fix: check resource before resume operation (#2272)

framework/app-service/.olares/config/cluster/deploy/appservice_deploy.yaml

@@ -170,7 +170,7 @@ spec:
       priorityClassName: "system-cluster-critical"
       containers:
       - name: app-service
-        image: beclab/app-service:0.4.63
+        image: beclab/app-service:0.4.65
         imagePullPolicy: IfNotPresent
         securityContext:
           runAsUser: 0

framework/app-service/pkg/apiserver/api/types.go

@@ -93,6 +93,7 @@ type RequirementResp struct {
 	Response
 	Resource string `json:"resource"`
 	Message  string `json:"message"`
+	Reason   string `json:"reason"`
 }
 
 // AppSource describe the source of an application, recommend,model,agent

framework/app-service/pkg/apiserver/handler_installer_install.go

@@ -343,23 +343,25 @@ func (h *installHandlerHelper) validate(isAdmin bool, installedApps []*v1alpha1.
 	}
 
 	//resourceType, err := CheckAppRequirement(h.h.kubeConfig, h.token, h.appConfig)
-	resourceType, err := apputils.CheckAppRequirement(h.token, h.appConfig)
+	resourceType, resourceConditionType, err := apputils.CheckAppRequirement(h.token, h.appConfig, v1alpha1.InstallOp)
 	if err != nil {
 		klog.Errorf("Failed to check app requirement err=%v", err)
 		h.resp.WriteHeaderAndEntity(http.StatusBadRequest, api.RequirementResp{
 			Response: api.Response{Code: 400},
-			Resource: resourceType,
+			Resource: resourceType.String(),
 			Message:  err.Error(),
+			Reason:   resourceConditionType.String(),
 		})
 		return
 	}
 
-	resourceType, err = apputils.CheckUserResRequirement(h.req.Request.Context(), h.appConfig, h.owner)
+	resourceType, resourceConditionType, err = apputils.CheckUserResRequirement(h.req.Request.Context(), h.appConfig, v1alpha1.InstallOp)
 	if err != nil {
 		h.resp.WriteHeaderAndEntity(http.StatusBadRequest, api.RequirementResp{
 			Response: api.Response{Code: 400},
-			Resource: resourceType,
+			Resource: resourceType.String(),
 			Message:  err.Error(),
+			Reason:   resourceConditionType.String(),
 		})
 		return
 	}

framework/app-service/pkg/apiserver/handler_suspend.go

@@ -1,13 +1,17 @@
 package apiserver
 
 import (
+	"encoding/json"
 	"errors"
 	"fmt"
+	"k8s.io/klog/v2"
+	"net/http"
 	"strconv"
 	"time"
 
 	"bytetrade.io/web3os/app-service/api/app.bytetrade.io/v1alpha1"
 	"bytetrade.io/web3os/app-service/pkg/apiserver/api"
+	"bytetrade.io/web3os/app-service/pkg/appcfg"
 	"bytetrade.io/web3os/app-service/pkg/appstate"
 	"bytetrade.io/web3os/app-service/pkg/constants"
 	"bytetrade.io/web3os/app-service/pkg/kubesphere"
@@ -99,6 +103,12 @@ func (h *Handler) suspend(req *restful.Request, resp *restful.Response) {
 func (h *Handler) resume(req *restful.Request, resp *restful.Response) {
 	app := req.PathParameter(ParamAppName)
 	owner := req.Attribute(constants.UserContextAttribute).(string)
+	token, err := h.GetUserServiceAccountToken(req.Request.Context(), owner)
+	if err != nil {
+		klog.Error("Failed to get user service account token: ", err)
+		api.HandleError(resp, req, err)
+		return
+	}
 
 	name, err := apputils.FmtAppMgrName(app, owner, "")
 	if err != nil {
@@ -116,6 +126,36 @@ func (h *Handler) resume(req *restful.Request, resp *restful.Response) {
 		api.HandleBadRequest(resp, req, fmt.Errorf("%s operation is not allowed for %s state", v1alpha1.ResumeOp, am.Status.State))
 		return
 	}
+	var appCfg *appcfg.ApplicationConfig
+	err = json.Unmarshal([]byte(am.Spec.Config), &appCfg)
+	if err != nil {
+		klog.Errorf("unmarshal to appConfig failed %v", err)
+		api.HandleError(resp, req, err)
+		return
+	}
+
+	resourceType, resourceConditionType, err := apputils.CheckAppRequirement(token, appCfg, v1alpha1.ResumeOp)
+	if err != nil {
+		klog.Errorf("Failed to check app requirement err=%v", err)
+		resp.WriteHeaderAndEntity(http.StatusBadRequest, api.RequirementResp{
+			Response: api.Response{Code: 400},
+			Resource: resourceType.String(),
+			Message:  err.Error(),
+			Reason:   resourceConditionType.String(),
+		})
+		return
+	}
+
+	resourceType, resourceConditionType, err = apputils.CheckUserResRequirement(req.Request.Context(), appCfg, v1alpha1.ResumeOp)
+	if err != nil {
+		resp.WriteHeaderAndEntity(http.StatusBadRequest, api.RequirementResp{
+			Response: api.Response{Code: 400},
+			Resource: resourceType.String(),
+			Message:  err.Error(),
+			Reason:   resourceConditionType.String(),
+		})
+		return
+	}
 
 	am.Spec.OpType = v1alpha1.ResumeOp
 	// if current user is admin, also resume server side

framework/app-service/pkg/apiserver/handler_webhook.go

@@ -1382,12 +1382,12 @@ func (h *Handler) installOpValidate(ctx context.Context, appConfig *appcfg.Appli
 	if err != nil {
 		return err
 	}
-	_, err = apputils.CheckAppRequirement("", appConfig)
+	_, _, err = apputils.CheckAppRequirement("", appConfig, v1alpha1.InstallOp)
 	if err != nil {
 		return err
 	}
 
-	_, err = apputils.CheckUserResRequirement(ctx, appConfig, appConfig.OwnerName)
+	_, _, err = apputils.CheckUserResRequirement(ctx, appConfig, v1alpha1.InstallOp)
 	if err != nil {
 		return err
 	}

framework/app-service/pkg/constants/constants.go

@@ -130,9 +130,6 @@ const (
 
 var (
 	empty = sets.Empty{}
-	// States represents the state for whole application lifecycle.
-	States = sets.String{"pending": empty, "downloading": empty, "installing": empty, "initializing": empty, "running": empty,
-		"uninstalling": empty, "upgrading": empty, "suspend": empty, "resuming": empty}
 	// Sources represents the source of the application.
 	Sources = sets.String{"market": empty, "custom": empty, "devbox": empty, "system": empty, "unknown": empty}
 	// ResourceTypes represents the type of application system supported.
@@ -152,6 +149,47 @@ var (
 	OLARES_APP_NAME = "olares-app"
 )
 
+type ResourceConditionType string
+
+const (
+	DiskPressure             ResourceConditionType = "DiskPressure"
+	SystemCPUPressure        ResourceConditionType = "SystemCPUPressure"
+	SystemMemoryPressure     ResourceConditionType = "SystemMemoryPressure"
+	SystemGPUNotAvailable    ResourceConditionType = "SystemGPUNotAvailable"
+	SystemGPUPressure        ResourceConditionType = "SystemGPUPressure"
+	K8sRequestCPUPressure    ResourceConditionType = "K8sReqeustCPUPressure"
+	K8sRequestMemoryPressure ResourceConditionType = "K8sRequestMemoryPressure"
+	UserCPUPressure          ResourceConditionType = "UserCPUPressure"
+	UserMemoryPressure       ResourceConditionType = "UserMemoryPressure"
+
+	DiskPressureMessage             string = "Insufficient disk space. Unable to %s the application. Please stop other running applications to free up storage."
+	SystemCPUPressureMessage        string = "Insufficient system CPU. Unable to %s the application. Please stop other running applications to free up resources."
+	SystemMemoryPressureMessage     string = "Insufficient system memory. Unable to %s the application. Please stop other running applications to free up memory."
+	SystemGPUNotAvailableMessage    string = "No available GPU found. Unable to %s the application."
+	SystemGPUPressureMessage        string = "Available GPU is insufficient to %s this application. The requested GPU memory cannot exceed the maximum GPU memory of the node."
+	K8sRequestCPUPressureMessage    string = "Available CPU is insufficient to %s this application. Please stop other applications to free up resources."
+	K8sRequestMemoryPressureMessage string = "Available memory is insufficient to %s this application. Please stop other applications to free up resources."
+	UserCPUPressureMessage          string = "Insufficient user CPU. Unable to %s the application. Please stop other running applications to free up resources."
+	UserMemoryPressureMessage       string = "Insufficient user memory. Unable to %s the application. Please stop other running applications to free up memory."
+)
+
+func (rct ResourceConditionType) String() string {
+	return string(rct)
+}
+
+type ResourceType string
+
+const (
+	Disk   ResourceType = "disk"
+	CPU    ResourceType = "cpu"
+	Memory ResourceType = "memory"
+	GPU    ResourceType = "gpu"
+)
+
+func (rt ResourceType) String() string {
+	return string(rt)
+}
+
 func init() {
 	flag.StringVar(&APIServerListenAddress, "listen", ":6755",
 		"app-service listening address")

framework/app-service/pkg/images/client.go

@@ -199,7 +199,7 @@ func (imc *ImageManagerClient) PollDownloadProgress(ctx context.Context, am *app
 
 			}
 			err = imc.updateProgress(ctx, am, &lastProgress, ret*100, am.Spec.OpType == appv1alpha1.UpgradeOp)
-			if err == nil && im.Status.State == "completed" {
+			if err == nil {
 				return nil
 			}
 

framework/app-service/pkg/images/jobs.go

@@ -242,6 +242,11 @@ func updateProgress(statuses []StatusInfo, ongoing *jobs, seen map[string]int64,
 	}
 	for _, status := range statuses {
 		klog.Infof("status: %s,ref: %v, offset: %v, Total: %v", status.Status, status.Ref, status.Offset, status.Total)
+		if !isLayerType(status.Ref) {
+			statusesLen--
+			continue
+		}
+
 		if status.Status == "exists" {
 			key := strings.Split(status.Ref, "-")[1]
 			offset += seen[key]
@@ -249,10 +254,6 @@ func updateProgress(statuses []StatusInfo, ongoing *jobs, seen map[string]int64,
 			continue
 		}
 
-		if !isLayerType(status.Ref) {
-			statusesLen--
-			continue
-		}
 		if status.Status == "done" {
 			offset += status.Total
 			doneLayer++
@@ -261,7 +262,7 @@ func updateProgress(statuses []StatusInfo, ongoing *jobs, seen map[string]int64,
 
 		offset += status.Offset
 	}
-	if doneLayer == statusesLen && doneLayer != 0 {
+	if doneLayer == statusesLen && statusesLen > 0 {
 		offset = imageSize
 	}
 	if imageSize != 0 {

framework/app-service/pkg/utils/app/validate.go

@@ -207,39 +207,41 @@ func CheckUserRole(appConfig *appcfg.ApplicationConfig, owner string) error {
 }
 
 // CheckAppRequirement check if the cluster has enough resources for application install/upgrade.
-func CheckAppRequirement(token string, appConfig *appcfg.ApplicationConfig) (string, error) {
+func CheckAppRequirement(token string, appConfig *appcfg.ApplicationConfig, op v1alpha1.OpType) (constants.ResourceType, constants.ResourceConditionType, error) {
 	metrics, _, err := GetClusterResource(token)
 	if err != nil {
-		return "", err
+		return "", "", err
 	}
 
+	klog.Infof("start to %s app %s", op, appConfig.AppName)
 	klog.Infof("Current resource=%s", utils.PrettyJSON(metrics))
 	klog.Infof("App required resource=%s", utils.PrettyJSON(appConfig.Requirement))
 
 	if appConfig.Requirement.Disk != nil &&
 		appConfig.Requirement.Disk.CmpInt64(int64(metrics.Disk.Total*0.9-metrics.Disk.Usage)) > 0 ||
 		int64(metrics.Disk.Total*0.9-metrics.Disk.Usage) < 5*1024*1024*1024 {
-		return "disk", errors.New("The app's DISK requirement cannot be satisfied")
+		return constants.Disk, constants.DiskPressure, fmt.Errorf(constants.DiskPressureMessage, op)
 	}
 
 	if appConfig.Requirement.Memory != nil &&
 		appConfig.Requirement.Memory.CmpInt64(int64(metrics.Memory.Total*0.9-metrics.Memory.Usage)) > 0 {
-		return "memory", errors.New("The app's MEMORY requirement cannot be satisfied")
+		return constants.Memory, constants.SystemMemoryPressure, fmt.Errorf(constants.SystemMemoryPressureMessage, op)
 	}
 	if appConfig.Requirement.CPU != nil {
 		availableCPU, _ := resource.ParseQuantity(strconv.FormatFloat(metrics.CPU.Total*0.9-metrics.CPU.Usage, 'f', -1, 64))
 		if appConfig.Requirement.CPU.Cmp(availableCPU) > 0 {
-			return "cpu", errors.New("The app's CPU requirement cannot be satisfied")
+			return constants.CPU, constants.SystemCPUPressure, fmt.Errorf(constants.SystemCPUPressureMessage, op)
 		}
 	}
 	if appConfig.Requirement.GPU != nil {
 		if !appConfig.Requirement.GPU.IsZero() && metrics.GPU.Total <= 0 {
-			return "gpu", errors.New("The app's GPU requirement cannot be satisfied")
+			return constants.GPU, constants.SystemGPUNotAvailable, fmt.Errorf(constants.SystemGPUNotAvailableMessage, op)
+
 		}
 		nodes, err := utils.GetNodeInfo(context.TODO())
 		if err != nil {
 			klog.Errorf("failed to get node info %v", err)
-			return "", err
+			return "", "", err
 		}
 		klog.Infof("nodes info: %#v", nodes)
 		var maxNodeGPUMem int64
@@ -254,13 +256,13 @@ func CheckAppRequirement(token string, appConfig *appcfg.ApplicationConfig) (str
 		}
 
 		if appConfig.Requirement.GPU.CmpInt64(maxNodeGPUMem) > 0 {
-			return "gpu", errors.New("The app's GPU requirement cannot found satisfied node")
+			return constants.GPU, constants.SystemGPUPressure, fmt.Errorf(constants.SystemGPUPressureMessage, op)
 		}
 	}
 
 	allocatedResources, err := getRequestResources()
 	if err != nil {
-		return "", err
+		return "", "", err
 	}
 	if len(allocatedResources) == 1 {
 		sufficientCPU, sufficientMemory := false, false
@@ -283,14 +285,14 @@ func CheckAppRequirement(token string, appConfig *appcfg.ApplicationConfig) (str
 			}
 		}
 		if !sufficientCPU {
-			return "cpu", errors.New("The app's CPU requirement specified in the kubernetes requests cannot be satisfied")
+			return constants.CPU, constants.K8sRequestCPUPressure, fmt.Errorf(constants.K8sRequestCPUPressureMessage, op)
 		}
 		if !sufficientMemory {
-			return "memory", errors.New("The app's MEMORY requirement specified in the kubernetes requests cannot be satisfied")
+			return constants.Memory, constants.K8sRequestMemoryPressure, fmt.Errorf(constants.K8sRequestMemoryPressureMessage, op)
 		}
 	}
 
-	return "", nil
+	return "", "", nil
 }
 
 func getRequestResources() (map[string]resources, error) {
@@ -450,22 +452,22 @@ func getValue(m *kubesphere.Metric) float64 {
 }
 
 // CheckUserResRequirement check if the user has enough resources for application install/upgrade.
-func CheckUserResRequirement(ctx context.Context, appConfig *appcfg.ApplicationConfig, username string) (string, error) {
-	metrics, err := prometheus.GetCurUserResource(ctx, username)
+func CheckUserResRequirement(ctx context.Context, appConfig *appcfg.ApplicationConfig, op v1alpha1.OpType) (constants.ResourceType, constants.ResourceConditionType, error) {
+	metrics, err := prometheus.GetCurUserResource(ctx, appConfig.OwnerName)
 	if err != nil {
-		return "", err
+		return "", "", err
 	}
 	switch {
 	case appConfig.Requirement.Memory != nil && metrics.Memory.Total != 0 &&
 		appConfig.Requirement.Memory.CmpInt64(int64(metrics.Memory.Total*0.9-metrics.Memory.Usage)) > 0:
-		return "memory", errors.New("The user's app MEMORY requirement cannot be satisfied")
+		return constants.Memory, constants.UserMemoryPressure, fmt.Errorf(constants.UserMemoryPressureMessage, op)
 	case appConfig.Requirement.CPU != nil && metrics.CPU.Total != 0:
 		availableCPU, _ := resource.ParseQuantity(strconv.FormatFloat(metrics.CPU.Total*0.9-metrics.CPU.Usage, 'f', -1, 64))
 		if appConfig.Requirement.CPU.Cmp(availableCPU) > 0 {
-			return "cpu", errors.New("The user's app CPU requirement cannot be satisfied")
+			return constants.CPU, constants.UserCPUPressure, fmt.Errorf(constants.UserCPUPressureMessage, op)
 		}
 	}
-	return "", nil
+	return "", "", nil
 }
 
 func CheckMiddlewareRequirement(ctx context.Context, ctrlClient client.Client, middleware *tapr.Middleware) (bool, error) {