DRA: refactor checkpointing

Co-authored-by: Kevin Klues <[email protected]>

Author: bart0sh

pkg/kubelet/checkpointmanager/checkpoint_manager.go

@@ -85,9 +85,6 @@ func (manager *impl) GetCheckpoint(checkpointKey string, checkpoint Checkpoint)
 		return err
 	}
 	err = checkpoint.UnmarshalCheckpoint(blob)
-	if err == nil {
-		err = checkpoint.VerifyChecksum()
-	}
 	return err
 }
 

pkg/kubelet/cm/dra/state/state_checkpoint.go renamed to pkg/kubelet/cm/dra/checkpoint/state_checkpoint.go

@@ -20,55 +20,19 @@ import (
 	"fmt"
 	"sync"
 
-	"k8s.io/apimachinery/pkg/types"
-	"k8s.io/apimachinery/pkg/util/sets"
 	"k8s.io/kubernetes/pkg/kubelet/checkpointmanager"
 	"k8s.io/kubernetes/pkg/kubelet/checkpointmanager/errors"
+	checkpointapi "k8s.io/kubernetes/pkg/kubelet/cm/dra/checkpoint/v1"
 )
 
 var _ CheckpointState = &stateCheckpoint{}
 
-// CheckpointState interface provides to get and store state
 type CheckpointState interface {
 	GetOrCreate() (ClaimInfoStateList, error)
 	Store(ClaimInfoStateList) error
 }
 
-// ClaimInfoState is used to store claim info state in a checkpoint
-// +k8s:deepcopy-gen=true
-type ClaimInfoState struct {
-	// ClaimUID is an UID of the resource claim
-	ClaimUID types.UID
-
-	// ClaimName is a name of the resource claim
-	ClaimName string
-
-	// Namespace is a claim namespace
-	Namespace string
-
-	// PodUIDs is a set of pod UIDs that reference a resource
-	PodUIDs sets.Set[string]
-
-	// DriverState contains information about all drivers which have allocation
-	// results in the claim, even if they don't provide devices for their results.
-	DriverState map[string]DriverState
-}
-
-// DriverState is used to store per-device claim info state in a checkpoint
-// +k8s:deepcopy-gen=true
-type DriverState struct {
-	Devices []Device
-}
-
-// Device is how a DRA driver described an allocated device in a claim
-// to kubelet. RequestName and CDI device IDs are optional.
-// +k8s:deepcopy-gen=true
-type Device struct {
-	PoolName     string
-	DeviceName   string
-	RequestNames []string
-	CDIDeviceIDs []string
-}
+type ClaimInfoStateList []checkpointapi.ClaimInfoState
 
 type stateCheckpoint struct {
 	sync.RWMutex
@@ -77,7 +41,7 @@ type stateCheckpoint struct {
 }
 
 // NewCheckpointState creates new State for keeping track of claim info  with checkpoint backend
-func NewCheckpointState(stateDir, checkpointName string) (*stateCheckpoint, error) {
+func NewCheckpointState(stateDir, checkpointName string) (CheckpointState, error) {
 	if len(checkpointName) == 0 {
 		return nil, fmt.Errorf("received empty string instead of checkpointName")
 	}
@@ -99,7 +63,7 @@ func (sc *stateCheckpoint) GetOrCreate() (ClaimInfoStateList, error) {
 	sc.Lock()
 	defer sc.Unlock()
 
-	checkpoint := NewDRAManagerCheckpoint()
+	checkpoint := checkpointapi.New(nil)
 	err := sc.checkpointManager.GetCheckpoint(sc.checkpointName, checkpoint)
 	if err == errors.ErrCheckpointNotFound {
 		sc.store(ClaimInfoStateList{})
@@ -109,7 +73,7 @@ func (sc *stateCheckpoint) GetOrCreate() (ClaimInfoStateList, error) {
 		return nil, fmt.Errorf("failed to get checkpoint %v: %w", sc.checkpointName, err)
 	}
 
-	return checkpoint.Entries, nil
+	return checkpoint.Data, nil
 }
 
 // saves state to a checkpoint
@@ -122,8 +86,7 @@ func (sc *stateCheckpoint) Store(claimInfoStateList ClaimInfoStateList) error {
 
 // saves state to a checkpoint, caller is responsible for locking
 func (sc *stateCheckpoint) store(claimInfoStateList ClaimInfoStateList) error {
-	checkpoint := NewDRAManagerCheckpoint()
-	checkpoint.Entries = claimInfoStateList
+	checkpoint := checkpointapi.New(claimInfoStateList)
 
 	err := sc.checkpointManager.CreateCheckpoint(sc.checkpointName, checkpoint)
 	if err != nil {