Feature limit caching to prescribed number of bytes per file (#6009)

* feature: we can check if a fileId is already in the cache

We using this to protect cache from adding the same needle to
the cache over and over.

* fuse mount: Do not start dowloader if needle is already in the cache

* added maxFilePartSizeInCache property to ChunkCache

If file very large only first maxFilePartSizeInCache bytes
are going to be put to the cache (subject to the needle size
constrains).

* feature: for large files put in cache no more than prescribed number of bytes

Before this patch only the first needle of a large file was intended for
caching. This patch uses maximum prescribed amount of bytes to be put in
cache. This allows to bypass default 2MB maximum for a file part stored
in the cache.

* added dummy mock methods to satisfy interfaces of ChunkCache

3 files changed, 15 insertions, 2 deletions

diff --git a/weed/filer/reader_at.go b/weed/filer/reader_at.go
index d475e6e11..b70942edc 100644
--- a/weed/filer/reader_at.go
+++ b/weed/filer/reader_at.go
@@ -199,7 +199,7 @@ func (c *ChunkReadAt) readChunkSliceAt(buffer []byte, chunkView *ChunkView, next
 		return fetchChunkRange(buffer, c.readerCache.lookupFileIdFn, chunkView.FileId, chunkView.CipherKey, chunkView.IsGzipped, int64(offset))
 	}
 
-	n, err = c.readerCache.ReadChunkAt(buffer, chunkView.FileId, chunkView.CipherKey, chunkView.IsGzipped, int64(offset), int(chunkView.ChunkSize), chunkView.ViewOffset == 0)
+	n, err = c.readerCache.ReadChunkAt(buffer, chunkView.FileId, chunkView.CipherKey, chunkView.IsGzipped, int64(offset), int(chunkView.ChunkSize), (uint64(chunkView.ViewOffset)+chunkView.ChunkSize) <= c.readerCache.chunkCache.GetMaxFilePartSizeInCache())
 	if c.lastChunkFid != chunkView.FileId {
 		if chunkView.OffsetInChunk == 0 { // start of a new chunk
 			if c.lastChunkFid != "" {
diff --git a/weed/filer/reader_at_test.go b/weed/filer/reader_at_test.go
index 8bc383184..0d95d1aad 100644
--- a/weed/filer/reader_at_test.go
+++ b/weed/filer/reader_at_test.go
@@ -31,6 +31,14 @@ func (m *mockChunkCache) ReadChunkAt(data []byte, fileId string, offset uint64)
 func (m *mockChunkCache) SetChunk(fileId string, data []byte) {
 }
 
+func (m *mockChunkCache) GetMaxFilePartSizeInCache() (uint64) {
+	return 0
+}
+
+func (m *mockChunkCache) IsInCache(fileId string, lockNeeded bool) (answer bool) {
+	return false
+}
+
 func TestReaderAt(t *testing.T) {
 
 	visibles := NewIntervalList[*VisibleInterval]()
diff --git a/weed/filer/reader_cache.go b/weed/filer/reader_cache.go
index a3df8e0bd..716e796c9 100644
--- a/weed/filer/reader_cache.go
+++ b/weed/filer/reader_cache.go
@@ -6,6 +6,7 @@ import (
 	"sync/atomic"
 	"time"
 
+	"github.com/seaweedfs/seaweedfs/weed/glog"
 	"github.com/seaweedfs/seaweedfs/weed/util/chunk_cache"
 	util_http "github.com/seaweedfs/seaweedfs/weed/util/http"
 	"github.com/seaweedfs/seaweedfs/weed/util/mem"
@@ -61,6 +62,10 @@ func (rc *ReaderCache) MaybeCache(chunkViews *Interval[*ChunkView]) {
 		if _, found := rc.downloaders[chunkView.FileId]; found {
 			continue
 		}
+		if rc.chunkCache.IsInCache(chunkView.FileId, true) {
+			glog.V(4).Infof("%s is in cache", chunkView.FileId)
+			continue
+		}
 
 		if len(rc.downloaders) >= rc.limit {
 			// abort when slots are filled
@@ -69,7 +74,7 @@ func (rc *ReaderCache) MaybeCache(chunkViews *Interval[*ChunkView]) {
 
 		// glog.V(4).Infof("prefetch %s offset %d", chunkView.FileId, chunkView.ViewOffset)
 		// cache this chunk if not yet
-		cacher := newSingleChunkCacher(rc, chunkView.FileId, chunkView.CipherKey, chunkView.IsGzipped, int(chunkView.ChunkSize), chunkView.ViewOffset == 0)
+		cacher := newSingleChunkCacher(rc, chunkView.FileId, chunkView.CipherKey, chunkView.IsGzipped, int(chunkView.ChunkSize), (uint64(chunkView.ViewOffset)+chunkView.ChunkSize) <= rc.chunkCache.GetMaxFilePartSizeInCache())
 		go cacher.startCaching()
 		<-cacher.cacheStartedCh
 		rc.downloaders[chunkView.FileId] = cacher