perf: speed up firecracker standby with diff snapshot reuse

lib/hypervisor/cloudhypervisor/cloudhypervisor.go

@@ -32,13 +32,19 @@ var _ hypervisor.Hypervisor = (*CloudHypervisor)(nil)
 
 // Capabilities returns the features supported by Cloud Hypervisor.
 func (c *CloudHypervisor) Capabilities() hypervisor.Capabilities {
+	return capabilities()
+}
+
+func capabilities() hypervisor.Capabilities {
 	return hypervisor.Capabilities{
-		SupportsSnapshot:       true,
-		SupportsHotplugMemory:  true,
-		SupportsPause:          true,
-		SupportsVsock:          true,
-		SupportsGPUPassthrough: true,
-		SupportsDiskIOLimit:    true,
+		SupportsSnapshot:            true,
+		SupportsHotplugMemory:       true,
+		SupportsPause:               true,
+		SupportsVsock:               true,
+		SupportsGPUPassthrough:      true,
+		SupportsDiskIOLimit:         true,
+		SupportsGracefulVMMShutdown: true,
+		SupportsSnapshotBaseReuse:   false,
 	}
 }
 

lib/hypervisor/cloudhypervisor/process.go

@@ -15,6 +15,7 @@ import (
 
 func init() {
 	hypervisor.RegisterSocketName(hypervisor.TypeCloudHypervisor, "ch.sock")
+	hypervisor.RegisterCapabilities(hypervisor.TypeCloudHypervisor, capabilities())
 	hypervisor.RegisterClientFactory(hypervisor.TypeCloudHypervisor, func(socketPath string) (hypervisor.Hypervisor, error) {
 		return New(socketPath)
 	})

lib/hypervisor/firecracker/config.go

@@ -200,10 +200,15 @@ func toRateLimiter(limit int64, burst int64) *rateLimiter {
 }
 
 func toSnapshotCreateParams(snapshotDir string) snapshotCreateParams {
+	snapshotType := "Full"
+	if _, err := os.Stat(snapshotMemoryPath(snapshotDir)); err == nil {
+		snapshotType = "Diff"
+	}
+
 	return snapshotCreateParams{
 		MemFilePath:  snapshotMemoryPath(snapshotDir),
 		SnapshotPath: snapshotStatePath(snapshotDir),
-		SnapshotType: "Full",
+		SnapshotType: snapshotType,
 	}
 }
 

lib/hypervisor/firecracker/config_test.go

@@ -1,6 +1,8 @@
 package firecracker
 
 import (
+	"os"
+	"path/filepath"
 	"testing"
 
 	"github.com/kernel/hypeman/lib/hypervisor"
@@ -59,10 +61,24 @@ func TestToNetworkInterfaces(t *testing.T) {
 }
 
 func TestSnapshotParamPaths(t *testing.T) {
-	create := toSnapshotCreateParams("/tmp/snapshot-latest")
-	assert.Equal(t, "/tmp/snapshot-latest/state", create.SnapshotPath)
-	assert.Equal(t, "/tmp/snapshot-latest/memory", create.MemFilePath)
-	assert.Equal(t, "Full", create.SnapshotType)
+	t.Run("uses full snapshots when no retained base exists", func(t *testing.T) {
+		snapshotDir := filepath.Join(t.TempDir(), "snapshot-latest")
+		create := toSnapshotCreateParams(snapshotDir)
+		assert.Equal(t, filepath.Join(snapshotDir, "state"), create.SnapshotPath)
+		assert.Equal(t, filepath.Join(snapshotDir, "memory"), create.MemFilePath)
+		assert.Equal(t, "Full", create.SnapshotType)
+	})
+
+	t.Run("uses diff snapshots when retained base memory exists", func(t *testing.T) {
+		snapshotDir := filepath.Join(t.TempDir(), "snapshot-latest")
+		require.NoError(t, os.MkdirAll(snapshotDir, 0755))
+		require.NoError(t, os.WriteFile(filepath.Join(snapshotDir, "memory"), []byte("base"), 0644))
+
+		create := toSnapshotCreateParams(snapshotDir)
+		assert.Equal(t, filepath.Join(snapshotDir, "state"), create.SnapshotPath)
+		assert.Equal(t, filepath.Join(snapshotDir, "memory"), create.MemFilePath)
+		assert.Equal(t, "Diff", create.SnapshotType)
+	})
 
 	load := toSnapshotLoadParams("/tmp/snapshot-latest", []networkOverride{
 		{IfaceID: "eth0", HostDevName: "hype-abc123"},

lib/hypervisor/firecracker/firecracker.go

@@ -47,13 +47,19 @@ func New(socketPath string) (*Firecracker, error) {
 var _ hypervisor.Hypervisor = (*Firecracker)(nil)
 
 func (f *Firecracker) Capabilities() hypervisor.Capabilities {
+	return capabilities()
+}
+
+func capabilities() hypervisor.Capabilities {
 	return hypervisor.Capabilities{
-		SupportsSnapshot:       true,
-		SupportsHotplugMemory:  false,
-		SupportsPause:          true,
-		SupportsVsock:          true,
-		SupportsGPUPassthrough: false,
-		SupportsDiskIOLimit:    true,
+		SupportsSnapshot:            true,
+		SupportsHotplugMemory:       false,
+		SupportsPause:               true,
+		SupportsVsock:               true,
+		SupportsGPUPassthrough:      false,
+		SupportsDiskIOLimit:         true,
+		SupportsGracefulVMMShutdown: false,
+		SupportsSnapshotBaseReuse:   true,
 	}
 }
 

lib/hypervisor/firecracker/process.go

@@ -25,6 +25,7 @@ const (
 
 func init() {
 	hypervisor.RegisterSocketName(hypervisor.TypeFirecracker, "fc.sock")
+	hypervisor.RegisterCapabilities(hypervisor.TypeFirecracker, capabilities())
 	hypervisor.RegisterClientFactory(hypervisor.TypeFirecracker, func(socketPath string) (hypervisor.Hypervisor, error) {
 		return New(socketPath)
 	})

lib/hypervisor/hypervisor.go

@@ -45,6 +45,10 @@ var socketNames = make(map[Type]string)
 // Registered by hypervisor packages when they use socket-based vsock routing.
 var vsockSocketNames = make(map[Type]string)
 
+// capabilitiesByType maps hypervisor types to their static capabilities.
+// Registered by each hypervisor package's init() function.
+var capabilitiesByType = make(map[Type]Capabilities)
+
 // RegisterSocketName registers the socket filename for a hypervisor type.
 // Called by each hypervisor implementation's init() function.
 func RegisterSocketName(t Type, name string) {
@@ -74,6 +78,17 @@ func VsockSocketNameForType(t Type) string {
 	return "vsock.sock"
 }
 
+// RegisterCapabilities registers static capabilities for a hypervisor type.
+func RegisterCapabilities(t Type, caps Capabilities) {
+	capabilitiesByType[t] = caps
+}
+
+// CapabilitiesForType returns static capabilities for a hypervisor type.
+func CapabilitiesForType(t Type) (Capabilities, bool) {
+	caps, ok := capabilitiesByType[t]
+	return caps, ok
+}
+
 // VMStarter handles the full VM startup sequence.
 // Each hypervisor implements its own startup flow:
 // - Cloud Hypervisor: starts process, configures via HTTP API, boots via HTTP API
@@ -197,6 +212,14 @@ type Capabilities struct {
 
 	// SupportsDiskIOLimit indicates if disk I/O rate limiting is available
 	SupportsDiskIOLimit bool
+
+	// SupportsGracefulVMMShutdown indicates the hypervisor exposes an API to
+	// ask the VMM process itself to exit cleanly.
+	SupportsGracefulVMMShutdown bool
+
+	// SupportsSnapshotBaseReuse indicates snapshots can safely reuse a retained
+	// on-disk base across restore/standby cycles.
+	SupportsSnapshotBaseReuse bool
 }
 
 // VsockDialer provides vsock connectivity to a guest VM.

lib/hypervisor/qemu/process.go

@@ -43,6 +43,7 @@ const (
 
 func init() {
 	hypervisor.RegisterSocketName(hypervisor.TypeQEMU, "qemu.sock")
+	hypervisor.RegisterCapabilities(hypervisor.TypeQEMU, capabilities())
 	hypervisor.RegisterClientFactory(hypervisor.TypeQEMU, func(socketPath string) (hypervisor.Hypervisor, error) {
 		return New(socketPath)
 	})

lib/hypervisor/qemu/qemu.go

@@ -37,13 +37,19 @@ var _ hypervisor.Hypervisor = (*QEMU)(nil)
 
 // Capabilities returns the features supported by QEMU.
 func (q *QEMU) Capabilities() hypervisor.Capabilities {
+	return capabilities()
+}
+
+func capabilities() hypervisor.Capabilities {
 	return hypervisor.Capabilities{
-		SupportsSnapshot:       true,  // Uses QMP migrate file:// for snapshot
-		SupportsHotplugMemory:  false, // Not implemented - balloon not configured
-		SupportsPause:          true,
-		SupportsVsock:          true,
-		SupportsGPUPassthrough: true,
-		SupportsDiskIOLimit:    true,
+		SupportsSnapshot:            true,  // Uses QMP migrate file:// for snapshot
+		SupportsHotplugMemory:       false, // Not implemented - balloon not configured
+		SupportsPause:               true,
+		SupportsVsock:               true,
+		SupportsGPUPassthrough:      true,
+		SupportsDiskIOLimit:         true,
+		SupportsGracefulVMMShutdown: true,
+		SupportsSnapshotBaseReuse:   false,
 	}
 }
 

lib/hypervisor/vz/client.go

@@ -71,13 +71,19 @@ type snapshotRequest struct {
 }
 
 func (c *Client) Capabilities() hypervisor.Capabilities {
+	return capabilities()
+}
+
+func capabilities() hypervisor.Capabilities {
 	return hypervisor.Capabilities{
-		SupportsSnapshot:       runtime.GOARCH == "arm64",
-		SupportsHotplugMemory:  false,
-		SupportsPause:          true,
-		SupportsVsock:          true,
-		SupportsGPUPassthrough: false,
-		SupportsDiskIOLimit:    false,
+		SupportsSnapshot:            runtime.GOARCH == "arm64",
+		SupportsHotplugMemory:       false,
+		SupportsPause:               true,
+		SupportsVsock:               true,
+		SupportsGPUPassthrough:      false,
+		SupportsDiskIOLimit:         false,
+		SupportsGracefulVMMShutdown: true,
+		SupportsSnapshotBaseReuse:   false,
 	}
 }
 

lib/hypervisor/vz/starter.go

@@ -24,6 +24,7 @@ import (
 
 func init() {
 	hypervisor.RegisterSocketName(hypervisor.TypeVZ, "vz.sock")
+	hypervisor.RegisterCapabilities(hypervisor.TypeVZ, capabilities())
 	hypervisor.RegisterVsockSocketName(hypervisor.TypeVZ, "vz.vsock")
 	hypervisor.RegisterVsockDialerFactory(hypervisor.TypeVZ, NewVsockDialer)
 	hypervisor.RegisterClientFactory(hypervisor.TypeVZ, func(socketPath string) (hypervisor.Hypervisor, error) {