Only heal on disks where we are sure that healing is needed (#7148)

cmd/storage-rest-client.go

@@ -29,6 +29,9 @@ import (
 	"encoding/gob"
 	"encoding/hex"
 
+	"fmt"
+	"strings"
+
 	"github.com/minio/minio/cmd/logger"
 	"github.com/minio/minio/cmd/rest"
 	xnet "github.com/minio/minio/pkg/net"
@@ -101,6 +104,15 @@ func toStorageErr(err error) error {
 	case errServerTimeMismatch.Error():
 		return errServerTimeMismatch
 	}
+	if strings.Contains(err.Error(), "Bitrot verification mismatch") {
+		var expected string
+		var received string
+		fmt.Sscanf(err.Error(), "Bitrot verification mismatch - expected %s received %s", &expected, &received)
+		// Go's Sscanf %s scans "," that comes after the expected hash, hence remove it. Providing "," in the format string does not help.
+		expected = strings.TrimSuffix(expected, ",")
+		bitrotErr := hashMismatchError{expected, received}
+		return bitrotErr
+	}
 	return err
 }
 

cmd/xl-v1-healing.go

@@ -22,6 +22,7 @@ import (
 	"io"
 	"path"
 	"sync"
+	"time"
 
 	"github.com/minio/minio/cmd/logger"
 	"github.com/minio/minio/pkg/madmin"
@@ -237,6 +238,30 @@ func listAllBuckets(storageDisks []StorageAPI) (buckets map[string]VolInfo,
 	return buckets, bucketsOcc, nil
 }
 
+// Only heal on disks where we are sure that healing is needed. We can expand
+// this list as and when we figure out more errors can be added to this list safely.
+func shouldHealObjectOnDisk(xlErr, dataErr error, meta xlMetaV1, quorumModTime time.Time) bool {
+	switch xlErr {
+	case errFileNotFound:
+		return true
+	case errCorruptedFormat:
+		return true
+	}
+	if xlErr == nil {
+		// If xl.json was read fine but there is some problem with the part.N files.
+		if dataErr == errFileNotFound {
+			return true
+		}
+		if _, ok := dataErr.(hashMismatchError); ok {
+			return true
+		}
+		if quorumModTime != meta.Stat.ModTime {
+			return true
+		}
+	}
+	return false
+}
+
 // 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) {
@@ -305,17 +330,13 @@ func healObject(ctx context.Context, storageDisks []StorageAPI, bucket string, o
 			driveState = madmin.DriveStateCorrupt
 		}
 
-		// an online disk without valid data/metadata is
-		// outdated and can be healed.
-		if errs[i] != errDiskNotFound && v == nil {
-			outDatedDisks[i] = storageDisks[i]
-			disksToHealCount++
-		}
 		var drive string
-		if v == nil {
-			if errs[i] != errDiskNotFound {
-				drive = outDatedDisks[i].String()
+		if storageDisks[i] != nil {
+			drive = storageDisks[i].String()
 		}
+		if shouldHealObjectOnDisk(errs[i], dataErrs[i], partsMetadata[i], modTime) {
+			outDatedDisks[i] = storageDisks[i]
+			disksToHealCount++
 			result.Before.Drives = append(result.Before.Drives, madmin.HealDriveInfo{
 				UUID:     "",
 				Endpoint: drive,
@@ -328,7 +349,6 @@ func healObject(ctx context.Context, storageDisks []StorageAPI, bucket string, o
 			})
 			continue
 		}
-		drive = v.String()
 		result.Before.Drives = append(result.Before.Drives, madmin.HealDriveInfo{
 			UUID:     "",
 			Endpoint: drive,