查询job信息

desc/hpc/pcm-hpc.api

@@ -116,6 +116,17 @@ type cancelJobReq {
     JobId string `form:"jobId"`
 }
 
+type jobInfoReq {
+    ClusterId int64 `form:"clusterId"`
+    JobId string `form:"jobId"`
+}
+
+type jobInfoResp {
+    JobId string `form:"jobId"`
+    JobState string `json:"jobState"`
+    CurrentWorkingDirectory string `json:"currentWorkingDirectory"`
+}
+
 type QueueAssetsResp {
     QueueAssets []QueueAsset `json:"queueAsset"`
 }

desc/pcm.api

@@ -209,6 +209,10 @@ service pcm {
 	@doc "删除超算任务"
 	@handler cancelJobHandler
 	delete /hpc/cancelJob (cancelJobReq)
+
+	@doc "查看job状态"
+	@handler jobInfoHandler
+	get /hpc/jobInfo (jobInfoReq) returns (jobInfoResp)
 }
 
 //cloud二级接口

go.mod

@@ -19,7 +19,7 @@ require (
 	github.com/robfig/cron/v3 v3.0.1
 	github.com/zeromicro/go-zero v1.7.3
 	gitlink.org.cn/JointCloud/pcm-ac v0.0.0-20240920093406-601f283f0185
-	gitlink.org.cn/JointCloud/pcm-hpc v0.0.0-20241026041404-af824802cfc8
+	gitlink.org.cn/JointCloud/pcm-hpc v0.0.0-20241104040331-d3a6eb951631
 	gitlink.org.cn/JointCloud/pcm-modelarts v0.0.0-20240918011543-482dcd609877
 	gitlink.org.cn/JointCloud/pcm-octopus v0.0.0-20240817071412-44397870b110
 	gitlink.org.cn/JointCloud/pcm-openstack v0.0.0-20240403033338-e7edabad4203

go.sum

@@ -467,8 +467,8 @@ github.com/zeromicro/go-zero v1.7.3 h1:yDUQF2DXDhUHc77/NZF6mzsoRPMBfldjPmG2O/ZSz
 github.com/zeromicro/go-zero v1.7.3/go.mod h1:9JIW3gHBGuc9LzvjZnNwINIq9QdiKu3AigajLtkJamQ=
 gitlink.org.cn/JointCloud/pcm-ac v0.0.0-20240920093406-601f283f0185 h1:B+YBB5xHlIAS6ILuaCGQwbOpr/L6LOHAlj9PeFUCetM=
 gitlink.org.cn/JointCloud/pcm-ac v0.0.0-20240920093406-601f283f0185/go.mod h1:3eECiw9O2bIFkkePlloKyLNXiqBAhOxNrDoGaaGseGY=
-gitlink.org.cn/JointCloud/pcm-hpc v0.0.0-20241026041404-af824802cfc8 h1:74Sgm3izTWGiENLQQKf/DUCHUds9vU4OigXvYi4d9Pc=
+gitlink.org.cn/JointCloud/pcm-hpc v0.0.0-20241104040331-d3a6eb951631 h1:udsOCXqZslipOlDEaxaVaG8gRZzieiRhTfSSnJYSc+E=
-gitlink.org.cn/JointCloud/pcm-hpc v0.0.0-20241026041404-af824802cfc8/go.mod h1:0tMb2cfE73vdFC3AZmPdfH7NwQYhpwsjdFyh2ZdOfY0=
+gitlink.org.cn/JointCloud/pcm-hpc v0.0.0-20241104040331-d3a6eb951631/go.mod h1:2/lG00ZhmS8wURzNSCTLexeGDvqYmZgHbCRnNCSqZaY=
 gitlink.org.cn/JointCloud/pcm-modelarts v0.0.0-20240918011543-482dcd609877 h1:a+1FpxqLPRojlAkJlAeRhKRbxajymXYgrM+s9bfQx0E=
 gitlink.org.cn/JointCloud/pcm-modelarts v0.0.0-20240918011543-482dcd609877/go.mod h1:/eOmBFZKWGoabG3sRVkVvIbLwsd2631k4jkUBR6x1AA=
 gitlink.org.cn/JointCloud/pcm-octopus v0.0.0-20240817071412-44397870b110 h1:GaXwr5sgDh0raHjUf9IewTvnRvajYea7zbLsaerYyXo=

internal/handler/hpc/jobinfohandler.go

@@ -0,0 +1,25 @@
+package hpc
+
+import (
+	"gitlink.org.cn/JointCloud/pcm-coordinator/pkg/repository/result"
+	"net/http"
+
+	"github.com/zeromicro/go-zero/rest/httpx"
+	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/logic/hpc"
+	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/svc"
+	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/types"
+)
+
+func JobInfoHandler(svcCtx *svc.ServiceContext) http.HandlerFunc {
+	return func(w http.ResponseWriter, r *http.Request) {
+		var req types.JobInfoReq
+		if err := httpx.Parse(r, &req); err != nil {
+			httpx.ErrorCtx(r.Context(), w, err)
+			return
+		}
+
+		l := hpc.NewJobInfoLogic(r.Context(), svcCtx)
+		resp, err := l.JobInfo(&req)
+		result.HttpResult(r, w, resp, err)
+	}
+}

internal/logic/hpc/jobinfologic.go

@@ -0,0 +1,56 @@
+package hpc
+
+import (
+	"context"
+	"github.com/pkg/errors"
+	"gitlink.org.cn/JointCloud/pcm-hpc/slurm"
+
+	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/svc"
+	"gitlink.org.cn/JointCloud/pcm-coordinator/internal/types"
+
+	"github.com/zeromicro/go-zero/core/logx"
+)
+
+type JobInfoLogic struct {
+	logx.Logger
+	ctx    context.Context
+	svcCtx *svc.ServiceContext
+}
+
+func NewJobInfoLogic(ctx context.Context, svcCtx *svc.ServiceContext) *JobInfoLogic {
+	return &JobInfoLogic{
+		Logger: logx.WithContext(ctx),
+		ctx:    ctx,
+		svcCtx: svcCtx,
+	}
+}
+
+func (l *JobInfoLogic) JobInfo(req *types.JobInfoReq) (resp *types.JobInfoResp, err error) {
+	resp = &types.JobInfoResp{}
+	var clusterInfo *types.ClusterInfo
+	tx := l.svcCtx.DbEngin.Raw("select * from t_cluster where id = ?", req.ClusterId).Scan(&clusterInfo)
+	if tx.Error != nil {
+		return nil, tx.Error
+	}
+	client, err := slurm.NewClient(slurm.ClientOptions{
+		URL:           clusterInfo.Server,
+		ClientVersion: clusterInfo.Version,
+		RestUserName:  clusterInfo.Username,
+		Token:         clusterInfo.Token})
+	if err != nil {
+		return nil, err
+	}
+	job, err := client.Job(slurm.JobOptions{})
+	if err != nil {
+		return nil, err
+	}
+	jobResp, _ := job.GetJob(slurm.GetJobReq{JobId: req.JobId})
+
+	if len(jobResp.Errors) != 0 {
+		return nil, errors.Errorf(jobResp.Errors[0].Description)
+	}
+	resp.JobId = jobResp.Jobs[0].JobId
+	resp.JobState = jobResp.Jobs[0].JobState
+	resp.CurrentWorkingDirectory = jobResp.Jobs[0].CurrentWorkingDirectory
+	return resp, nil
+}