feat(config): 新增集群区分AI功能和HPC功能配置 (#1148)

###改动集群配置文件新增配置项，配置为true时才会在对应的门户系统展示集群选项 ```yaml hpc: enabled: true ai: enabled: false ``` AI 跳转由门户改为 SCOW HPC ![b473480d-8392-4108-a37d-84e1d27fecf0](https://github.com/PKUHPC/SCOW/assets/130351655/8a5a67d5-5253-428f-8b13-41dad0b031b0) 管理系统跳转门户改为SCOW HPC 跳转AI 为 SCOW AI ![cd0ab7ac-f5bc-4bb9-92ad-1e6f8ff67ee8](https://github.com/PKUHPC/SCOW/assets/130351655/4864336a-1055-40ae-9e4d-e8d99a6dfc3e)
PKUHPC · Mar 5, 2024 · 02d6a18 · 02d6a18
1 parent f02687a
commit 02d6a18
Show file tree

Hide file tree

Showing 17 changed files with 92 additions and 14 deletions.
diff --git a/.changeset/dry-spoons-sit.md b/.changeset/dry-spoons-sit.md
@@ -0,0 +1,5 @@
+---
+"@scow/config": patch
+---
+
+集群配置新增 hpc 和 ai enabled属性 区分 Ai 集群和 HPC 集群或者是融合集群
diff --git a/.changeset/proud-planes-give.md b/.changeset/proud-planes-give.md
@@ -0,0 +1,8 @@
+---
+"@scow/portal-server": patch
+"@scow/portal-web": patch
+"@scow/mis-web": patch
+"@scow/ai": patch
+---
+
+新增集群区分 AI 功能和 HPC 功能配置
diff --git a/apps/ai/src/app/(auth)/jobs/[clusterId]/LaunchAppForm.tsx b/apps/ai/src/app/(auth)/jobs/[clusterId]/LaunchAppForm.tsx
@@ -366,7 +366,9 @@ export const LaunchAppForm = (props: Props) => {
             account: account,
             partition: partition,
             nodeCount: nodeCount,
-            coreCount: coreCount,
+            coreCount: gpuCount ?
+              gpuCount * Math.floor(currentPartitionInfo!.cores / currentPartitionInfo!.gpus) :
+              coreCount,
             gpuCount: gpuCount,
             maxTime: maxTime,
             memory: memorySize,

diff --git a/apps/ai/src/app/(auth)/layout.tsx b/apps/ai/src/app/(auth)/layout.tsx
@@ -95,7 +95,7 @@ export default function Layout(
             icon={<DesktopOutlined style={{ paddingRight: 2 }} />}
             link={publicConfig.PORTAL_URL}
             // linkText={t("baseLayout.linkTextAI")}
-            linkText="门户"
+            linkText="SCOW HPC"
           />
           {/* {
             systemLanguageConfig.isUsingI18n ? (

diff --git a/apps/ai/src/server/config/clusters.ts b/apps/ai/src/server/config/clusters.ts
@@ -13,5 +13,5 @@
 import { getClusterConfigs } from "@scow/config/build/cluster";
 import { logger } from "src/server/utils/logger";
 
-export const clusters = getClusterConfigs(undefined, logger);
+export const clusters = getClusterConfigs(undefined, logger, ["ai"]);
 
diff --git a/apps/ai/src/server/trpc/route/config.ts b/apps/ai/src/server/trpc/route/config.ts
@@ -31,7 +31,7 @@ import { z } from "zod";
 
 
 const configPath = USE_MOCK ? join(__dirname, "config") : undefined;
-const clustersInit = getClusterConfigs(configPath, console);
+const clustersInit = getClusterConfigs(configPath, console, ["ai"]);
 Object.keys(clustersInit).map((id) => clustersInit[id].loginNodes = clustersInit[id].loginNodes.map(getLoginNode));
 
 export const clusters = clustersInit;

diff --git a/apps/cli/assets/init-full/config/clusters/hpc01.yaml b/apps/cli/assets/init-full/config/clusters/hpc01.yaml
@@ -23,3 +23,11 @@ adapterUrl: localhost:8972
 
 #   # 是否在这个机器上自动配置nginx
 #   autoSetupNginx: true
+
+# 集群在HPC或是否启用，默认为true
+hpc:
+  enabled: true
+
+# 集群在AI或是否启用，默认为false
+ai:
+  enabled: false
diff --git a/apps/mis-web/src/i18n/en.ts b/apps/mis-web/src/i18n/en.ts
@@ -156,7 +156,7 @@ export default {
         operationLog: "Operation Log",
         statistic: "Statistic",
       },
-      navLinkTextPortal: "Portal",
+      navLinkTextPortal: "SCOW HPC",
       navLinkTextAI: "SCOW AI",
       dashboard: "Dashboard",
       user: {

diff --git a/apps/mis-web/src/i18n/zh_cn.ts b/apps/mis-web/src/i18n/zh_cn.ts
@@ -156,7 +156,7 @@ export default {
         operationLog:"操作日志",
         statistic: "平台数据统计",
       },
-      navLinkTextPortal: "门户",
+      navLinkTextPortal: "SCOW HPC",
       navLinkTextAI: "SCOW AI",
       dashboard: "仪表盘",
       user: {

diff --git a/apps/portal-server/src/config/clusters.ts b/apps/portal-server/src/config/clusters.ts
@@ -13,5 +13,5 @@
 import { getClusterConfigs } from "@scow/config/build/cluster";
 import { logger } from "src/utils/logger";
 
-export const clusters = getClusterConfigs(undefined, logger);
+export const clusters = getClusterConfigs(undefined, logger, ["hpc"]);
 
diff --git a/apps/portal-server/src/utils/desktops.ts b/apps/portal-server/src/utils/desktops.ts
@@ -24,7 +24,8 @@ import { Logger } from "ts-log";
 
 export function getDesktopConfig(cluster: string): LoginDeskopConfigSchema {
 
-  return { ...getPortalConfig().loginDesktop, ...getClusterConfigs()[cluster].loginDesktop };
+  return { ...getPortalConfig().loginDesktop,
+    ...getClusterConfigs(undefined, undefined, ["hpc"])[cluster].loginDesktop };
 }
 
 export function ensureEnabled(cluster: string) {

diff --git a/apps/portal-server/src/utils/turbovnc.ts b/apps/portal-server/src/utils/turbovnc.ts
@@ -22,7 +22,7 @@ export function getTurboVNCPath(cluster: string) {
 
   const commonTurboVNCPath = getPortalConfig().turboVNCPath;
 
-  const clusterTurboVNCPath = getClusterConfigs()[cluster].turboVNCPath;
+  const clusterTurboVNCPath = getClusterConfigs(undefined, undefined, ["hpc"])[cluster].turboVNCPath;
 
   return clusterTurboVNCPath || commonTurboVNCPath;
 

diff --git a/apps/portal-web/config.js b/apps/portal-web/config.js
@@ -130,7 +130,7 @@ const buildRuntimeConfig = async (phase, basePath) => {
 
   const configPath = mockEnv ? join(__dirname, "config") : undefined;
 
-  const clusters = getClusterConfigs(configPath, console);
+  const clusters = getClusterConfigs(configPath, console, ["hpc"]);
 
   Object.keys(clusters).map((id) => clusters[id].loginNodes = clusters[id].loginNodes.map(getLoginNode));
 

diff --git a/docs/docs/deploy/config/cluster-config.md b/docs/docs/deploy/config/cluster-config.md
@@ -55,7 +55,14 @@ crossClusterFileTransfer:
   enabled: true
   # 传输节点的地址(ip地址:端口号)
   transferNode: localhost:22222
-
+
+# 集群在HPC或是否启用，默认为true
+hpc:
+  enabled: true
+
+# 集群在AI或是否启用，默认为false
+ai:
+  enabled: false
 ```
 
 ## 注意

diff --git a/libs/config/package.json b/libs/config/package.json
@@ -18,6 +18,9 @@
     "@scow/lib-config": "workspace:*",
     "@sinclair/typebox": "0.32.13"
   },
+  "devDependencies": {
+    "ts-log": "2.2.5"
+  },
   "volta": {
     "extends": "../../package.json"
   }

diff --git a/libs/config/src/cluster.ts b/libs/config/src/cluster.ts
@@ -10,10 +10,11 @@
  * See the Mulan PSL v2 for more details.
  */
 
-import { GetConfigFn, getDirConfig } from "@scow/lib-config";
+import { getDirConfig } from "@scow/lib-config";
 import { Static, Type } from "@sinclair/typebox";
 import { DEFAULT_CONFIG_BASE_PATH } from "src/constants";
 import { createI18nStringSchema } from "src/i18n";
+import { Logger } from "ts-log";
 
 const CLUSTER_CONFIG_BASE_PATH = "clusters";
 
@@ -100,6 +101,15 @@ export const ClusterConfigSchema = Type.Object({
     enabled: Type.Boolean({ description: "是否开启跨集群传输功能", default: false }),
     transferNode: Type.Optional(Type.String({ description: "跨集群传输文件的节点" })),
   })),
+
+  hpc: Type.Object({
+    enabled: Type.Boolean({ description: "是否在HPC中启用" }),
+  }, { description: "集群在HPC中是否启用, 默认启用", default: { enabled: true } }),
+
+  ai: Type.Object({
+    enabled: Type.Boolean({ description: "是否在AI中启用" }),
+  }, { description: "集群在AI中是否启用, 默认不启用", default: { enabled: false } }),
+
   k8s: Type.Optional(Type.Object({
     runtime: Type.Enum(k8sRuntime, { description: "k8s 集群运行时, ai系统的镜像功能的命令取决于该值, 可选 docker 或者 containerd",
       default: "containerd" }),
@@ -109,8 +119,19 @@ export const ClusterConfigSchema = Type.Object({
 
 export type ClusterConfigSchema = Static<typeof ClusterConfigSchema>;
 
-export const getClusterConfigs: GetConfigFn<Record<string, ClusterConfigSchema>> =
-  (baseConfigPath, logger) => {
+
+export type ClusterType = "hpc" | "ai";
+
+/**
+ * @param
+ * type: 获取的集群类型，如果不传则返回所有集群，如果传入则返回指定类型的集群，例如：["hpc", "ai"] 返回所有HPC和AI集群
+ */
+export type GetClusterConfigFn<T> = (baseConfigPath?: string, logger?: Logger, type?: ClusterType[]) => T;
+
+export const getClusterConfigs: GetClusterConfigFn<Record<string, ClusterConfigSchema>> =
+  (baseConfigPath, logger, clusterType) => {
+
+    const types: ClusterType[] = clusterType ?? ["hpc", "ai"];
 
     const config = getDirConfig(
       ClusterConfigSchema,
@@ -144,5 +165,24 @@ export const getClusterConfigs: GetConfigFn<Record<string, ClusterConfigSchema>>
       throw new Error("login node address must be unique across all clusters and all login nodes.");
     }
 
+
+    for (const cluster in config) {
+      if (Object.hasOwnProperty.call(config, cluster)) {
+        const clusterInfo = config[cluster];
+        if (clusterInfo) {
+          let enabled = false;
+          for (const type of types) {
+            if (clusterInfo[type].enabled) {
+              enabled = true;
+              break;
+            }
+          }
+          if (!enabled) {
+            delete config[cluster];
+          }
+        }
+      }
+    }
+
     return config;
   };
diff --git a/pnpm-lock.yaml b/pnpm-lock.yaml