Continue healing other objects even if objects without quorum exist (#5851)

fixes #5815

cmd/xl-sets.go

@@ -1379,19 +1379,8 @@ func (s *xlSets) listObjectsHeal(ctx context.Context, bucket, prefix, marker, de
 			return loi, toObjectErr(walkResult.err, bucket, prefix)
 		}
 		var objInfo ObjectInfo
-		var err error
+		objInfo.Bucket = bucket
-		if hasSuffix(walkResult.entry, slashSeparator) {
+		objInfo.Name = walkResult.entry
-			objInfo, err = s.getHashedSet(walkResult.entry).getObjectInfoDir(ctx, bucket, walkResult.entry)
-		} else {
-			objInfo, err = s.getHashedSet(walkResult.entry).getObjectInfo(ctx, bucket, walkResult.entry)
-		}
-		if err != nil {
-			// Ignore errFileNotFound
-			if err == errFileNotFound {
-				continue
-			}
-			return loi, toObjectErr(err, bucket, prefix)
-		}
 		nextMarker = objInfo.Name
 		objInfos = append(objInfos, objInfo)
 		i++

cmd/xl-v1-healing.go

@@ -284,12 +284,21 @@ func healObject(ctx context.Context, storageDisks []StorageAPI, bucket string, o
 
 	partsMetadata, errs := readAllXLMetadata(ctx, storageDisks, bucket, object)
 
-	// readQuorum suffices for xl.json since we use monotonic
+	errCount := 0
-	// system time to break the tie when a split-brain situation
+	for _, err := range errs {
-	// arises.
+		if err != nil {
+			errCount++
+		}
+	}
+
+	if errCount == len(errs) {
+		// Only if we get errors from all the disks we return error. Else we need to
+		// continue to return filled madmin.HealResultItem struct which includes info
+		// on what disks the file is available etc.
 		if reducedErr := reduceReadQuorumErrs(ctx, errs, nil, quorum); reducedErr != nil {
 			return result, toObjectErr(reducedErr, bucket, object)
 		}
+	}
 
 	// List of disks having latest version of the object xl.json
 	// (by modtime).
@@ -536,17 +545,11 @@ func healObject(ctx context.Context, storageDisks []StorageAPI, bucket string, o
 // 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) (hr madmin.HealResultItem, err error) {
-
 	// FIXME: Metadata is read again in the healObject() call below.
 	// Read metadata files from all the disks
 	partsMetadata, errs := readAllXLMetadata(ctx, xl.getDisks(), bucket, object)
 
-	// get read quorum for this object
+	latestXLMeta, _ := getLatestXLMeta(partsMetadata, errs)
-	var readQuorum int
-	readQuorum, _, err = objectQuorumFromMeta(xl, partsMetadata, errs)
-	if err != nil {
-		return hr, err
-	}
 
 	// Lock the object before healing.
 	objectLock := xl.nsMutex.NewNSLock(bucket, object)
@@ -556,5 +559,5 @@ func (xl xlObjects) HealObject(ctx context.Context, bucket, object string, dryRu
 	defer objectLock.RUnlock()
 
 	// Heal the object.
-	return healObject(ctx, xl.getDisks(), bucket, object, readQuorum, dryRun)
+	return healObject(ctx, xl.getDisks(), bucket, object, latestXLMeta.Erasure.DataBlocks, dryRun)
 }

cmd/xl-v1-healing_test.go

@@ -144,7 +144,7 @@ func TestHealObjectXL(t *testing.T) {
 	// Try healing now, expect to receive errDiskNotFound.
 	_, err = obj.HealObject(context.Background(), bucket, object, false)
 	// since majority of xl.jsons are not available, object quorum can't be read properly and error will be errXLReadQuorum
-	if err != errXLReadQuorum {
+	if _, ok := err.(InsufficientReadQuorum); !ok {
-		t.Errorf("Expected %v but received %v", errDiskNotFound, err)
+		t.Errorf("Expected %v but received %v", InsufficientWriteQuorum{}, err)
 	}
 }

cmd/xl-v1-metadata.go

@@ -387,8 +387,7 @@ func (m xlMetaV1) ObjectToPartOffset(ctx context.Context, offset int64) (partInd
 }
 
 // pickValidXLMeta - picks one valid xlMeta content and returns from a
-// slice of xlmeta content. If no value is found this function panics
+// slice of xlmeta content.
-// and dies.
 func pickValidXLMeta(ctx context.Context, metaArr []xlMetaV1, modTime time.Time) (xmv xlMetaV1, e error) {
 	// Pick latest valid metadata.
 	for _, meta := range metaArr {