Merge pull request #18757 from swordqiu/hotfix/qj-hot-add-mem-for-esxi

fix: hot add mem check for esxi vm

pkg/apis/compute/guest_const.go

@@ -433,6 +433,10 @@ const (
 	VM_METADATA_OS_VERSION          = "os_version"
 	VM_METADATA_CGROUP_CPUSET       = "cgroup_cpuset"
 	VM_METADATA_ENABLE_MEMCLEAN     = "enable_memclean"
+	VM_METADATA_HOTPLUG_CPU_MEM     = "hotplug_cpu_mem"
+	VM_METADATA_HOT_REMOVE_NIC      = "hot_remove_nic"
+	VM_METADATA_START_VMEM_MB       = "start_vmem_mb"
+	VM_METADATA_START_VCPU_COUNT    = "start_vcpu_count"
 )
 
 func Hypervisors2HostTypes(hypervisors []string) []string {

pkg/compute/guestdrivers/base.go

@@ -321,7 +321,7 @@ func (drv *SBaseGuestDriver) IsSupportPublicIp() bool {
 	return false
 }
 
-func (drv *SBaseGuestDriver) NeedStopForChangeSpec(ctx context.Context, guest *models.SGuest, cpuChanged, memChanged bool) bool {
+func (drv *SBaseGuestDriver) NeedStopForChangeSpec(ctx context.Context, guest *models.SGuest, addCpu int, addMemMb int) bool {
 	return false
 }
 

pkg/compute/guestdrivers/esxi.go

@@ -18,6 +18,7 @@ import (
 	"context"
 	"fmt"
 	"sort"
+	"strconv"
 	"time"
 
 	"yunion.io/x/cloudmux/pkg/cloudprovider"
@@ -861,3 +862,23 @@ func (drv *SESXiGuestDriver) RequestUndeployGuestOnHost(ctx context.Context, gue
 	})
 	return nil
 }
+
+func (drv *SESXiGuestDriver) NeedStopForChangeSpec(ctx context.Context, guest *models.SGuest, addCpu int, addMemMb int) bool {
+	// https://kb.vmware.com/s/article/2008405
+	startVmem := guest.VmemSize
+	vmemMbStr := guest.GetMetadata(ctx, api.VM_METADATA_START_VMEM_MB, nil)
+	if len(vmemMbStr) > 0 {
+		vmemMb, _ := strconv.Atoi(vmemMbStr)
+		if vmemMb > 0 {
+			startVmem = int(vmemMb)
+		}
+	}
+	maxAllowVmem := 16 * startVmem
+	if startVmem <= 3*1024 {
+		maxAllowVmem = 3 * 1024
+	}
+	if guest.VmemSize+addMemMb > maxAllowVmem {
+		return true
+	}
+	return false
+}

pkg/compute/guestdrivers/kvm.go

@@ -425,10 +425,8 @@ func (self *SKVMGuestDriver) RequestAssociateEip(ctx context.Context, userCred m
 	return nil
 }
 
-func (self *SKVMGuestDriver) NeedStopForChangeSpec(ctx context.Context, guest *models.SGuest, cpuChanged, memChanged bool) bool {
-	return guest.GetMetadata(ctx, "hotplug_cpu_mem", nil) != "enable" || apis.IsARM(guest.OsArch)
-	// (memChanged && guest.GetMetadata(ctx, "__hugepage", nil) == "native") ||
-	// apis.IsARM(guest.OsArch)
+func (self *SKVMGuestDriver) NeedStopForChangeSpec(ctx context.Context, guest *models.SGuest, addCpu int, addMemMb int) bool {
+	return guest.GetMetadata(ctx, api.VM_METADATA_HOTPLUG_CPU_MEM, nil) != "enable" || apis.IsARM(guest.OsArch) && addMemMb > 0
 }
 
 func (self *SKVMGuestDriver) RequestChangeVmConfig(ctx context.Context, guest *models.SGuest, task taskman.ITask, instanceType string, vcpuCount, cpuSockets, vmemSize int64) error {
@@ -866,7 +864,7 @@ func (self *SKVMGuestDriver) RequestCancelLiveMigrate(ctx context.Context, guest
 }
 
 func (self *SKVMGuestDriver) ValidateDetachNetwork(ctx context.Context, userCred mcclient.TokenCredential, guest *models.SGuest) error {
-	if guest.Status == api.VM_RUNNING && guest.GetMetadata(ctx, "hot_remove_nic", nil) != "enable" {
+	if guest.Status == api.VM_RUNNING && guest.GetMetadata(ctx, api.VM_METADATA_HOT_REMOVE_NIC, nil) != "enable" {
 		return httperrors.NewBadRequestError("Guest %s can't hot remove nic", guest.GetName())
 	}
 	return nil

pkg/compute/models/guest_actions.go

@@ -2762,7 +2762,7 @@ func (self *SGuest) PerformChangeConfig(ctx context.Context, userCred mcclient.T
 		confs.Set("cpu_sockets", jsonutils.NewInt(int64(*input.CpuSockets)))
 	}
 
-	if self.Status == api.VM_RUNNING && (cpuChanged || memChanged || cpuSocketsChanged) && self.GetDriver().NeedStopForChangeSpec(ctx, self, cpuChanged, memChanged) {
+	if self.Status == api.VM_RUNNING && (cpuChanged || memChanged || cpuSocketsChanged) && self.GetDriver().NeedStopForChangeSpec(ctx, self, addCpu, addMem) {
 		return nil, httperrors.NewInvalidStatusError("cannot change CPU/Memory spec in status %s", self.Status)
 	}
 

pkg/compute/models/guestdrivers.go

@@ -186,7 +186,7 @@ type IGuestDriver interface {
 	IsSupportPublicIp() bool
 	ValidateCreateEip(ctx context.Context, userCred mcclient.TokenCredential, input api.ServerCreateEipInput) error
 
-	NeedStopForChangeSpec(ctx context.Context, guest *SGuest, cpuChanged, memChanged bool) bool
+	NeedStopForChangeSpec(ctx context.Context, guest *SGuest, addCpu int, addMemMb int) bool
 
 	OnGuestChangeCpuMemFailed(ctx context.Context, guest *SGuest, data *jsonutils.JSONDict, task taskman.ITask) error
 	IsSupportGuestClone() bool

pkg/compute/tasks/guest_start_task.go

@@ -58,6 +58,10 @@ func (self *GuestStartTask) RequestStart(ctx context.Context, guest *models.SGue
 
 func (task *GuestStartTask) OnStartComplete(ctx context.Context, obj db.IStandaloneModel, data jsonutils.JSONObject) {
 	guest := obj.(*models.SGuest)
+	// save start mem and cpu
+	guest.SetMetadata(ctx, api.VM_METADATA_START_VCPU_COUNT, guest.VcpuCount, task.UserCred)
+	guest.SetMetadata(ctx, api.VM_METADATA_START_VMEM_MB, guest.VmemSize, task.UserCred)
+	// sync Vpc Topology
 	isVpc, err := guest.IsOneCloudVpcNetwork()
 	if err != nil {
 		log.Errorf("IsOneCloudVpcNetwork fail: %s", err)
@@ -68,6 +72,7 @@ func (task *GuestStartTask) OnStartComplete(ctx context.Context, obj db.IStandal
 			log.Errorf("vpcagent.VpcAgent.DoSync fail %s", err)
 		}
 	}
+	// log
 	db.OpsLog.LogEvent(guest, db.ACT_START, guest.GetShortDesc(ctx), task.UserCred)
 	logclient.AddActionLogWithStartable(task, guest, logclient.ACT_VM_START, guest.GetShortDesc(ctx), task.UserCred, true)
 	task.taskComplete(ctx, guest)

pkg/hostman/guestman/qemu-kvm.go

@@ -759,8 +759,8 @@ func (s *SKVMGuestInstance) onImportGuestMonitorConnected(ctx context.Context) {
 		log.Infof("Guest %s qemu version %s", s.Id, version)
 		s.QemuVersion = version
 		meta := jsonutils.NewDict()
-		meta.Set("hotplug_cpu_mem", jsonutils.NewString("disable"))
-		meta.Set("hot_remove_nic", jsonutils.NewString("disable"))
+		meta.Set(api.VM_METADATA_HOTPLUG_CPU_MEM, jsonutils.NewString("disable"))
+		meta.Set(api.VM_METADATA_HOT_REMOVE_NIC, jsonutils.NewString("disable"))
 		meta.Set("__qemu_version", jsonutils.NewString(s.GetQemuVersionStr()))
 		s.SyncMetadata(meta)
 		s.SyncStatus("")
@@ -2602,8 +2602,8 @@ func (s *SKVMGuestInstance) OnResumeSyncMetadataInfo() {
 	meta.Set("__qemu_version", jsonutils.NewString(s.GetQemuVersionStr()))
 	meta.Set("__vnc_port", jsonutils.NewInt(int64(s.GetVncPort())))
 	meta.Set("__enable_cgroup_cpuset", jsonutils.JSONTrue)
-	meta.Set("hotplug_cpu_mem", jsonutils.NewString("enable"))
-	meta.Set("hot_remove_nic", jsonutils.NewString("enable"))
+	meta.Set(api.VM_METADATA_HOTPLUG_CPU_MEM, jsonutils.NewString("enable"))
+	meta.Set(api.VM_METADATA_HOT_REMOVE_NIC, jsonutils.NewString("enable"))
 	if len(s.VncPassword) > 0 {
 		meta.Set("__vnc_password", jsonutils.NewString(s.VncPassword))
 	}