revert: "fix: ai和hpc在提交作业和应用前检查一下是否重名" (#1359)

Reverts #1356
PKUHPC · Jul 15, 2024 · 56f5a69 · 56f5a69
1 parent e9c8bfa
commit 56f5a69
Show file tree

Hide file tree

Showing 9 changed files with 6 additions and 117 deletions.
diff --git a/.changeset/tasty-files-end.md b/.changeset/tasty-files-end.md
diff --git a/apps/ai/src/server/trpc/route/jobs/apps.ts b/apps/ai/src/server/trpc/route/jobs/apps.ts
@@ -299,24 +299,6 @@ export const createAppSession = procedure
       model, mountPoints = [], account, partition, coreCount, nodeCount, gpuCount, memory,
       maxTime, workingDirectory, customAttributes } = input;
 
-    const userId = user.identityId;
-    const client = getAdapterClient(clusterId);
-
-    // 检查是否存在同名的作业
-    const existedJobName = await asyncClientCall(client.job, "getJobs", {
-      fields: ["job_id"],
-      filter: {
-        users: [userId], accounts: [],states: [],jobName:appJobName,
-      },
-    }).then((resp) => resp.jobs);
-
-    if (existedJobName.length) {
-      throw new TRPCError({
-        code: "CONFLICT",
-        message: `appJobName ${appJobName} is already existed`,
-      });
-    }
-
     const apps = getClusterAppConfigs(clusterId);
     const app = checkAppExist(apps, appId);
 
@@ -389,7 +371,7 @@ export const createAppSession = procedure
       throw clusterNotFound(clusterId);
     }
 
-
+    const userId = user.identityId;
     return await sshConnect(host, userId, logger, async (ssh) => {
       const homeDir = await getUserHomedir(ssh, userId, logger);
 
@@ -478,7 +460,7 @@ export const createAppSession = procedure
 
       // 将entry.sh写入后将路径传给适配器后启动容器
       await sftpWriteFile(sftp)(remoteEntryPath, entryScript);
-
+      const client = getAdapterClient(clusterId);
       const reply = await asyncClientCall(client.job, "submitJob", {
         userId,
         jobName: appJobName,

diff --git a/apps/ai/src/server/trpc/route/jobs/jobs.ts b/apps/ai/src/server/trpc/route/jobs/jobs.ts
@@ -128,23 +128,6 @@ procedure
         throw clusterNotFound(clusterId);
       }
 
-      const client = getAdapterClient(clusterId);
-
-      // 检查是否存在同名的作业
-      const existedJobName = await asyncClientCall(client.job, "getJobs", {
-        fields: ["job_id"],
-        filter: {
-          users: [userId], accounts: [],states: [],jobName:trainJobName,
-        },
-      }).then((resp) => resp.jobs);
-
-      if (existedJobName.length) {
-        throw new TRPCError({
-          code: "CONFLICT",
-          message: `trainJobName ${trainJobName} is already existed`,
-        });
-      }
-
       const em = await forkEntityManager();
       const {
         datasetVersion,
@@ -191,6 +174,7 @@ procedure
         const entryScript = command;
         await sftpWriteFile(sftp)(remoteEntryPath, entryScript);
 
+        const client = getAdapterClient(clusterId);
         const reply = await asyncClientCall(client.job, "submitJob", {
           userId,
           jobName: trainJobName,

diff --git a/apps/portal-server/src/clusterops/app.ts b/apps/portal-server/src/clusterops/app.ts
@@ -78,26 +78,6 @@ export const appOps = (cluster: string): AppOps => {
       const { appId, userId, account, coreCount, nodeCount, gpuCount, memory, maxTime, proxyBasePath,
         partition, qos, customAttributes, appJobName } = request;
 
-      // 检查是否存在同名的作业
-      const existedJobName = await callOnOne(
-        cluster,
-        logger,
-        async (client) => await asyncClientCall(client.job, "getJobs", {
-          fields: ["job_id"],
-          filter: {
-            users: [userId], accounts: [], states: [],jobName:appJobName,
-          },
-        }),
-      ).then((resp) => resp.jobs);
-
-      if (existedJobName.length) {
-        throw new DetailedError({
-          code: Status.ALREADY_EXISTS,
-          message: `appJobName ${appJobName} is already existed`,
-          details: [errorInfo("ALREADY EXISTS")],
-        });
-      }
-
       const memoryMb = memory ? Number(memory.slice(0, -2)) : undefined;
 
 

diff --git a/apps/portal-server/src/services/job.ts b/apps/portal-server/src/services/job.ts
@@ -233,26 +233,6 @@ export const jobServiceServer = plugin((server) => {
         , errorOutput, memory, scriptOutput } = request;
       await checkActivatedClusters({ clusterIds: cluster });
 
-      // 检查是否存在同名的作业
-      const existedJobName = await callOnOne(
-        cluster,
-        logger,
-        async (client) => await asyncClientCall(client.job, "getJobs", {
-          fields: ["job_id"],
-          filter: {
-            users: [userId], accounts: [], states: [],jobName,
-          },
-        }),
-      ).then((resp) => resp.jobs);
-
-      if (existedJobName.length) {
-        throw {
-          code: Status.ALREADY_EXISTS,
-          message: "already exists",
-          details: `jobName ${jobName} is already existed`,
-        } as ServiceError;
-      }
-
       // make sure working directory exists
       const host = getClusterLoginNode(cluster);
       if (!host) { throw clusterNotFound(cluster); }

diff --git a/apps/portal-web/src/pageComponents/app/LaunchAppForm.tsx b/apps/portal-web/src/pageComponents/app/LaunchAppForm.tsx
@@ -116,7 +116,7 @@ export const LaunchAppForm: React.FC<Props> = ({ clusterId, appId, attributes, a
       maxTime,
       customAttributes: customFormKeyValue,
     } })
-      .httpError(500, (e) => {
+      .httpError(409, (e) => {
         if (e.code === "SBATCH_FAILED") {
           createErrorModal(e.message);
         } else {
@@ -137,13 +137,6 @@ export const LaunchAppForm: React.FC<Props> = ({ clusterId, appId, attributes, a
           throw e;
         }
       })
-      .httpError(409, (e) => {
-        if (e.code === "ALREADY_EXISTS") {
-          createErrorModal(e.message);
-        } else {
-          throw e;
-        }
-      })
       .then(() => {
         message.success(t(p("successMessage")));
         Router.push(`/apps/${clusterId}/sessions`);

diff --git a/apps/portal-web/src/pageComponents/job/SubmitJobForm.tsx b/apps/portal-web/src/pageComponents/job/SubmitJobForm.tsx
@@ -132,16 +132,6 @@ export const SubmitJobForm: React.FC<Props> = ({ initial = initialValues, submit
           throw e;
         }
       })
-      .httpError(409, (e) => {
-        if (e.code === "ALREADY_EXISTS") {
-          modal.error({
-            title: t(p("errorMessage")),
-            content: e.message,
-          });
-        } else {
-          throw e;
-        }
-      })
       .then(({ jobId }) => {
         message.success(t(p("successMessage")) + jobId);
         Router.push("/jobs/runningJobs");

diff --git a/apps/portal-web/src/pages/api/app/createAppSession.ts b/apps/portal-web/src/pages/api/app/createAppSession.ts
@@ -55,17 +55,12 @@ export const CreateAppSessionSchema = typeboxRouteSchema({
       message: Type.String(),
     }),
 
-    409: Type.Object({
-      code: Type.Literal("ALREADY_EXISTS"),
-      message: Type.String(),
-    }),
-
     404: Type.Object({
       code: Type.Literal("APP_NOT_FOUND"),
       message: Type.String(),
     }),
 
-    500: Type.Object({
+    409: Type.Object({
       code: Type.Literal("SBATCH_FAILED"),
       message: Type.String(),
     }),
@@ -131,13 +126,11 @@ export default /* #__PURE__*/route(CreateAppSessionSchema, async (req, res) => {
     if (errors[0] && errors[0].$type === "google.rpc.ErrorInfo") {
       switch (errors[0].reason) {
         case "SBATCH_FAILED":
-          return { 500: { code: "SBATCH_FAILED" as const, message: ex.details } };
+          return { 409: { code: "SBATCH_FAILED" as const, message: ex.details } };
         case "NOT FOUND":
           return { 404: { code: "APP_NOT_FOUND" as const, message: ex.details } };
         case "INVALID ARGUMENT":
           return { 400: { code: "INVALID_INPUT" as const, message: ex.details } };
-        case "ALREADY EXISTS":
-          return { 409: { code: "ALREADY_EXISTS" as const, message: ex.details } };
         default:
           return e;
       }

diff --git a/apps/portal-web/src/pages/api/job/submitJob.ts b/apps/portal-web/src/pages/api/job/submitJob.ts
@@ -65,11 +65,6 @@ export const SubmitJobSchema = typeboxRouteSchema({
       message: Type.String(),
     }),
 
-    409: Type.Object({
-      code: Type.Literal("ALREADY_EXISTS"),
-      message: Type.String(),
-    }),
-
     500: Type.Object({
       code: Type.Literal("SCHEDULER_FAILED"),
       message: Type.String(),
@@ -142,7 +137,6 @@ export default route(SubmitJobSchema, async (req, res) => {
     .catch(handlegRPCError({
       [status.INTERNAL]: (err) => ({ 500: { code: "SCHEDULER_FAILED", message: err.details } } as const),
       [status.NOT_FOUND]: (err) => ({ 404: { code: "NOT_FOUND", message: err.details } } as const),
-      [status.ALREADY_EXISTS]: (err) => ({ 409: { code: "ALREADY_EXISTS", message: err.details } } as const),
     },
     async () => await callLog(
       { ...logInfo,