一份 kubectl config 配置文件。 ![image](https://github.com/PKUHPC/SCOW/assets/140392039/8c01af88-ffac-41fd-9600-1658a7b6c24d) ![image](https://github.com/PKUHPC/SCOW/assets/140392039/1e9b433d-b18c-4cc1-88c1-b667f70c2070) # ai 增加 vnc Ai模块新增vnc应用 --------- Co-authored-by: Miracle575 <longsijie@icode.pku.edu.cn>
PKUHPC · Apr 23, 2024 · e312efb · e312efb
1 parent a4d36e2
commit e312efb
Show file tree

Hide file tree

Showing 31 changed files with 1,326 additions and 163 deletions.
diff --git a/.changeset/lemon-fans-peel.md b/.changeset/lemon-fans-peel.md
@@ -0,0 +1,7 @@
+---
+"@scow/config": patch
+"@scow/cli": patch
+"@scow/ai": patch
+---
+
+AI 模块支持创建 vnc 类型应用
diff --git a/.changeset/pretty-toes-greet.md b/.changeset/pretty-toes-greet.md
@@ -0,0 +1,10 @@
+---
+"@scow/scheduler-adapter-protos": patch
+"@scow/test-adapter": patch
+"@scow/config": patch
+"@scow/cli": patch
+"@scow/ai": patch
+"@scow/docs": patch
+---
+
+ai 增加 vnc 功能，以 shell 方式进入容器功能和提交作业的优化
diff --git a/.changeset/witty-camels-share.md b/.changeset/witty-camels-share.md
@@ -0,0 +1,5 @@
+---
+"@scow/ai": patch
+---
+
+ai 新增以 shell 的方式进入容器的功能
diff --git a/apps/ai/assets/app/vnc_entry.sh b/apps/ai/assets/app/vnc_entry.sh
@@ -1 +1,5 @@
-//TODO
+#!/bin/bash
+
+export PORT=$1
+export HOST=$2
+export SVCPORT=$3
diff --git a/apps/ai/next.config.mjs b/apps/ai/next.config.mjs
@@ -27,7 +27,7 @@ export default async () => {
     // HACK setup ws proxy
     setTimeout(() => {
       const url = `http://localhost:${process.env.PORT || 3000}${join(BASE_PATH, "/api/setup")}`;
-      console.log("Calling setup url to initialize proxy and shell server", url);
+      console.log("Calling setup url to initialize proxy and job shell server", url);
 
       fetch(url).then(async (res) => {
         console.log("Call completed. Response: ", await res.text());

diff --git a/apps/ai/package.json b/apps/ai/package.json
@@ -37,6 +37,7 @@
     "@ddadaal/tsgrpc-client": "0.17.7",
     "@ddadaal/tsgrpc-common": "0.2.5",
     "@grpc/grpc-js": "1.10.6",
+    "@kubernetes/client-node": "^0.20.0",
     "@mikro-orm/cli": "6.1.12",
     "@mikro-orm/core": "6.1.12",
     "@mikro-orm/migrations": "6.1.12",
@@ -48,9 +49,9 @@
     "@scow/lib-decimal": "workspace:*",
     "@scow/lib-operation-log": "workspace:*",
     "@scow/lib-scheduler-adapter": "workspace:*",
+    "@scow/lib-server": "workspace:*",
     "@scow/lib-ssh": "workspace:*",
     "@scow/lib-web": "workspace:*",
-    "@scow/lib-server": "workspace:*",
     "@scow/scheduler-adapter-protos": "workspace:*",
     "@scow/utils": "workspace:*",
     "@scow/rich-error-model": "workspace:*",
@@ -87,8 +88,8 @@
     "swagger-ui-react": "5.13.0",
     "trpc-openapi": "1.2.0",
     "ws": "8.16.0",
-    "xterm": "5.3.0",
-    "xterm-addon-fit": "0.8.0",
+    "@xterm/xterm": "5.5.0",
+    "@xterm/addon-fit": "0.10.0",
     "zod": "3.22.4",
     "shell-quote": "1.8.1",
     "replace-in-file": "7.1.0"

diff --git a/apps/ai/src/app/(auth)/dashboard/page.tsx b/apps/ai/src/app/(auth)/dashboard/page.tsx
@@ -12,8 +12,8 @@
 
 "use client";
 import { join } from "path";
+import { usePublicConfig } from "src/app/(auth)/context";
 import { Head } from "src/utils/head";
-import { trpc } from "src/utils/trpc";
 import { styled } from "styled-components";
 
 const Logo = styled.div`
@@ -25,24 +25,22 @@ const Logo = styled.div`
 
 export default function Page() {
 
-  const { data } = trpc.config.publicConfig.useQuery();
+  const { publicConfig: { BASE_PATH } } = usePublicConfig();
 
   return (
     <div>
       <Head title={"dashboard"} />
       {
-        data ? (
-          <Logo>
-            <img
-              alt="logo"
-              src={join(data.BASE_PATH, "/api/logo?type=logo")}
-              style={{
-                objectFit: "contain",
-                maxWidth: "50%",
-              }}
-            />
-          </Logo>
-        ) : undefined
+        <Logo>
+          <img
+            alt="logo"
+            src={join(BASE_PATH, "/api/logo?type=logo")}
+            style={{
+              objectFit: "contain",
+              maxWidth: "50%",
+            }}
+          />
+        </Logo>
       }
     </div>
   );

diff --git a/apps/ai/src/app/(auth)/jobShell/[clusterId]/[jobId]/page.tsx b/apps/ai/src/app/(auth)/jobShell/[clusterId]/[jobId]/page.tsx
@@ -0,0 +1,94 @@
+/**
+ * Copyright (c) 2022 Peking University and Peking University Institute for Computing and Digital Economy
+ * SCOW is licensed under Mulan PSL v2.
+ * You can use this software according to the terms and conditions of the Mulan PSL v2.
+ * You may obtain a copy of Mulan PSL v2 at:
+ *          http://license.coscl.org.cn/MulanPSL2
+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND,
+ * EITHER EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT,
+ * MERCHANTABILITY OR FIT FOR A PARTICULAR PURPOSE.
+ * See the Mulan PSL v2 for more details.
+ */
+
+"use client";
+
+import "xterm/css/xterm.css";
+
+import { Button, Space } from "antd";
+import dynamic from "next/dynamic";
+import { usePublicConfig } from "src/app/(auth)/context";
+import { Head } from "src/utils/head";
+import { styled } from "styled-components";
+
+const Container = styled.div`
+  position: fixed;
+  left: 0;
+  top: 0;
+  height: 100%;
+  width: 100%;
+  z-index: 2000;
+  display: flex;
+  flex-direction: column;
+`;
+
+const Header = styled.div`
+  padding: 8px 16px;
+  display: flex;
+  justify-content: space-between;
+  background-color: #333;
+
+  h2 { color: white; margin: 0px; }
+
+  .ant-popover-content p {
+    margin: 0;
+  }
+`;
+
+
+const TerminalContainer = styled.div`
+  display: flex;
+  flex: 1;
+  height: 100%;
+`;
+
+const Black = styled.div`
+  height: 100%;
+  background-color: black;
+`;
+
+const JobShellComponent = dynamic(
+  () => import("src/components/shell/JobShell").then((x) => x.JobShell), {
+    ssr: false,
+    loading: Black,
+  });
+
+export default function Page({ params }: {params: {clusterId: string, jobId: string}}) {
+
+  const { clusterId, jobId } = params;
+  const { publicConfig, user } = usePublicConfig();
+
+  const clusterName = publicConfig.CLUSTERS.find((x) => x.id === clusterId)?.name || clusterId;
+
+  return (
+    <Container>
+      <Head title={`${clusterId}的终端`} />
+      <Header>
+        <h2>
+          {`用户 ${user.identityId} 连接到集群 ${clusterName} 的作业 ${jobId}`}
+        </h2>
+        <Space wrap>
+          <Button onClick={() => window.location.reload()}>
+            {"刷新并重新连接"}
+          </Button>
+        </Space>
+      </Header>
+      <TerminalContainer>
+        <JobShellComponent
+          user={user}
+          cluster={clusterId}
+          jobId={jobId}
+        />
+      </TerminalContainer>
+    </Container>
+  );
+};
diff --git a/apps/ai/src/app/(auth)/jobs/[clusterId]/AppSessionsTable.tsx b/apps/ai/src/app/(auth)/jobs/[clusterId]/AppSessionsTable.tsx
@@ -14,6 +14,7 @@
 
 import { ExclamationCircleOutlined } from "@ant-design/icons";
 import { App, Button, Checkbox, Form, Input, Popconfirm, Space, Table, TableColumnsType, Tooltip } from "antd";
+import Link from "next/link";
 import { useRouter } from "next/navigation";
 import { join } from "path";
 import React, { useCallback, useEffect, useMemo, useState } from "react";
@@ -96,7 +97,7 @@ export const AppSessionsTable: React.FC<Props> = ({ cluster, status }) => {
     {
       title: "类型",
       dataIndex: "jobType",
-      width: "10%",
+      width: "8%",
       render: (_, record) => {
         if (record.jobType === JobType.APP) {
           return "应用";
@@ -107,19 +108,20 @@ export const AppSessionsTable: React.FC<Props> = ({ cluster, status }) => {
     {
       title: "应用",
       dataIndex: "appId",
+      width: "8%",
       render: (appId: string, record) => record.appName ?? appId,
       sorter: (a, b) => (!a.submitTime || !b.submitTime) ? -1 : compareDateTime(a.submitTime, b.submitTime),
     },
     {
       title: "提交时间",
       dataIndex: "submitTime",
-      width: "15%",
+      width: "200px",
       render: (_, record) => record.submitTime ? formatDateTime(record.submitTime) : "",
     },
     {
       title: "状态",
       dataIndex: "state",
-      width: "12%",
+      width: "120px",
       render: (_, record) => (
         record.reason ? (
           <Tooltip title={record.reason}>
@@ -140,14 +142,15 @@ export const AppSessionsTable: React.FC<Props> = ({ cluster, status }) => {
     },
     ...(unfinished ? [{
       title: "剩余时间",
+      width: "100px",
       dataIndex: "remainingTime",
     },
     ] : []),
     {
       title: "操作",
       key: "action",
       fixed:"right",
-      width: "10%",
+      width: "350px",
       render: (_, record) => (
         <Space>
           {
@@ -160,6 +163,9 @@ export const AppSessionsTable: React.FC<Props> = ({ cluster, status }) => {
                     refreshToken={connectivityRefreshToken}
                   />
                 )}
+                <Link href={`/jobShell/${cluster.id}/${record.jobId}`} target="_blank">
+                  {"进入容器"}
+                </Link>
                 <Popconfirm
                   title="确定结束这个任务吗"
                   onConfirm={

diff --git a/apps/ai/src/app/(auth)/jobs/[clusterId]/ConnectToAppLink.tsx b/apps/ai/src/app/(auth)/jobs/[clusterId]/ConnectToAppLink.tsx
@@ -19,6 +19,7 @@ import { useEffect } from "react";
 import { DisabledA } from "src/components/DisabledA";
 import { AppSession } from "src/server/trpc/route/jobs/apps";
 import { trpc } from "src/utils/trpc";
+import { openDesktop } from "src/utils/vnc";
 
 import { usePublicConfig } from "../../context";
 
@@ -31,7 +32,7 @@ export interface Props {
 export const ConnectTopAppLink: React.FC<Props> = ({
   session, cluster, refreshToken,
 }) => {
-  const { publicConfig: { BASE_PATH } } = usePublicConfig();
+  const { publicConfig: { BASE_PATH, NOVNC_CLIENT_URL } } = usePublicConfig();
   const { message } = App.useApp();
 
   const { data, refetch } = trpc.jobs.checkAppConnectivity.useQuery({ clusterId: cluster, jobId: session.jobId }, {
@@ -100,7 +101,8 @@ export const ConnectTopAppLink: React.FC<Props> = ({
       }
 
     } else {
-      // TODO: vnc app
+      const { host, port, password } = reply;
+      openDesktop(BASE_PATH, NOVNC_CLIENT_URL, cluster, host, port, password);
       return;
     }