Merge pull request #278 from kthcloud/dev

fix gpu not attaching in cs, fix jobs not using entry and exit func (some activities stuck)
kthcloud · Nov 18, 2023 · eca9b06 · eca9b06
2 parents 201e56b + cbc9e4c
commit eca9b06
Show file tree

Hide file tree

Showing 4 changed files with 62 additions and 15 deletions.
diff --git a/pkg/jobs/runner.go b/pkg/jobs/runner.go
@@ -42,7 +42,7 @@ func (runner *Runner) Run() {
 			}
 		}
 
-		go wrapper(jobDef.JobFunc, runner.Job)
+		go wrapper(jobDef)
 	} else {
 		utils.PrettyPrintError(fmt.Errorf("unknown job type: %s", runner.Job.Type))
 
@@ -54,39 +54,69 @@ func (runner *Runner) Run() {
 	}
 }
 
-func wrapper(fn func(job *jobModel.Job) error, job *jobModel.Job) {
-	err := fn(job)
+func wrapper(def *JobDefinition) {
+	if def.EntryFunc != nil {
+		err := def.EntryFunc(def.Job)
+		if err != nil {
+			utils.PrettyPrintError(fmt.Errorf("error executing job (%s) entry function. details: %w", def.Job.Type, err))
+
+			err = jobModel.New().MarkFailed(def.Job.ID, err.Error())
+			if err != nil {
+				utils.PrettyPrintError(fmt.Errorf("error marking job as failed. details: %w", err))
+				return
+			}
+			return
+		}
+	}
+
+	defer func() {
+		if def.ExitFunc != nil {
+			err := def.ExitFunc(def.Job)
+			if err != nil {
+				utils.PrettyPrintError(fmt.Errorf("error executing job (%s) exit function. details: %w", def.Job.Type, err))
+
+				err = jobModel.New().MarkFailed(def.Job.ID, err.Error())
+				if err != nil {
+					utils.PrettyPrintError(fmt.Errorf("error marking job as failed. details: %w", err))
+					return
+				}
+				return
+			}
+		}
+	}()
+
+	err := def.JobFunc(def.Job)
 
 	if err != nil {
 		if strings.HasPrefix(err.Error(), "failed") {
 			err = errors.Unwrap(err)
-			utils.PrettyPrintError(fmt.Errorf("failed job (%s). details: %w", job.Type, err))
+			utils.PrettyPrintError(fmt.Errorf("failed job (%s). details: %w", def.Job.Type, err))
 
-			err = jobModel.New().MarkFailed(job.ID, err.Error())
+			err = jobModel.New().MarkFailed(def.Job.ID, err.Error())
 			if err != nil {
 				utils.PrettyPrintError(fmt.Errorf("error marking job as failed. details: %w", err))
 				return
 			}
 		} else if strings.HasPrefix(err.Error(), "terminated") {
 			err = errors.Unwrap(err)
-			utils.PrettyPrintError(fmt.Errorf("terminated job (%s). details: %w", job.Type, err))
+			utils.PrettyPrintError(fmt.Errorf("terminated job (%s). details: %w", def.Job.Type, err))
 
-			err = jobModel.New().MarkTerminated(job.ID, err.Error())
+			err = jobModel.New().MarkTerminated(def.Job.ID, err.Error())
 			if err != nil {
 				utils.PrettyPrintError(fmt.Errorf("error marking job as terminated. details: %w", err))
 				return
 			}
 		} else {
-			utils.PrettyPrintError(fmt.Errorf("error executing job (%s). details: %w", job.Type, err))
+			utils.PrettyPrintError(fmt.Errorf("error executing job (%s). details: %w", def.Job.Type, err))
 
-			err = jobModel.New().MarkFailed(job.ID, err.Error())
+			err = jobModel.New().MarkFailed(def.Job.ID, err.Error())
 			if err != nil {
 				utils.PrettyPrintError(fmt.Errorf("error marking job as failed. details: %w", err))
 				return
 			}
 		}
 	} else {
-		err = jobModel.New().MarkCompleted(job.ID)
+		err = jobModel.New().MarkCompleted(def.Job.ID)
 		if err != nil {
 			utils.PrettyPrintError(fmt.Errorf("error marking job as completed. details: %w", err))
 			return

diff --git a/pkg/subsystems/cs/vm.go b/pkg/subsystems/cs/vm.go
@@ -9,6 +9,7 @@ import (
 	"gopkg.in/yaml.v3"
 	"log"
 	"math/rand"
+	"net/url"
 	"strings"
 	"time"
 )
@@ -120,11 +121,11 @@ func (client *Client) UpdateVM(public *models.VmPublic) (*models.VmPublic, error
 		params.SetName(public.Name)
 		params.SetDisplayname(public.Name)
 
-		//if public.ExtraConfig == "" {
-		//	params.SetExtraconfig(url.QueryEscape("none"))
-		//} else {
-		//	params.SetExtraconfig(url.QueryEscape(public.ExtraConfig))
-		//}
+		if public.ExtraConfig == "" {
+			params.SetExtraconfig(url.QueryEscape("none"))
+		} else {
+			params.SetExtraconfig(url.QueryEscape(public.ExtraConfig))
+		}
 
 		_, err = client.CsClient.VirtualMachine.UpdateVirtualMachine(params)
 		if err != nil {

diff --git a/service/resources/cs_generator.go b/service/resources/cs_generator.go
@@ -51,6 +51,8 @@ func (cr *CsGenerator) VMs() []models.VmPublic {
 		if csVM := &cr.v.vm.Subsystems.CS.VM; service.Created(csVM) {
 			csVM.Name = cr.v.vm.Name
 			csVM.ServiceOfferingID = cr.v.vm.Subsystems.CS.ServiceOffering.ID
+			csVM.ExtraConfig = cr.v.vm.Subsystems.CS.VM.ExtraConfig
+
 			res = append(res, *csVM)
 			return res
 		} else {
@@ -59,6 +61,7 @@ func (cr *CsGenerator) VMs() []models.VmPublic {
 				ServiceOfferingID: cr.v.vm.Subsystems.CS.ServiceOffering.ID,
 				TemplateID:        constants.TemplateID,
 				Tags:              createTags(cr.v.vm.Name, cr.v.vm.Name),
+				ExtraConfig:       cr.v.vm.Subsystems.CS.VM.ExtraConfig,
 			})
 			return res
 		}

diff --git a/service/vm_service/cs_service/cs_service.go b/service/vm_service/cs_service/cs_service.go
@@ -5,6 +5,7 @@ import (
 	"fmt"
 	configModels "go-deploy/models/config"
 	vmModel "go-deploy/models/sys/vm"
+	gpuModel "go-deploy/models/sys/vm/gpu"
 	"go-deploy/pkg/config"
 	"go-deploy/pkg/subsystems/cs/commands"
 	csModels "go-deploy/pkg/subsystems/cs/models"
@@ -341,6 +342,18 @@ func Repair(id string) error {
 
 	// only repair if the vm is stopped to prevent downtime for the user
 	if status == "Stopped" {
+		var gpu *gpuModel.GPU
+		if context.VM.GpuID != "" {
+			gpu, err = gpuModel.New().GetByID(context.VM.GpuID)
+			if err != nil {
+				return makeError(err)
+			}
+		}
+
+		if gpu != nil {
+			vm.ExtraConfig = CreateExtraConfig(gpu)
+		}
+
 		err = resources.SsRepairer(
 			context.Client.ReadVM,
 			context.Client.CreateVM,