heal: Enable periodic bitrot scan configuration (minio#14464)

cmd/admin-heal-ops.go

@@ -701,7 +701,7 @@ func (h *healSequence) queueHealTask(source healSource, healType madmin.HealItem
 	if source.opts != nil {
 		task.opts = *source.opts
 	} else {
-		task.opts.ScanMode = globalHealConfig.ScanMode()
+		task.opts.ScanMode = madmin.HealNormalScan
 	}
 
 	h.mutex.Lock()

cmd/data-scanner.go

@@ -21,6 +21,7 @@ import (
 	"bytes"
 	"context"
 	"encoding/binary"
+	"encoding/json"
 	"errors"
 	"fmt"
 	"io/fs"
@@ -103,6 +104,63 @@ func (s *safeDuration) Get() time.Duration {
 	return s.t
 }
 
+func getCycleScanMode(currentCycle, bitrotStartCycle uint64, bitrotStartTime time.Time) madmin.HealScanMode {
+	bitrotCycle := globalHealConfig.BitrotScanCycle()
+	switch bitrotCycle {
+	case -1:
+		return madmin.HealNormalScan
+	case 0:
+		return madmin.HealDeepScan
+	}
+
+	if currentCycle-bitrotStartCycle < healObjectSelectProb {
+		return madmin.HealDeepScan
+	}
+
+	if time.Since(bitrotStartTime) > bitrotCycle {
+		return madmin.HealDeepScan
+	}
+
+	return madmin.HealNormalScan
+}
+
+type backgroundHealInfo struct {
+	BitrotStartTime  time.Time           `json:"bitrotStartTime"`
+	BitrotStartCycle uint64              `json:"bitrotStartCycle"`
+	CurrentScanMode  madmin.HealScanMode `json:"currentScanMode"`
+}
+
+func readBackgroundHealInfo(ctx context.Context, objAPI ObjectLayer) backgroundHealInfo {
+	// Get last healing information
+	buf, err := readConfig(ctx, objAPI, backgroundHealInfoPath)
+	if err != nil {
+		if !errors.Is(err, errConfigNotFound) {
+			logger.LogIf(ctx, err)
+		}
+		return backgroundHealInfo{}
+	}
+	var info backgroundHealInfo
+	err = json.Unmarshal(buf, &info)
+	if err != nil {
+		logger.LogIf(ctx, err)
+		return backgroundHealInfo{}
+	}
+	return info
+}
+
+func saveBackgroundHealInfo(ctx context.Context, objAPI ObjectLayer, info backgroundHealInfo) {
+	b, err := json.Marshal(info)
+	if err != nil {
+		logger.LogIf(ctx, err)
+		return
+	}
+	// Get last healing information
+	err = saveConfig(ctx, objAPI, backgroundHealInfoPath, b)
+	if err != nil {
+		logger.LogIf(ctx, err)
+	}
+}
+
 // runDataScanner will start a data scanner.
 // The function will block until the context is canceled.
 // There should only ever be one scanner running per cluster.
@@ -145,12 +203,24 @@ func runDataScanner(pctx context.Context, objAPI ObjectLayer) {
 				console.Debugln("starting scanner cycle")
 			}
 
+			bgHealInfo := readBackgroundHealInfo(ctx, objAPI)
+			scanMode := getCycleScanMode(nextBloomCycle, bgHealInfo.BitrotStartCycle, bgHealInfo.BitrotStartTime)
+			if bgHealInfo.CurrentScanMode != scanMode {
+				newHealInfo := bgHealInfo
+				newHealInfo.CurrentScanMode = scanMode
+				if scanMode == madmin.HealDeepScan {
+					newHealInfo.BitrotStartTime = time.Now().UTC()
+					newHealInfo.BitrotStartCycle = nextBloomCycle
+				}
+				saveBackgroundHealInfo(ctx, objAPI, newHealInfo)
+			}
+
 			// Wait before starting next cycle and wait on startup.
 			results := make(chan DataUsageInfo, 1)
 			go storeDataUsageInBackend(ctx, objAPI, results)
 			bf, err := globalNotificationSys.updateBloomFilter(ctx, nextBloomCycle)
 			logger.LogIf(ctx, err)
-			err = objAPI.NSScanner(ctx, bf, results, uint32(nextBloomCycle))
+			err = objAPI.NSScanner(ctx, bf, results, uint32(nextBloomCycle), scanMode)
 			logger.LogIf(ctx, err)
 			if err == nil {
 				// Store new cycle...
@@ -182,6 +252,7 @@ type folderScanner struct {
 	dataUsageScannerDebug bool
 	healFolderInclude     uint32 // Include a clean folder one in n cycles.
 	healObjectSelect      uint32 // Do a heal check on an object once every n cycles. Must divide into healFolderInclude
+	scanMode              madmin.HealScanMode
 
 	disks       []StorageAPI
 	disksQuorum int
@@ -250,7 +321,7 @@ var globalScannerStats scannerStats
 // The returned cache will always be valid, but may not be updated from the existing.
 // Before each operation sleepDuration is called which can be used to temporarily halt the scanner.
 // If the supplied context is canceled the function will return at the first chance.
-func scanDataFolder(ctx context.Context, poolIdx, setIdx int, basePath string, cache dataUsageCache, getSize getSizeFn) (dataUsageCache, error) {
+func scanDataFolder(ctx context.Context, poolIdx, setIdx int, basePath string, cache dataUsageCache, getSize getSizeFn, scanMode madmin.HealScanMode) (dataUsageCache, error) {
 	t := UTCNow()
 
 	logPrefix := color.Green("data-usage: ")
@@ -279,6 +350,7 @@ func scanDataFolder(ctx context.Context, poolIdx, setIdx int, basePath string, c
 		dataUsageScannerDebug: intDataUpdateTracker.debug,
 		healFolderInclude:     0,
 		healObjectSelect:      0,
+		scanMode:              scanMode,
 		updates:               cache.Info.updates,
 	}
 
@@ -482,12 +554,15 @@ func (f *folderScanner) scanFolder(ctx context.Context, folder cachedFolder, int
 				debug:       f.dataUsageScannerDebug,
 				lifeCycle:   activeLifeCycle,
 				replication: replicationCfg,
-				heal:        thisHash.modAlt(f.oldCache.Info.NextCycle/folder.objectHealProbDiv, f.healObjectSelect/folder.objectHealProbDiv) && globalIsErasure,
 			}
+
+			item.heal.enabled = thisHash.modAlt(f.oldCache.Info.NextCycle/folder.objectHealProbDiv, f.healObjectSelect/folder.objectHealProbDiv) && globalIsErasure
+			item.heal.bitrot = f.scanMode == madmin.HealDeepScan
+
 			// if the drive belongs to an erasure set
 			// that is already being healed, skip the
 			// healing attempt on this drive.
-			item.heal = item.heal && f.healObjectSelect > 0
+			item.heal.enabled = item.heal.enabled && f.healObjectSelect > 0
 
 			sz, err := f.getSize(item)
 			if err != nil {
@@ -821,8 +896,11 @@ type scannerItem struct {
 	replication replicationConfig
 	lifeCycle   *lifecycle.Lifecycle
 	Typ         fs.FileMode
-	heal        bool // Has the object been selected for heal check?
-	debug       bool
+	heal        struct {
+		enabled bool
+		bitrot  bool
+	} // Has the object been selected for heal check?
+	debug bool
 }
 
 type sizeSummary struct {
@@ -874,9 +952,13 @@ func (i *scannerItem) applyHealing(ctx context.Context, o ObjectLayer, oi Object
 			console.Debugf(applyActionsLogPrefix+" heal checking: %v/%v\n", i.bucket, i.objectPath())
 		}
 	}
+	scanMode := madmin.HealNormalScan
+	if i.heal.bitrot {
+		scanMode = madmin.HealDeepScan
+	}
 	healOpts := madmin.HealOpts{
 		Remove:   healDeleteDangling,
-		ScanMode: globalHealConfig.ScanMode(),
+		ScanMode: scanMode,
 	}
 	res, err := o.HealObject(ctx, i.bucket, i.objectPath(), oi.VersionID, healOpts)
 	if err != nil && !errors.Is(err, NotImplemented{}) {
@@ -1040,7 +1122,7 @@ func (i *scannerItem) applyActions(ctx context.Context, o ObjectLayer, oi Object
 	// from the current deployment, which means we don't have to call healing
 	// routine even if we are asked to do via heal flag.
 	if !applied {
-		if i.heal {
+		if i.heal.enabled {
 			size = i.applyHealing(ctx, o, oi)
 		}
 		// replicate only if lifecycle rules are not applied.

cmd/data-usage.go

@@ -35,6 +35,8 @@ const (
 	dataUsageBloomName     = ".bloomcycle.bin"
 	dataUsageBloomNamePath = bucketMetaPrefix + SlashSeparator + dataUsageBloomName
 
+	backgroundHealInfoPath = bucketMetaPrefix + SlashSeparator + ".background-heal.json"
+
 	dataUsageCacheName = ".usage-cache.bin"
 )
 

cmd/data-usage_test.go

@@ -67,7 +67,7 @@ func TestDataUsageUpdate(t *testing.T) {
 		return
 	}
 
-	got, err := scanDataFolder(context.Background(), 0, 0, base, dataUsageCache{Info: dataUsageCacheInfo{Name: bucket}}, getSize)
+	got, err := scanDataFolder(context.Background(), 0, 0, base, dataUsageCache{Info: dataUsageCacheInfo{Name: bucket}}, getSize, 0)
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -178,7 +178,7 @@ func TestDataUsageUpdate(t *testing.T) {
 	}
 	// Changed dir must be picked up in this many cycles.
 	for i := 0; i < dataUsageUpdateDirCycles; i++ {
-		got, err = scanDataFolder(context.Background(), 0, 0, base, got, getSize)
+		got, err = scanDataFolder(context.Background(), 0, 0, base, got, getSize, 0)
 		got.Info.NextCycle++
 		if err != nil {
 			t.Fatal(err)
@@ -289,7 +289,7 @@ func TestDataUsageUpdatePrefix(t *testing.T) {
 		}
 		return
 	}
-	got, err := scanDataFolder(context.Background(), 0, 0, base, dataUsageCache{Info: dataUsageCacheInfo{Name: "bucket"}}, getSize)
+	got, err := scanDataFolder(context.Background(), 0, 0, base, dataUsageCache{Info: dataUsageCacheInfo{Name: "bucket"}}, getSize, 0)
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -423,7 +423,7 @@ func TestDataUsageUpdatePrefix(t *testing.T) {
 	}
 	// Changed dir must be picked up in this many cycles.
 	for i := 0; i < dataUsageUpdateDirCycles; i++ {
-		got, err = scanDataFolder(context.Background(), 0, 0, base, got, getSize)
+		got, err = scanDataFolder(context.Background(), 0, 0, base, got, getSize, 0)
 		got.Info.NextCycle++
 		if err != nil {
 			t.Fatal(err)
@@ -575,7 +575,7 @@ func TestDataUsageCacheSerialize(t *testing.T) {
 		}
 		return
 	}
-	want, err := scanDataFolder(context.Background(), 0, 0, base, dataUsageCache{Info: dataUsageCacheInfo{Name: bucket}}, getSize)
+	want, err := scanDataFolder(context.Background(), 0, 0, base, dataUsageCache{Info: dataUsageCacheInfo{Name: bucket}}, getSize, 0)
 	if err != nil {
 		t.Fatal(err)
 	}

cmd/erasure-server-pool.go

@@ -531,7 +531,7 @@ func (z *erasureServerPools) StorageInfo(ctx context.Context) (StorageInfo, []er
 	return storageInfo, errs
 }
 
-func (z *erasureServerPools) NSScanner(ctx context.Context, bf *bloomFilter, updates chan<- DataUsageInfo, wantCycle uint32) error {
+func (z *erasureServerPools) NSScanner(ctx context.Context, bf *bloomFilter, updates chan<- DataUsageInfo, wantCycle uint32, healScanMode madmin.HealScanMode) error {
 	// Updates must be closed before we return.
 	defer close(updates)
 
@@ -576,7 +576,7 @@ func (z *erasureServerPools) NSScanner(ctx context.Context, bf *bloomFilter, upd
 					}
 				}()
 				// Start scanner. Blocks until done.
-				err := erObj.nsScanner(ctx, allBuckets, bf, wantCycle, updates)
+				err := erObj.nsScanner(ctx, allBuckets, bf, wantCycle, updates, healScanMode)
 				if err != nil {
 					logger.LogIf(ctx, err)
 					mu.Lock()

cmd/erasure.go

@@ -344,7 +344,7 @@ func (er erasureObjects) cleanupDeletedObjects(ctx context.Context) {
 
 // nsScanner will start scanning buckets and send updated totals as they are traversed.
 // Updates are sent on a regular basis and the caller *must* consume them.
-func (er erasureObjects) nsScanner(ctx context.Context, buckets []BucketInfo, bf *bloomFilter, wantCycle uint32, updates chan<- dataUsageCache) error {
+func (er erasureObjects) nsScanner(ctx context.Context, buckets []BucketInfo, bf *bloomFilter, wantCycle uint32, updates chan<- dataUsageCache, healScanMode madmin.HealScanMode) error {
 	if len(buckets) == 0 {
 		return nil
 	}
@@ -490,7 +490,7 @@ func (er erasureObjects) nsScanner(ctx context.Context, buckets []BucketInfo, bf
 				// Calc usage
 				before := cache.Info.LastUpdate
 				var err error
-				cache, err = disk.NSScanner(ctx, cache, updates)
+				cache, err = disk.NSScanner(ctx, cache, updates, healScanMode)
 				cache.Info.BloomFilter = nil
 				if err != nil {
 					if !cache.Info.LastUpdate.IsZero() && cache.Info.LastUpdate.After(before) {

cmd/fs-v1.go

@@ -235,7 +235,7 @@ func (fs *FSObjects) StorageInfo(ctx context.Context) (StorageInfo, []error) {
 }
 
 // NSScanner returns data usage stats of the current FS deployment
-func (fs *FSObjects) NSScanner(ctx context.Context, bf *bloomFilter, updates chan<- DataUsageInfo, wantCycle uint32) error {
+func (fs *FSObjects) NSScanner(ctx context.Context, bf *bloomFilter, updates chan<- DataUsageInfo, wantCycle uint32, _ madmin.HealScanMode) error {
 	defer close(updates)
 	// Load bucket totals
 	var totalCache dataUsageCache
@@ -396,7 +396,7 @@ func (fs *FSObjects) scanBucket(ctx context.Context, bucket string, cache dataUs
 		}
 
 		return sizeSummary{totalSize: fi.Size(), versions: 1}, nil
-	})
+	}, 0)
 
 	return cache, err
 }

cmd/gateway-unsupported.go

@@ -48,7 +48,7 @@ func (a GatewayUnsupported) LocalStorageInfo(ctx context.Context) (StorageInfo,
 }
 
 // NSScanner - scanner is not implemented for gateway
-func (a GatewayUnsupported) NSScanner(ctx context.Context, bf *bloomFilter, updates chan<- DataUsageInfo, wantCycle uint32) error {
+func (a GatewayUnsupported) NSScanner(ctx context.Context, bf *bloomFilter, updates chan<- DataUsageInfo, wantCycle uint32, scanMode madmin.HealScanMode) error {
 	logger.CriticalIf(ctx, errors.New("not implemented"))
 	return NotImplemented{}
 }

cmd/global-heal.go

@@ -44,8 +44,7 @@ func newBgHealSequence() *healSequence {
 
 	hs := madmin.HealOpts{
 		// Remove objects that do not have read-quorum
-		Remove:   healDeleteDangling,
-		ScanMode: globalHealConfig.ScanMode(),
+		Remove: healDeleteDangling,
 	}
 
 	return &healSequence{
@@ -165,7 +164,7 @@ func mustGetHealSequence(ctx context.Context) *healSequence {
 // healErasureSet lists and heals all objects in a specific erasure set
 func (er *erasureObjects) healErasureSet(ctx context.Context, buckets []string, tracker *healingTracker) error {
 	bgSeq := mustGetHealSequence(ctx)
-	scanMode := globalHealConfig.ScanMode()
+	scanMode := madmin.HealNormalScan
 
 	// Make sure to copy since `buckets slice`
 	// is modified in place by tracker.

cmd/mrf.go

@@ -185,7 +185,7 @@ func (m *mrfState) healRoutine() {
 	defer idler.Stop()
 
 	mrfHealingOpts := madmin.HealOpts{
-		ScanMode: globalHealConfig.ScanMode(),
+		ScanMode: madmin.HealNormalScan,
 		Remove:   healDeleteDangling,
 	}
 

cmd/naughty-disk_test.go

@@ -22,6 +22,8 @@ import (
 	"io"
 	"sync"
 	"time"
+
+	"github.com/minio/madmin-go"
 )
 
 // naughtyDisk wraps a POSIX disk and returns programmed errors
@@ -110,8 +112,8 @@ func (d *naughtyDisk) SetDiskID(id string) {
 	d.disk.SetDiskID(id)
 }
 
-func (d *naughtyDisk) NSScanner(ctx context.Context, cache dataUsageCache, updates chan<- dataUsageEntry) (info dataUsageCache, err error) {
-	return d.disk.NSScanner(ctx, cache, updates)
+func (d *naughtyDisk) NSScanner(ctx context.Context, cache dataUsageCache, updates chan<- dataUsageEntry, scanMode madmin.HealScanMode) (info dataUsageCache, err error) {
+	return d.disk.NSScanner(ctx, cache, updates, scanMode)
 }
 
 func (d *naughtyDisk) DiskInfo(ctx context.Context) (info DiskInfo, err error) {

cmd/object-api-interface.go

@@ -171,7 +171,7 @@ type ObjectLayer interface {
 
 	// Storage operations.
 	Shutdown(context.Context) error
-	NSScanner(ctx context.Context, bf *bloomFilter, updates chan<- DataUsageInfo, wantCycle uint32) error
+	NSScanner(ctx context.Context, bf *bloomFilter, updates chan<- DataUsageInfo, wantCycle uint32, scanMode madmin.HealScanMode) error
 	BackendInfo() madmin.BackendInfo
 	StorageInfo(ctx context.Context) (StorageInfo, []error)
 	LocalStorageInfo(ctx context.Context) (StorageInfo, []error)

cmd/storage-interface.go

@@ -21,6 +21,8 @@ import (
 	"context"
 	"io"
 	"time"
+
+	"github.com/minio/madmin-go"
 )
 
 // StorageAPI interface.
@@ -64,7 +66,7 @@ type StorageAPI interface {
 	// has never been replaced.
 	Healing() *healingTracker
 	DiskInfo(ctx context.Context) (info DiskInfo, err error)
-	NSScanner(ctx context.Context, cache dataUsageCache, updates chan<- dataUsageEntry) (dataUsageCache, error)
+	NSScanner(ctx context.Context, cache dataUsageCache, updates chan<- dataUsageEntry, scanMode madmin.HealScanMode) (dataUsageCache, error)
 
 	// Volume operations.
 	MakeVol(ctx context.Context, volume string) (err error)
@@ -142,7 +144,7 @@ func (p *unrecognizedDisk) Healing() *healingTracker {
 	return nil
 }
 
-func (p *unrecognizedDisk) NSScanner(ctx context.Context, cache dataUsageCache, updates chan<- dataUsageEntry) (dataUsageCache, error) {
+func (p *unrecognizedDisk) NSScanner(ctx context.Context, cache dataUsageCache, updates chan<- dataUsageEntry, scanMode madmin.HealScanMode) (dataUsageCache, error) {
 	return dataUsageCache{}, errDiskNotFound
 }