Fix standby snapshot validation balloon handling

Author: cursoragent

lib/instances/snapshot.go

@@ -272,7 +272,7 @@ func (m *manager) restoreSnapshot(ctx context.Context, id string, snapshotID str
 iferr!=nil {
 returnnil, err
 	}
-starter, targetHypervisorVersion, err:=m.prepareSnapshotTarget(ctx, rec.StoredMetadata, targetHypervisor)
+starter, targetHypervisorVersion, err:=m.prepareSnapshotTarget(ctx, rec.Snapshot.Kind, rec.StoredMetadata, targetHypervisor)
 iferr!=nil {
 returnnil, err
 	}
@@ -387,7 +387,7 @@ func (m *manager) forkSnapshot(ctx context.Context, snapshotID string, req ForkS
 iferr!=nil {
 returnnil, err
 	}
-starter, targetHypervisorVersion, err:=m.prepareSnapshotTarget(ctx, rec.StoredMetadata, targetHypervisor)
+starter, targetHypervisorVersion, err:=m.prepareSnapshotTarget(ctx, rec.Snapshot.Kind, rec.StoredMetadata, targetHypervisor)
 iferr!=nil {
 returnnil, err
 	}
@@ -519,12 +519,12 @@ func (m *manager) replaceInstanceWithSnapshotPayload(snapshotID, instanceID stri
 returnnil
 }
 
-func (m*manager) prepareSnapshotTarget(ctx context.Context, sourceStoredMetadata, target hypervisor.Type) (hypervisor.VMStarter, string, error) {
+func (m*manager) prepareSnapshotTarget(ctx context.Context, snapshotKindSnapshotKind, sourceStoredMetadata, target hypervisor.Type) (hypervisor.VMStarter, string, error) {
 starter, err:=m.getVMStarter(target)
 iferr!=nil {
 returnnil, "", fmt.Errorf("get vm starter: %w", err)
 	}
-iferr:=m.validateStoredVMConfig(starter, source, target); err!=nil {
+iferr:=m.validateStoredVMConfig(starter, snapshotKind, source, target); err!=nil {
 returnnil, "", err
 	}
 

lib/instances/vm_config_validation.go

@@ -34,19 +34,25 @@ func (m *manager) validateCreateVMConfig(starter hypervisor.VMStarter, req Creat
 	))
 }
 
-// validateStoredVMConfig applies the same backend validation before a stopped
-// snapshot is restored or forked onto a target hypervisor.
-func (m*manager) validateStoredVMConfig(starter hypervisor.VMStarter, metaStoredMetadata, hvType hypervisor.Type) error {
+// validateStoredVMConfig applies backend validation before a snapshot is
+// restored or forked onto a target hypervisor.
+func (m*manager) validateStoredVMConfig(starter hypervisor.VMStarter, snapshotKindSnapshotKind, metaStoredMetadata, hvType hypervisor.Type) error {
 pciDeviceCount:=len(meta.Devices)
 ifmeta.GPUMdevUUID!=""||meta.GPUProfile!="" {
 pciDeviceCount++
 	}
-returnvalidatePlannedVMConfig(starter, hvType,m.plannedVMConfig(
+config:=m.plannedVMConfig(
 meta.HotplugSize,
 meta.Volumes,
 meta.NetworkEnabled,
 pciDeviceCount,
-	))
+	)
+ifsnapshotKind==SnapshotKindStandby {
+// Standby restore/fork reuses the frozen snapshot device model from
+// qemu-config.json, so live host balloon policy must not affect preflight.
+config.GuestMemory= hypervisor.GuestMemoryConfig{}
+	}
+returnvalidatePlannedVMConfig(starter, hvType, config)
 }
 
 func (m*manager) plannedVMConfig(

lib/instances/vm_config_validation_test.go

@@ -1,8 +1,13 @@
 package instances
 
 import (
+"context"
+"fmt"
 "testing"
 
+"github.com/kernel/hypeman/lib/guestmemory"
+"github.com/kernel/hypeman/lib/hypervisor"
+"github.com/kernel/hypeman/lib/paths"
 "github.com/stretchr/testify/assert"
 "github.com/stretchr/testify/require"
 )
@@ -32,3 +37,74 @@ func TestPlannedVMConfigWithoutOptionalDevices(t *testing.T) {
 assert.Empty(t, config.Networks)
 assert.Empty(t, config.PCIDevices)
 }
+
+funcTestPrepareSnapshotTargetStandbyValidationIgnoresLiveBalloonPolicy(t*testing.T) {
+t.Parallel()
+
+hvType:=hypervisor.Type("snapshot-validation-test")
+starter:=&snapshotValidationStarter{}
+m:=&manager{
+vmStarters: map[hypervisor.Type]hypervisor.VMStarter{
+hvType: starter,
+ },
+guestMemoryPolicy: guestmemory.Policy{
+Enabled: true,
+ReclaimEnabled: true,
+ },
+	}
+source:=StoredMetadata{
+HypervisorType: hvType,
+HypervisorVersion: "test",
+Volumes: []VolumeAttachment{
+ {VolumeID: "v1"},
+ {VolumeID: "v2"},
+ {VolumeID: "v3"},
+ {VolumeID: "v4"},
+ },
+	}
+
+_, _, err:=m.prepareSnapshotTarget(context.Background(), SnapshotKindStopped, source, hvType)
+require.ErrorIs(t, err, ErrInvalidRequest)
+assert.True(t, starter.lastConfig.GuestMemory.EnableBalloon, "stopped validation should use live balloon policy")
+
+_, _, err=m.prepareSnapshotTarget(context.Background(), SnapshotKindStandby, source, hvType)
+require.NoError(t, err)
+assert.False(t, starter.lastConfig.GuestMemory.EnableBalloon, "standby validation should not inject live balloon policy")
+}
+
+typesnapshotValidationStarterstruct {
+lastConfig hypervisor.VMConfig
+}
+
+func (s*snapshotValidationStarter) ValidateConfig(config hypervisor.VMConfig) error {
+s.lastConfig=config
+devices:=len(config.Disks) +len(config.Networks)
+ifconfig.VsockCID>0 {
+devices++
+	}
+ifconfig.GuestMemory.EnableBalloon {
+devices++
+	}
+ifdevices>8 {
+returnfmt.Errorf("too many virtio-mmio devices: %d", devices)
+	}
+returnnil
+}
+
+func (*snapshotValidationStarter) SocketName() string { return"snapshot-validation.sock" }
+func (*snapshotValidationStarter) GetBinaryPath(*paths.Paths, string) (string, error) {
+return"", nil
+}
+func (*snapshotValidationStarter) GetVersion(*paths.Paths) (string, error) { return"test", nil }
+func (*snapshotValidationStarter) ResolveVersion(*paths.Paths, string) (string, error) {
+return"test", nil
+}
+func (*snapshotValidationStarter) StartVM(context.Context, *paths.Paths, string, string, hypervisor.VMConfig) (int, hypervisor.Hypervisor, error) {
+return0, nil, nil
+}
+func (*snapshotValidationStarter) RestoreVM(context.Context, *paths.Paths, string, string, string, hypervisor.RestoreOptions) (int, hypervisor.Hypervisor, error) {
+return0, nil, nil
+}
+func (*snapshotValidationStarter) PrepareFork(context.Context, hypervisor.ForkPrepareRequest) (hypervisor.ForkPrepareResult, error) {
+return hypervisor.ForkPrepareResult{}, nil
+}