heal: Avoid marking a bucket as done when remote drives are offline (#19587)

cmd/erasure-healing.go

@@ -44,8 +44,8 @@ const (
 	healingMetricCheckAbandonedParts
 )
 
-func (er erasureObjects) listAndHeal(bucket, prefix string, scanMode madmin.HealScanMode, healEntry func(string, metaCacheEntry, madmin.HealScanMode) error) error {
-	ctx, cancel := context.WithCancel(context.Background())
+func (er erasureObjects) listAndHeal(ctx context.Context, bucket, prefix string, scanMode madmin.HealScanMode, healEntry func(string, metaCacheEntry, madmin.HealScanMode) error) error {
+	ctx, cancel := context.WithCancel(ctx)
 	defer cancel()
 
 	disks, _ := er.getOnlineDisksWithHealing(false)

cmd/erasure-server-pool.go

@@ -2270,7 +2270,7 @@ func (z *erasureServerPools) HealObjects(ctx context.Context, bucket, prefix str
 			go func(idx int, set *erasureObjects) {
 				defer wg.Done()
 
-				errs[idx] = set.listAndHeal(bucket, prefix, opts.ScanMode, healEntry)
+				errs[idx] = set.listAndHeal(ctx, bucket, prefix, opts.ScanMode, healEntry)
 			}(idx, set)
 		}
 		wg.Wait()

cmd/erasure.go

@@ -272,11 +272,11 @@ func (er erasureObjects) LocalStorageInfo(ctx context.Context, metrics bool) Sto
 }
 
 // getOnlineDisksWithHealingAndInfo - returns online disks and overall healing status.
-// Disks are randomly ordered, but in the following groups:
+// Disks are ordered in the following groups:
 // - Non-scanning disks
 // - Non-healing disks
 // - Healing disks (if inclHealing is true)
-func (er erasureObjects) getOnlineDisksWithHealingAndInfo(inclHealing bool) (newDisks []StorageAPI, newInfos []DiskInfo, healing bool) {
+func (er erasureObjects) getOnlineDisksWithHealingAndInfo(inclHealing bool) (newDisks []StorageAPI, newInfos []DiskInfo, healing int) {
 	var wg sync.WaitGroup
 	disks := er.getDisks()
 	infos := make([]DiskInfo, len(disks))
@@ -315,7 +315,7 @@ func (er erasureObjects) getOnlineDisksWithHealingAndInfo(inclHealing bool) (new
 			continue
 		}
 		if info.Healing {
-			healing = true
+			healing++
 			if inclHealing {
 				healingDisks = append(healingDisks, disks[i])
 				healingInfos = append(healingInfos, infos[i])
@@ -343,9 +343,9 @@ func (er erasureObjects) getOnlineDisksWithHealingAndInfo(inclHealing bool) (new
 	return newDisks, newInfos, healing
 }
 
-func (er erasureObjects) getOnlineDisksWithHealing(inclHealing bool) (newDisks []StorageAPI, healing bool) {
-	newDisks, _, healing = er.getOnlineDisksWithHealingAndInfo(inclHealing)
-	return
+func (er erasureObjects) getOnlineDisksWithHealing(inclHealing bool) ([]StorageAPI, bool) {
+	newDisks, _, healing := er.getOnlineDisksWithHealingAndInfo(inclHealing)
+	return newDisks, healing > 0
 }
 
 // Clean-up previously deleted objects. from .minio.sys/tmp/.trash/

cmd/global-heal.go

@@ -259,12 +259,17 @@ func (er *erasureObjects) healErasureSet(ctx context.Context, buckets []string,
 				bucket, humanize.Ordinal(er.setIndex+1))
 		}
 
-		disks, _ := er.getOnlineDisksWithHealing(false)
-		if len(disks) == 0 {
-			// No object healing necessary
-			tracker.bucketDone(bucket)
-			healingLogIf(ctx, tracker.update(ctx))
-			continue
+		disks, _, healing := er.getOnlineDisksWithHealingAndInfo(true)
+		if len(disks) == healing {
+			// All drives in this erasure set were reformatted for some reasons, abort healing and mark it as successful
+			healingLogIf(ctx, errors.New("all drives are in healing state, aborting.."))
+			return nil
+		}
+
+		disks = disks[:len(disks)-healing] // healing drives are always at the end of the list
+
+		if len(disks) < er.setDriveCount/2 {
+			return fmt.Errorf("not enough drives (found=%d, healing=%d, total=%d) are available to heal `%s`", len(disks), healing, er.setDriveCount, tracker.disk.String())
 		}
 
 		rand.Shuffle(len(disks), func(i, j int) {
@@ -465,27 +470,24 @@ func (er *erasureObjects) healErasureSet(ctx context.Context, buckets []string,
 			waitForLowHTTPReq()
 		}
 
-		actualBucket, prefix := path2BucketObject(bucket)
-
 		// How to resolve partial results.
 		resolver := metadataResolutionParams{
 			dirQuorum: 1,
 			objQuorum: 1,
-			bucket:    actualBucket,
+			bucket:    bucket,
 		}
 
 		err = listPathRaw(ctx, listPathRawOptions{
 			disks:          disks,
 			fallbackDisks:  fallbackDisks,
-			bucket:         actualBucket,
-			path:           prefix,
+			bucket:         bucket,
 			recursive:      true,
 			forwardTo:      forwardTo,
 			minDisks:       1,
 			reportNotFound: false,
 			agreed: func(entry metaCacheEntry) {
 				jt.Take()
-				go healEntry(actualBucket, entry)
+				go healEntry(bucket, entry)
 			},
 			partial: func(entries metaCacheEntries, _ []error) {
 				entry, ok := entries.resolve(&resolver)
@@ -495,7 +497,7 @@ func (er *erasureObjects) healErasureSet(ctx context.Context, buckets []string,
 					entry, _ = entries.firstFound()
 				}
 				jt.Take()
-				go healEntry(actualBucket, *entry)
+				go healEntry(bucket, *entry)
 			},
 			finished: nil,
 		})