Add normal/deep type of heal scanning (#7251)

Healing scan used to read all objects parts to check for bitrot
checksum. This commit will add a quicker way of healing scan
by only checking if parts are actually present in disks or not.

cmd/admin-heal-ops.go

@@ -575,7 +575,7 @@ func (h *healSequence) healMinioSysMeta(metaPrefix string) func() error {
 			if h.isQuitting() {
 				return errHealStopSignalled
 			}
-			res, herr := objectAPI.HealObject(h.ctx, bucket, object, h.settings.DryRun, h.settings.Remove)
+			res, herr := objectAPI.HealObject(h.ctx, bucket, object, h.settings.DryRun, h.settings.Remove, h.settings.ScanMode)
 			// Object might have been deleted, by the time heal
 			// was attempted we ignore this object an move on.
 			if isErrObjectNotFound(herr) {
@@ -718,7 +718,7 @@ func (h *healSequence) healObject(bucket, object string) error {
 		return errServerNotInitialized
 	}
 
-	hri, err := objectAPI.HealObject(h.ctx, bucket, object, h.settings.DryRun, h.settings.Remove)
+	hri, err := objectAPI.HealObject(h.ctx, bucket, object, h.settings.DryRun, h.settings.Remove, h.settings.ScanMode)
 	if isErrObjectNotFound(err) {
 		return nil
 	}

cmd/fs-v1.go

@@ -1240,7 +1240,7 @@ func (fs *FSObjects) HealFormat(ctx context.Context, dryRun bool) (madmin.HealRe
 }
 
 // HealObject - no-op for fs. Valid only for XL.
-func (fs *FSObjects) HealObject(ctx context.Context, bucket, object string, dryRun, remove bool) (
+func (fs *FSObjects) HealObject(ctx context.Context, bucket, object string, dryRun, remove bool, scanMode madmin.HealScanMode) (
 	res madmin.HealResultItem, err error) {
 	logger.LogIf(ctx, NotImplemented{})
 	return res, NotImplemented{}

cmd/fs-v1_test.go

@@ -22,6 +22,8 @@ import (
 	"os"
 	"path/filepath"
 	"testing"
+
+	"github.com/minio/minio/pkg/madmin"
 )
 
 // Tests for if parent directory is object
@@ -390,7 +392,7 @@ func TestFSHealObject(t *testing.T) {
 	defer os.RemoveAll(disk)
 
 	obj := initFSObjects(disk, t)
-	_, err := obj.HealObject(context.Background(), "bucket", "object", false, false)
+	_, err := obj.HealObject(context.Background(), "bucket", "object", false, false, madmin.HealDeepScan)
 	if err == nil || !isSameType(err, NotImplemented{}) {
 		t.Fatalf("Heal Object should return NotImplemented error ")
 	}

cmd/gateway-unsupported.go

@@ -102,7 +102,7 @@ func (a GatewayUnsupported) ListBucketsHeal(ctx context.Context) (buckets []Buck
 }
 
 // HealObject - Not implemented stub
-func (a GatewayUnsupported) HealObject(ctx context.Context, bucket, object string, dryRun, remove bool) (h madmin.HealResultItem, e error) {
+func (a GatewayUnsupported) HealObject(ctx context.Context, bucket, object string, dryRun, remove bool, scanMode madmin.HealScanMode) (h madmin.HealResultItem, e error) {
 	return h, NotImplemented{}
 }
 

cmd/object-api-interface.go

@@ -88,7 +88,7 @@ type ObjectLayer interface {
 	ReloadFormat(ctx context.Context, dryRun bool) error
 	HealFormat(ctx context.Context, dryRun bool) (madmin.HealResultItem, error)
 	HealBucket(ctx context.Context, bucket string, dryRun, remove bool) (madmin.HealResultItem, error)
-	HealObject(ctx context.Context, bucket, object string, dryRun, remove bool) (madmin.HealResultItem, error)
+	HealObject(ctx context.Context, bucket, object string, dryRun, remove bool, scanMode madmin.HealScanMode) (madmin.HealResultItem, error)
 	ListBucketsHeal(ctx context.Context) (buckets []BucketInfo, err error)
 	HealObjects(ctx context.Context, bucket, prefix string, healObjectFn func(string, string) error) error
 

cmd/xl-sets.go

@@ -1296,8 +1296,8 @@ func (s *xlSets) HealBucket(ctx context.Context, bucket string, dryRun, remove b
 }
 
 // HealObject - heals inconsistent object on a hashedSet based on object name.
-func (s *xlSets) HealObject(ctx context.Context, bucket, object string, dryRun, remove bool) (madmin.HealResultItem, error) {
-	return s.getHashedSet(object).HealObject(ctx, bucket, object, dryRun, remove)
+func (s *xlSets) HealObject(ctx context.Context, bucket, object string, dryRun, remove bool, scanMode madmin.HealScanMode) (madmin.HealResultItem, error) {
+	return s.getHashedSet(object).HealObject(ctx, bucket, object, dryRun, remove, scanMode)
 }
 
 // Lists all buckets which need healing.

cmd/xl-v1-healing-common.go

@@ -22,6 +22,7 @@ import (
 	"time"
 
 	"github.com/minio/minio/cmd/logger"
+	"github.com/minio/minio/pkg/madmin"
 )
 
 // commonTime returns a maximally occurring time from a list of time.
@@ -158,7 +159,7 @@ func getLatestXLMeta(ctx context.Context, partsMetadata []xlMetaV1, errs []error
 // - slice of errors about the state of data files on disk - can have
 //   a not-found error or a hash-mismatch error.
 func disksWithAllParts(ctx context.Context, onlineDisks []StorageAPI, partsMetadata []xlMetaV1, errs []error, bucket,
-	object string) ([]StorageAPI, []error) {
+	object string, scanMode madmin.HealScanMode) ([]StorageAPI, []error) {
 	availableDisks := make([]StorageAPI, len(onlineDisks))
 	dataErrs := make([]error, len(onlineDisks))
 
@@ -168,27 +169,38 @@ func disksWithAllParts(ctx context.Context, onlineDisks []StorageAPI, partsMetad
 			continue
 		}
 
-		erasureInfo := partsMetadata[i].Erasure
-		erasure, err := NewErasure(ctx, erasureInfo.DataBlocks, erasureInfo.ParityBlocks, erasureInfo.BlockSize)
-		if err != nil {
-			dataErrs[i] = err
-			continue
-		}
-
-		// disk has a valid xl.json but may not have all the
-		// parts. This is considered an outdated disk, since
-		// it needs healing too.
-		for _, part := range partsMetadata[i].Parts {
-			checksumInfo := erasureInfo.GetChecksumInfo(part.Name)
-			tillOffset := erasure.ShardFileTillOffset(0, part.Size, part.Size)
-			err = bitrotCheckFile(onlineDisk, bucket, pathJoin(object, part.Name), tillOffset, checksumInfo.Algorithm, checksumInfo.Hash, erasure.ShardSize())
+		switch scanMode {
+		case madmin.HealDeepScan:
+			erasureInfo := partsMetadata[i].Erasure
+			erasure, err := NewErasure(ctx, erasureInfo.DataBlocks, erasureInfo.ParityBlocks, erasureInfo.BlockSize)
 			if err != nil {
-				isCorrupt := strings.HasPrefix(err.Error(), "Bitrot verification mismatch - expected ")
-				if !isCorrupt && err != errFileNotFound && err != errVolumeNotFound {
-					logger.LogIf(ctx, err)
-				}
 				dataErrs[i] = err
-				break
+				continue
+			}
+
+			// disk has a valid xl.json but may not have all the
+			// parts. This is considered an outdated disk, since
+			// it needs healing too.
+			for _, part := range partsMetadata[i].Parts {
+				checksumInfo := erasureInfo.GetChecksumInfo(part.Name)
+				tillOffset := erasure.ShardFileTillOffset(0, part.Size, part.Size)
+				err = bitrotCheckFile(onlineDisk, bucket, pathJoin(object, part.Name), tillOffset, checksumInfo.Algorithm, checksumInfo.Hash, erasure.ShardSize())
+				if err != nil {
+					isCorrupt := strings.HasPrefix(err.Error(), "Bitrot verification mismatch - expected ")
+					if !isCorrupt && err != errFileNotFound && err != errVolumeNotFound {
+						logger.LogIf(ctx, err)
+					}
+					dataErrs[i] = err
+					break
+				}
+			}
+		case madmin.HealNormalScan:
+			for _, part := range partsMetadata[i].Parts {
+				_, err := onlineDisk.StatFile(bucket, pathJoin(object, part.Name))
+				if err != nil {
+					dataErrs[i] = err
+					break
+				}
 			}
 		}
 

cmd/xl-v1-healing-common_test.go

@@ -23,6 +23,8 @@ import (
 	"path/filepath"
 	"testing"
 	"time"
+
+	"github.com/minio/minio/pkg/madmin"
 )
 
 // validates functionality provided to find most common
@@ -239,7 +241,7 @@ func TestListOnlineDisks(t *testing.T) {
 				i+1, test.expectedTime, modTime)
 		}
 
-		availableDisks, newErrs := disksWithAllParts(context.Background(), onlineDisks, partsMetadata, test.errs, bucket, object)
+		availableDisks, newErrs := disksWithAllParts(context.Background(), onlineDisks, partsMetadata, test.errs, bucket, object, madmin.HealDeepScan)
 		test.errs = newErrs
 
 		if test._tamperBackend != noTamper {
@@ -291,7 +293,7 @@ func TestDisksWithAllParts(t *testing.T) {
 		t.Fatalf("Failed to read xl meta data %v", err)
 	}
 
-	filteredDisks, errs := disksWithAllParts(ctx, xlDisks, partsMetadata, errs, bucket, object)
+	filteredDisks, errs := disksWithAllParts(ctx, xlDisks, partsMetadata, errs, bucket, object, madmin.HealDeepScan)
 
 	if len(filteredDisks) != len(xlDisks) {
 		t.Errorf("Unexpected number of disks: %d", len(filteredDisks))
@@ -328,7 +330,7 @@ func TestDisksWithAllParts(t *testing.T) {
 	}
 
 	errs = make([]error, len(xlDisks))
-	filteredDisks, errs = disksWithAllParts(ctx, xlDisks, partsMetadata, errs, bucket, object)
+	filteredDisks, errs = disksWithAllParts(ctx, xlDisks, partsMetadata, errs, bucket, object, madmin.HealDeepScan)
 
 	if len(filteredDisks) != len(xlDisks) {
 		t.Errorf("Unexpected number of disks: %d", len(filteredDisks))

cmd/xl-v1-healing.go

@@ -208,7 +208,8 @@ func shouldHealObjectOnDisk(xlErr, dataErr error, meta xlMetaV1, quorumModTime t
 
 // Heals an object by re-writing corrupt/missing erasure blocks.
 func healObject(ctx context.Context, storageDisks []StorageAPI, bucket string, object string,
-	quorum int, dryRun bool) (result madmin.HealResultItem, err error) {
+	quorum int, dryRun bool, scanMode madmin.HealScanMode) (result madmin.HealResultItem, err error) {
+
 	partsMetadata, errs := readAllXLMetadata(ctx, storageDisks, bucket, object)
 
 	errCount := 0
@@ -232,7 +233,7 @@ func healObject(ctx context.Context, storageDisks []StorageAPI, bucket string, o
 	latestDisks, modTime := listOnlineDisks(storageDisks, partsMetadata, errs)
 
 	// List of disks having all parts as per latest xl.json.
-	availableDisks, dataErrs := disksWithAllParts(ctx, latestDisks, partsMetadata, errs, bucket, object)
+	availableDisks, dataErrs := disksWithAllParts(ctx, latestDisks, partsMetadata, errs, bucket, object, scanMode)
 
 	// Initialize heal result object
 	result = madmin.HealResultItem{
@@ -621,7 +622,7 @@ func (xl xlObjects) isObjectDangling(metaArr []xlMetaV1, errs []error) (validMet
 // FIXME: If an object object was deleted and one disk was down,
 // and later the disk comes back up again, heal on the object
 // should delete it.
-func (xl xlObjects) HealObject(ctx context.Context, bucket, object string, dryRun bool, remove bool) (hr madmin.HealResultItem, err error) {
+func (xl xlObjects) HealObject(ctx context.Context, bucket, object string, dryRun bool, remove bool, scanMode madmin.HealScanMode) (hr madmin.HealResultItem, err error) {
 	// Create context that also contains information about the object and bucket.
 	// The top level handler might not have this information.
 	reqInfo := logger.GetReqInfo(ctx)
@@ -670,5 +671,5 @@ func (xl xlObjects) HealObject(ctx context.Context, bucket, object string, dryRu
 	defer objectLock.RUnlock()
 
 	// Heal the object.
-	return healObject(healCtx, xl.getDisks(), bucket, object, latestXLMeta.Erasure.DataBlocks, dryRun)
+	return healObject(healCtx, xl.getDisks(), bucket, object, latestXLMeta.Erasure.DataBlocks, dryRun, scanMode)
 }

cmd/xl-v1-healing_test.go

@@ -21,6 +21,8 @@ import (
 	"context"
 	"path/filepath"
 	"testing"
+
+	"github.com/minio/minio/pkg/madmin"
 )
 
 // Tests undoes and validates if the undoing completes successfully.
@@ -114,7 +116,7 @@ func TestHealObjectXL(t *testing.T) {
 		t.Fatalf("Failed to delete a file - %v", err)
 	}
 
-	_, err = obj.HealObject(context.Background(), bucket, object, false, false)
+	_, err = obj.HealObject(context.Background(), bucket, object, false, false, madmin.HealNormalScan)
 	if err != nil {
 		t.Fatalf("Failed to heal object - %v", err)
 	}
@@ -130,7 +132,7 @@ func TestHealObjectXL(t *testing.T) {
 	}
 
 	// Try healing now, expect to receive errDiskNotFound.
-	_, err = obj.HealObject(context.Background(), bucket, object, false, false)
+	_, err = obj.HealObject(context.Background(), bucket, object, false, false, madmin.HealDeepScan)
 	// since majority of xl.jsons are not available, object quorum can't be read properly and error will be errXLReadQuorum
 	if _, ok := err.(InsufficientReadQuorum); !ok {
 		t.Errorf("Expected %v but received %v", InsufficientReadQuorum{}, err)

cmd/xl-v1-object_test.go

@@ -28,6 +28,7 @@ import (
 	"time"
 
 	humanize "github.com/dustin/go-humanize"
+	"github.com/minio/minio/pkg/madmin"
 )
 
 func TestRepeatPutObjectPart(t *testing.T) {
@@ -308,7 +309,7 @@ func TestHealing(t *testing.T) {
 		t.Fatal(err)
 	}
 
-	_, err = xl.HealObject(context.Background(), bucket, object, false, false)
+	_, err = xl.HealObject(context.Background(), bucket, object, false, false, madmin.HealNormalScan)
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -337,7 +338,7 @@ func TestHealing(t *testing.T) {
 		t.Fatal(err)
 	}
 
-	_, err = xl.HealObject(context.Background(), bucket, object, false, false)
+	_, err = xl.HealObject(context.Background(), bucket, object, false, false, madmin.HealDeepScan)
 	if err != nil {
 		t.Fatal(err)
 	}

pkg/madmin/heal-commands.go

@@ -26,11 +26,22 @@ import (
 	"time"
 )
 
+// HealScanMode represents the type of healing scan
+type HealScanMode int
+
+const (
+	// HealNormalScan checks if parts are present and not outdated
+	HealNormalScan HealScanMode = iota
+	// HealDeepScan checks for parts bitrot checksums
+	HealDeepScan
+)
+
 // HealOpts - collection of options for a heal sequence
 type HealOpts struct {
-	Recursive bool `json:"recursive"`
-	DryRun    bool `json:"dryRun"`
-	Remove    bool `json:"remove"`
+	Recursive bool         `json:"recursive"`
+	DryRun    bool         `json:"dryRun"`
+	Remove    bool         `json:"remove"`
+	ScanMode  HealScanMode `json:"scanMode"`
 }
 
 // HealStartSuccess - holds information about a successfully started