Avoid cache GC of writebacks before commit syncs (#13860)

Save part.1 for writebacks in a separate folder
and move it to cache dir atomically while saving
the cache metadata. This is to avoid GC mistaking
part.1 as orphaned cache entries and purging them.

This PR also fixes object size being overwritten during
retries for write-back mode.

cmd/disk-cache-backend.go

@@ -59,10 +59,13 @@ const (
 	cacheExpiryDays  = 90 * time.Hour * 24 // defaults to 90 days
 	// SSECacheEncrypted is the metadata key indicating that the object
 	// is a cache entry encrypted with cache KMS master key in globalCacheKMS.
-	SSECacheEncrypted               = "X-Minio-Internal-Encrypted-Cache"
-	cacheMultipartDir               = "multipart"
+	SSECacheEncrypted = "X-Minio-Internal-Encrypted-Cache"
+	cacheMultipartDir = "multipart"
+	cacheWritebackDir = "writeback"
+
 	cacheStaleUploadCleanupInterval = time.Hour * 24
 	cacheStaleUploadExpiry          = time.Hour * 24
+	cacheWBStaleUploadExpiry        = time.Hour * 24 * 7
 )
 
 // CacheChecksumInfoV1 - carries checksums of individual blocks on disk.
@@ -105,15 +108,15 @@ func (r *RangeInfo) Empty() bool {
 	return r.Range == "" && r.File == "" && r.Size == 0
 }
 
-func (m *cacheMeta) ToObjectInfo(bucket, object string) (o ObjectInfo) {
+func (m *cacheMeta) ToObjectInfo() (o ObjectInfo) {
 	if len(m.Meta) == 0 {
 		m.Meta = make(map[string]string)
 		m.Stat.ModTime = timeSentinel
 	}
 
 	o = ObjectInfo{
-		Bucket:            bucket,
-		Name:              object,
+		Bucket:            m.Bucket,
+		Name:              m.Object,
 		CacheStatus:       CacheHit,
 		CacheLookupStatus: CacheHit,
 	}
@@ -381,7 +384,7 @@ func (c *diskCache) purge(ctx context.Context) {
 		lastAtime := lastAtimeFn(meta.PartNumbers, pathJoin(c.dir, name))
 		// stat all cached file ranges.
 		cachedRngFiles := fiStatRangesFn(meta.Ranges, pathJoin(c.dir, name))
-		objInfo := meta.ToObjectInfo("", "")
+		objInfo := meta.ToObjectInfo()
 		// prevent gc from clearing un-synced commits. This metadata is present when
 		// cache writeback commit setting is enabled.
 		status, ok := objInfo.UserDefined[writeBackStatusHeader]
@@ -408,6 +411,9 @@ func (c *diskCache) purge(ctx context.Context) {
 		}
 
 		for fname, fi := range cachedRngFiles {
+			if fi == nil {
+				continue
+			}
 			if cc != nil {
 				if cc.isStale(objInfo.ModTime) {
 					removeAll(fname)
@@ -425,9 +431,11 @@ func (c *diskCache) purge(ctx context.Context) {
 		}
 		// clean up stale cache.json files for objects that never got cached but access count was maintained in cache.json
 		fi, err := os.Stat(pathJoin(cacheDir, cacheMetaJSONFile))
-		if err != nil || (fi.ModTime().Before(expiry) && len(cachedRngFiles) == 0) {
+		if err != nil || (fi != nil && fi.ModTime().Before(expiry) && len(cachedRngFiles) == 0) {
 			removeAll(cacheDir)
-			scorer.adjustSaveBytes(-fi.Size())
+			if fi != nil {
+				scorer.adjustSaveBytes(-fi.Size())
+			}
 			// Proceed to next file.
 			return nil
 		}
@@ -486,7 +494,7 @@ func (c *diskCache) Stat(ctx context.Context, bucket, object string) (oi ObjectI
 	if partial {
 		return oi, numHits, errFileNotFound
 	}
-	oi = meta.ToObjectInfo("", "")
+	oi = meta.ToObjectInfo()
 	oi.Bucket = bucket
 	oi.Name = object
 
@@ -521,7 +529,7 @@ func (c *diskCache) statRange(ctx context.Context, bucket, object string, rs *HT
 		return
 	}
 
-	oi = meta.ToObjectInfo("", "")
+	oi = meta.ToObjectInfo()
 	oi.Bucket = bucket
 	oi.Name = object
 	if !partial {
@@ -573,12 +581,16 @@ func (c *diskCache) statCache(ctx context.Context, cacheObjPath string) (meta *c
 	if _, err := os.Stat(pathJoin(cacheObjPath, cacheDataFile)); err == nil {
 		partial = false
 	}
+	if writebackInProgress(meta.Meta) {
+		partial = false
+	}
 	return meta, partial, meta.Hits, nil
 }
 
 // saves object metadata to disk cache
 // incHitsOnly is true if metadata update is incrementing only the hit counter
-func (c *diskCache) SaveMetadata(ctx context.Context, bucket, object string, meta map[string]string, actualSize int64, rs *HTTPRangeSpec, rsFileName string, incHitsOnly bool) error {
+// finalizeWB is true only if metadata update accompanied by moving part from temp location to cache dir.
+func (c *diskCache) SaveMetadata(ctx context.Context, bucket, object string, meta map[string]string, actualSize int64, rs *HTTPRangeSpec, rsFileName string, incHitsOnly, finalizeWB bool) error {
 	cachedPath := getCacheSHADir(c.dir, bucket, object)
 	cLock := c.NewNSLockFn(cachedPath)
 	lkctx, err := cLock.GetLock(ctx, globalOperationTimeout)
@@ -587,7 +599,18 @@ func (c *diskCache) SaveMetadata(ctx context.Context, bucket, object string, met
 	}
 	ctx = lkctx.Context()
 	defer cLock.Unlock(lkctx.Cancel)
-	return c.saveMetadata(ctx, bucket, object, meta, actualSize, rs, rsFileName, incHitsOnly)
+	if err = c.saveMetadata(ctx, bucket, object, meta, actualSize, rs, rsFileName, incHitsOnly); err != nil {
+		return err
+	}
+	// move part saved in writeback directory and cache.json atomically
+	if finalizeWB {
+		wbdir := getCacheWriteBackSHADir(c.dir, bucket, object)
+		if err = renameAll(pathJoin(wbdir, cacheDataFile), pathJoin(cachedPath, cacheDataFile)); err != nil {
+			return err
+		}
+		removeAll(wbdir) // cleanup writeback/shadir
+	}
+	return nil
 }
 
 // saves object metadata to disk cache
@@ -702,6 +725,11 @@ func getCacheSHADir(dir, bucket, object string) string {
 	return pathJoin(dir, getSHA256Hash([]byte(pathJoin(bucket, object))))
 }
 
+// returns temporary writeback cache location.
+func getCacheWriteBackSHADir(dir, bucket, object string) string {
+	return pathJoin(dir, minioMetaBucket, "writeback", getSHA256Hash([]byte(pathJoin(bucket, object))))
+}
+
 // Cache data to disk with bitrot checksum added for each block of 1MB
 func (c *diskCache) bitrotWriteToCache(cachePath, fileName string, reader io.Reader, size uint64) (int64, string, error) {
 	if err := os.MkdirAll(cachePath, 0777); err != nil {
@@ -846,6 +874,11 @@ func (c *diskCache) put(ctx context.Context, bucket, object string, data io.Read
 	if !c.diskSpaceAvailable(size) {
 		return oi, errDiskFull
 	}
+
+	if writeback {
+		cachePath = getCacheWriteBackSHADir(c.dir, bucket, object)
+	}
+
 	if err := os.MkdirAll(cachePath, 0777); err != nil {
 		return oi, err
 	}
@@ -1131,6 +1164,9 @@ func (c *diskCache) Get(ctx context.Context, bucket, object string, rs *HTTPRang
 		}()
 	} else {
 		go func() {
+			if writebackInProgress(objInfo.UserDefined) {
+				cacheObjPath = getCacheWriteBackSHADir(c.dir, bucket, object)
+			}
 			filePath := pathJoin(cacheObjPath, cacheFile)
 			err := c.bitrotReadFromCache(ctx, filePath, startOffset, length, pw)
 			if err != nil {
@@ -1199,7 +1235,7 @@ func (c *diskCache) scanCacheWritebackFailures(ctx context.Context) {
 			return nil
 		}
 
-		objInfo := meta.ToObjectInfo("", "")
+		objInfo := meta.ToObjectInfo()
 		status, ok := objInfo.UserDefined[writeBackStatusHeader]
 		if !ok || status == CommitComplete.String() {
 			return nil
@@ -1497,6 +1533,8 @@ func (c *diskCache) CompleteMultipartUpload(ctx context.Context, bucket, object,
 	}
 	uploadMeta.Stat.Size = objectSize
 	uploadMeta.Stat.ModTime = roi.ModTime
+	uploadMeta.Bucket = bucket
+	uploadMeta.Object = object
 	// if encrypted - make sure ETag updated
 
 	uploadMeta.Meta["etag"] = roi.ETag
@@ -1532,7 +1570,7 @@ func (c *diskCache) CompleteMultipartUpload(ctx context.Context, bucket, object,
 	}
 	renameAll(pathJoin(uploadIDDir, cacheMetaJSONFile), pathJoin(cachePath, cacheMetaJSONFile))
 	removeAll(uploadIDDir) // clean up any unused parts in the uploadIDDir
-	return uploadMeta.ToObjectInfo(bucket, object), nil
+	return uploadMeta.ToObjectInfo(), nil
 }
 
 func (c *diskCache) AbortUpload(bucket, object, uploadID string) (err error) {
@@ -1579,9 +1617,6 @@ func getMultipartCacheSHADir(dir, bucket, object string) string {
 
 // clean up stale cache multipart uploads according to cleanup interval.
 func (c *diskCache) cleanupStaleUploads(ctx context.Context) {
-	if !c.commitWritethrough {
-		return
-	}
 	timer := time.NewTimer(cacheStaleUploadCleanupInterval)
 	defer timer.Stop()
 	for {
@@ -1605,6 +1640,22 @@ func (c *diskCache) cleanupStaleUploads(ctx context.Context) {
 					return nil
 				})
 			})
+			// clean up of writeback folder where cache.json no longer exists in the main c.dir/<sha256(bucket,object> path
+			// and if past upload expiry window.
+			readDirFn(pathJoin(c.dir, minioMetaBucket, cacheWritebackDir), func(shaDir string, typ os.FileMode) error {
+				wbdir := pathJoin(c.dir, minioMetaBucket, cacheWritebackDir, shaDir)
+				cachedir := pathJoin(c.dir, shaDir)
+				if _, err := os.Stat(cachedir); os.IsNotExist(err) {
+					fi, err := os.Stat(wbdir)
+					if err != nil {
+						return nil
+					}
+					if now.Sub(fi.ModTime()) > cacheWBStaleUploadExpiry {
+						return removeAll(wbdir)
+					}
+				}
+				return nil
+			})
 		}
 	}
 }

cmd/disk-cache-utils.go

@@ -585,8 +585,8 @@ func cacheMultiWriter(w1 io.Writer, w2 *io.PipeWriter) io.Writer {
 	return &multiWriter{backendWriter: w1, cacheWriter: w2}
 }
 
-// skipETagVerification returns true if writeback commit is not complete
-func skipETagVerification(m map[string]string) bool {
+// writebackInProgress returns true if writeback commit is not complete
+func writebackInProgress(m map[string]string) bool {
 	if v, ok := m[writeBackStatusHeader]; ok {
 		switch cacheCommitStatus(v) {
 		case CommitPending, CommitFailed:

cmd/disk-cache.go

@@ -132,7 +132,7 @@ type cacheObjects struct {
 
 func (c *cacheObjects) incHitsToMeta(ctx context.Context, dcache *diskCache, bucket, object string, size int64, eTag string, rs *HTTPRangeSpec) error {
 	metadata := map[string]string{"etag": eTag}
-	return dcache.SaveMetadata(ctx, bucket, object, metadata, size, rs, "", true)
+	return dcache.SaveMetadata(ctx, bucket, object, metadata, size, rs, "", true, false)
 }
 
 // Backend metadata could have changed through server side copy - reset cache metadata if that is the case
@@ -159,7 +159,7 @@ func (c *cacheObjects) updateMetadataIfChanged(ctx context.Context, dcache *disk
 		bkObjectInfo.ETag != cacheObjInfo.ETag ||
 		bkObjectInfo.ContentType != cacheObjInfo.ContentType ||
 		!bkObjectInfo.Expires.Equal(cacheObjInfo.Expires) {
-		return dcache.SaveMetadata(ctx, bucket, object, getMetadata(bkObjectInfo), bkObjectInfo.Size, nil, "", false)
+		return dcache.SaveMetadata(ctx, bucket, object, getMetadata(bkObjectInfo), bkObjectInfo.Size, nil, "", false, false)
 	}
 	return c.incHitsToMeta(ctx, dcache, bucket, object, cacheObjInfo.Size, cacheObjInfo.ETag, rs)
 }
@@ -277,7 +277,7 @@ func (c *cacheObjects) GetObjectNInfo(ctx context.Context, bucket, object string
 			return bReader, err
 		}
 		// serve cached content without ETag verification if writeback commit is not yet complete
-		if skipETagVerification(cacheReader.ObjInfo.UserDefined) {
+		if writebackInProgress(cacheReader.ObjInfo.UserDefined) {
 			return cacheReader, nil
 		}
 	}
@@ -427,7 +427,7 @@ func (c *cacheObjects) GetObjectInfo(ctx context.Context, bucket, object string,
 			return cachedObjInfo, nil
 		}
 		// serve cache metadata without ETag verification if writeback commit is not yet complete
-		if skipETagVerification(cachedObjInfo.UserDefined) {
+		if writebackInProgress(cachedObjInfo.UserDefined) {
 			return cachedObjInfo, nil
 		}
 	}
@@ -794,6 +794,7 @@ func (c *cacheObjects) uploadObject(ctx context.Context, oi ObjectInfo) {
 	opts.UserDefined[xhttp.ContentMD5] = oi.UserDefined["content-md5"]
 	objInfo, err := c.InnerPutObjectFn(ctx, oi.Bucket, oi.Name, NewPutObjReader(hashReader), opts)
 	wbCommitStatus := CommitComplete
+	size := objInfo.Size
 	if err != nil {
 		wbCommitStatus = CommitFailed
 	}
@@ -804,12 +805,13 @@ func (c *cacheObjects) uploadObject(ctx context.Context, oi ObjectInfo) {
 		retryCnt, _ = strconv.Atoi(meta[writeBackRetryHeader])
 		retryCnt++
 		meta[writeBackRetryHeader] = strconv.Itoa(retryCnt)
+		size = cReader.ObjInfo.Size
 	} else {
 		delete(meta, writeBackRetryHeader)
 	}
 	meta[writeBackStatusHeader] = wbCommitStatus.String()
 	meta["etag"] = oi.ETag
-	dcache.SaveMetadata(ctx, oi.Bucket, oi.Name, meta, objInfo.Size, nil, "", false)
+	dcache.SaveMetadata(ctx, oi.Bucket, oi.Name, meta, size, nil, "", false, wbCommitStatus == CommitComplete)
 	if retryCnt > 0 {
 		// slow down retries
 		time.Sleep(time.Second * time.Duration(retryCnt%10+1))

cmd/disk-cache_test.go

@@ -25,7 +25,7 @@ import (
 // Tests ToObjectInfo function.
 func TestCacheMetadataObjInfo(t *testing.T) {
 	m := cacheMeta{Meta: nil}
-	objInfo := m.ToObjectInfo("testbucket", "testobject")
+	objInfo := m.ToObjectInfo()
 	if objInfo.Size != 0 {
 		t.Fatal("Unexpected object info value for Size", objInfo.Size)
 	}