Remove locks on usage cache (#16786)

cmd/data-usage-cache.go

@@ -23,6 +23,7 @@ import (
 	"errors"
 	"fmt"
 	"io"
+	"math/rand"
 	"net/http"
 	"path"
 	"path/filepath"
@@ -847,37 +848,54 @@ type objectIO interface {
 
 // load the cache content with name from minioMetaBackgroundOpsBucket.
 // Only backend errors are returned as errors.
+// The loader is optimistic and has no locking, but tries 5 times before giving up.
 // If the object is not found or unable to deserialize d is cleared and nil error is returned.
 func (d *dataUsageCache) load(ctx context.Context, store objectIO, name string) error {
 	// Abandon if more than 5 minutes, so we don't hold up scanner.
 	ctx, cancel := context.WithTimeout(ctx, 5*time.Minute)
 	defer cancel()
 
-	r, err := store.GetObjectNInfo(ctx, dataUsageBucket, name, nil, http.Header{}, readLock, ObjectOptions{})
+	// Caches are read+written without locks,
-	if err != nil {
+	retries := 0
-		switch err.(type) {
+	for retries < 5 {
-		case ObjectNotFound:
+		r, err := store.GetObjectNInfo(ctx, dataUsageBucket, name, nil, http.Header{}, noLock, ObjectOptions{NoLock: true})
-		case BucketNotFound:
+		if err != nil {
-		case InsufficientReadQuorum:
+			switch err.(type) {
-		case StorageErr:
+			case ObjectNotFound, BucketNotFound:
-		default:
+			case InsufficientReadQuorum, StorageErr:
-			return toObjectErr(err, dataUsageBucket, name)
+				retries++
+				time.Sleep(time.Duration(rand.Int63n(int64(time.Second))))
+				continue
+			default:
+				return toObjectErr(err, dataUsageBucket, name)
+			}
+			*d = dataUsageCache{}
+			return nil
 		}
-		*d = dataUsageCache{}
+		if err := d.deserialize(r); err != nil {
+			r.Close()
+			retries++
+			time.Sleep(time.Duration(rand.Int63n(int64(time.Second))))
+			continue
+		}
+		r.Close()
 		return nil
 	}
-	defer r.Close()
+	*d = dataUsageCache{}
-	if err := d.deserialize(r); err != nil {
-		*d = dataUsageCache{}
-		logger.LogOnceIf(ctx, err, err.Error())
-	}
 	return nil
 }
 
+// Maximum running concurrent saves on server.
+var maxConcurrentScannerSaves = make(chan struct{}, 4)
+
 // save the content of the cache to minioMetaBackgroundOpsBucket with the provided name.
+// Note that no locking is done when saving.
 func (d *dataUsageCache) save(ctx context.Context, store objectIO, name string) error {
 	var r io.Reader
-
+	maxConcurrentScannerSaves <- struct{}{}
+	defer func() {
+		<-maxConcurrentScannerSaves
+	}()
 	// If big, do streaming...
 	size := int64(-1)
 	if len(d.Cache) > 10000 {
@@ -909,7 +927,7 @@ func (d *dataUsageCache) save(ctx context.Context, store objectIO, name string)
 		dataUsageBucket,
 		name,
 		NewPutObjReader(hr),
-		ObjectOptions{})
+		ObjectOptions{NoLock: true})
 	if isErrBucketNotFound(err) {
 		return nil
 	}

cmd/erasure.go

@@ -449,6 +449,7 @@ func (er erasureObjects) nsScanner(ctx context.Context, buckets []BucketInfo, wa
 	// Start one scanner per disk
 	var wg sync.WaitGroup
 	wg.Add(len(disks))
+
 	for i := range disks {
 		go func(i int) {
 			defer wg.Done()
@@ -518,7 +519,6 @@ func (er erasureObjects) nsScanner(ctx context.Context, buckets []BucketInfo, wa
 				if r := cache.root(); r != nil {
 					root = cache.flatten(*r)
 				}
-				t := time.Now()
 				select {
 				case <-ctx.Done():
 					return
@@ -528,9 +528,7 @@ func (er erasureObjects) nsScanner(ctx context.Context, buckets []BucketInfo, wa
 					Entry:  root,
 				}:
 				}
-				// We want to avoid synchronizing up all writes in case
+
-				// the results are piled up.
-				time.Sleep(time.Duration(float64(time.Since(t)) * rand.Float64()))
 				// Save cache
 				logger.LogIf(ctx, cache.save(ctx, er, cacheName))
 			}