Chris Lu
5 years ago
9 changed files with 378 additions and 35 deletions
-
10weed/command/mount.go
-
6weed/command/mount_std.go
-
7weed/command/webdav.go
-
31weed/filesys/wfs.go
-
9weed/server/webdav_server.go
-
111weed/util/chunk_cache/chunk_cache.go
-
36weed/util/chunk_cache/chunk_cache_in_memory.go
-
145weed/util/chunk_cache/chunk_cache_on_disk.go
-
58weed/util/chunk_cache/chunk_cache_on_disk_test.go
@ -1,36 +1,115 @@ |
|||
package chunk_cache |
|||
|
|||
import ( |
|||
"time" |
|||
"fmt" |
|||
"path" |
|||
"sort" |
|||
"sync" |
|||
|
|||
"github.com/karlseguin/ccache" |
|||
"github.com/chrislusf/seaweedfs/weed/glog" |
|||
"github.com/chrislusf/seaweedfs/weed/storage" |
|||
"github.com/chrislusf/seaweedfs/weed/storage/needle" |
|||
) |
|||
|
|||
// a global cache for recently accessed file chunks
|
|||
type ChunkCache struct { |
|||
cache *ccache.Cache |
|||
memCache *ChunkCacheInMemory |
|||
diskCaches []*ChunkCacheVolume |
|||
sync.RWMutex |
|||
} |
|||
|
|||
func NewChunkCache(maxEntries int64) *ChunkCache { |
|||
pruneCount := maxEntries >> 3 |
|||
if pruneCount <= 0 { |
|||
pruneCount = 500 |
|||
func NewChunkCache(maxEntries int64, dir string, diskSizeMB int64, segmentCount int) *ChunkCache { |
|||
c := &ChunkCache{ |
|||
memCache: NewChunkCacheInMemory(maxEntries), |
|||
} |
|||
return &ChunkCache{ |
|||
cache: ccache.New(ccache.Configure().MaxSize(maxEntries).ItemsToPrune(uint32(pruneCount))), |
|||
|
|||
volumeCount, volumeSize := int(diskSizeMB/30000), int64(30000) |
|||
if volumeCount < segmentCount { |
|||
volumeCount, volumeSize = segmentCount, diskSizeMB/int64(segmentCount) |
|||
} |
|||
|
|||
for i := 0; i < volumeCount; i++ { |
|||
fileName := path.Join(dir, fmt.Sprintf("cache_%d", i)) |
|||
diskCache, err := LoadOrCreateChunkCacheVolume(fileName, volumeSize*1024*1024) |
|||
if err != nil { |
|||
glog.Errorf("failed to add cache %s : %v", fileName, err) |
|||
} else { |
|||
c.diskCaches = append(c.diskCaches, diskCache) |
|||
} |
|||
} |
|||
|
|||
// keep newest cache to the front
|
|||
sort.Slice(c.diskCaches, func(i, j int) bool { |
|||
return c.diskCaches[i].lastModTime.After(c.diskCaches[j].lastModTime) |
|||
}) |
|||
|
|||
return c |
|||
} |
|||
|
|||
func (c *ChunkCache) GetChunk(fileId string) []byte { |
|||
item := c.cache.Get(fileId) |
|||
if item == nil { |
|||
func (c *ChunkCache) GetChunk(fileId string) (data []byte) { |
|||
c.RLock() |
|||
defer c.RUnlock() |
|||
|
|||
if data = c.memCache.GetChunk(fileId); data != nil { |
|||
return data |
|||
} |
|||
|
|||
fid, err := needle.ParseFileIdFromString(fileId) |
|||
if err != nil { |
|||
glog.Errorf("failed to parse file id %s", fileId) |
|||
return nil |
|||
} |
|||
data := item.Value().([]byte) |
|||
item.Extend(time.Hour) |
|||
return data |
|||
for _, diskCache := range c.diskCaches { |
|||
data, err = diskCache.GetNeedle(fid.Key) |
|||
if err == storage.ErrorNotFound { |
|||
continue |
|||
} |
|||
if err != nil { |
|||
glog.Errorf("failed to read cache file %s id %s", diskCache.fileName, fileId) |
|||
continue |
|||
} |
|||
if len(data) != 0 { |
|||
return |
|||
} |
|||
} |
|||
return nil |
|||
} |
|||
|
|||
func (c *ChunkCache) SetChunk(fileId string, data []byte) { |
|||
c.cache.Set(fileId, data, time.Hour) |
|||
c.Lock() |
|||
defer c.Unlock() |
|||
|
|||
c.memCache.SetChunk(fileId, data) |
|||
|
|||
if len(c.diskCaches) == 0 { |
|||
return |
|||
} |
|||
|
|||
if c.diskCaches[0].fileSize+int64(len(data)) > c.diskCaches[0].sizeLimit { |
|||
t, resetErr := c.diskCaches[len(c.diskCaches)-1].Reset() |
|||
if resetErr != nil { |
|||
glog.Errorf("failed to reset cache file %s", c.diskCaches[len(c.diskCaches)-1].fileName) |
|||
return |
|||
} |
|||
for i := len(c.diskCaches) - 1; i > 0; i-- { |
|||
c.diskCaches[i] = c.diskCaches[i-1] |
|||
} |
|||
c.diskCaches[0] = t |
|||
} |
|||
|
|||
fid, err := needle.ParseFileIdFromString(fileId) |
|||
if err != nil { |
|||
glog.Errorf("failed to parse file id %s", fileId) |
|||
return |
|||
} |
|||
c.diskCaches[0].WriteNeedle(fid.Key, data) |
|||
|
|||
} |
|||
|
|||
func (c *ChunkCache) Shutdown() { |
|||
c.Lock() |
|||
defer c.Unlock() |
|||
for _, diskCache := range c.diskCaches { |
|||
diskCache.Shutdown() |
|||
} |
|||
} |
@ -0,0 +1,36 @@ |
|||
package chunk_cache |
|||
|
|||
import ( |
|||
"time" |
|||
|
|||
"github.com/karlseguin/ccache" |
|||
) |
|||
|
|||
// a global cache for recently accessed file chunks
|
|||
type ChunkCacheInMemory struct { |
|||
cache *ccache.Cache |
|||
} |
|||
|
|||
func NewChunkCacheInMemory(maxEntries int64) *ChunkCacheInMemory { |
|||
pruneCount := maxEntries >> 3 |
|||
if pruneCount <= 0 { |
|||
pruneCount = 500 |
|||
} |
|||
return &ChunkCacheInMemory{ |
|||
cache: ccache.New(ccache.Configure().MaxSize(maxEntries).ItemsToPrune(uint32(pruneCount))), |
|||
} |
|||
} |
|||
|
|||
func (c *ChunkCacheInMemory) GetChunk(fileId string) []byte { |
|||
item := c.cache.Get(fileId) |
|||
if item == nil { |
|||
return nil |
|||
} |
|||
data := item.Value().([]byte) |
|||
item.Extend(time.Hour) |
|||
return data |
|||
} |
|||
|
|||
func (c *ChunkCacheInMemory) SetChunk(fileId string, data []byte) { |
|||
c.cache.Set(fileId, data, time.Hour) |
|||
} |
@ -0,0 +1,145 @@ |
|||
package chunk_cache |
|||
|
|||
import ( |
|||
"fmt" |
|||
"os" |
|||
"time" |
|||
|
|||
"github.com/syndtr/goleveldb/leveldb/opt" |
|||
|
|||
"github.com/chrislusf/seaweedfs/weed/glog" |
|||
"github.com/chrislusf/seaweedfs/weed/storage" |
|||
"github.com/chrislusf/seaweedfs/weed/storage/backend" |
|||
"github.com/chrislusf/seaweedfs/weed/storage/types" |
|||
"github.com/chrislusf/seaweedfs/weed/util" |
|||
) |
|||
|
|||
// This implements an on disk cache
|
|||
// The entries are an FIFO with a size limit
|
|||
|
|||
type ChunkCacheVolume struct { |
|||
DataBackend backend.BackendStorageFile |
|||
nm storage.NeedleMapper |
|||
fileName string |
|||
smallBuffer []byte |
|||
sizeLimit int64 |
|||
lastModTime time.Time |
|||
fileSize int64 |
|||
} |
|||
|
|||
func LoadOrCreateChunkCacheVolume(fileName string, preallocate int64) (*ChunkCacheVolume, error) { |
|||
|
|||
v := &ChunkCacheVolume{ |
|||
smallBuffer: make([]byte, types.NeedlePaddingSize), |
|||
fileName: fileName, |
|||
sizeLimit: preallocate, |
|||
} |
|||
|
|||
var err error |
|||
|
|||
if exists, canRead, canWrite, modTime, fileSize := util.CheckFile(v.fileName + ".dat"); exists { |
|||
if !canRead { |
|||
return nil, fmt.Errorf("cannot read cache file %s.dat", v.fileName) |
|||
} |
|||
if !canWrite { |
|||
return nil, fmt.Errorf("cannot write cache file %s.dat", v.fileName) |
|||
} |
|||
if dataFile, err := os.OpenFile(v.fileName+".dat", os.O_RDWR|os.O_CREATE, 0644); err != nil { |
|||
return nil, fmt.Errorf("cannot create cache file %s.dat: %v", v.fileName, err) |
|||
} else { |
|||
v.DataBackend = backend.NewDiskFile(dataFile) |
|||
v.lastModTime = modTime |
|||
v.fileSize = fileSize |
|||
} |
|||
} else { |
|||
if v.DataBackend, err = backend.CreateVolumeFile(v.fileName+".dat", preallocate, 0); err != nil { |
|||
return nil, fmt.Errorf("cannot create cache file %s.dat: %v", v.fileName, err) |
|||
} |
|||
v.lastModTime = time.Now() |
|||
} |
|||
|
|||
var indexFile *os.File |
|||
if indexFile, err = os.OpenFile(v.fileName+".idx", os.O_RDWR|os.O_CREATE, 0644); err != nil { |
|||
return nil, fmt.Errorf("cannot write cache index %s.idx: %v", v.fileName, err) |
|||
} |
|||
|
|||
glog.V(0).Infoln("loading leveldb", v.fileName+".ldb") |
|||
opts := &opt.Options{ |
|||
BlockCacheCapacity: 2 * 1024 * 1024, // default value is 8MiB
|
|||
WriteBuffer: 1 * 1024 * 1024, // default value is 4MiB
|
|||
CompactionTableSizeMultiplier: 10, // default value is 1
|
|||
} |
|||
if v.nm, err = storage.NewLevelDbNeedleMap(v.fileName+".ldb", indexFile, opts); err != nil { |
|||
return nil, fmt.Errorf("loading leveldb %s error: %v", v.fileName+".ldb", err) |
|||
} |
|||
|
|||
return v, nil |
|||
|
|||
} |
|||
|
|||
func (v *ChunkCacheVolume) Shutdown() { |
|||
if v.DataBackend != nil { |
|||
v.DataBackend.Close() |
|||
v.DataBackend = nil |
|||
} |
|||
if v.nm != nil { |
|||
v.nm.Close() |
|||
v.nm = nil |
|||
} |
|||
} |
|||
|
|||
func (v *ChunkCacheVolume) destroy() { |
|||
v.Shutdown() |
|||
os.Remove(v.fileName + ".dat") |
|||
os.Remove(v.fileName + ".idx") |
|||
os.RemoveAll(v.fileName + ".ldb") |
|||
} |
|||
|
|||
func (v *ChunkCacheVolume) Reset() (*ChunkCacheVolume, error) { |
|||
v.destroy() |
|||
return LoadOrCreateChunkCacheVolume(v.fileName, v.sizeLimit) |
|||
} |
|||
|
|||
func (v *ChunkCacheVolume) GetNeedle(key types.NeedleId) ([]byte, error) { |
|||
|
|||
nv, ok := v.nm.Get(key) |
|||
if !ok { |
|||
return nil, storage.ErrorNotFound |
|||
} |
|||
data := make([]byte, nv.Size) |
|||
if readSize, readErr := v.DataBackend.ReadAt(data, nv.Offset.ToAcutalOffset()); readErr != nil { |
|||
return nil, fmt.Errorf("read %s.dat [%d,%d): %v", |
|||
v.fileName, nv.Offset.ToAcutalOffset(), nv.Offset.ToAcutalOffset()+int64(nv.Size), readErr) |
|||
} else { |
|||
if readSize != int(nv.Size) { |
|||
return nil, fmt.Errorf("read %d, expected %d", readSize, nv.Size) |
|||
} |
|||
} |
|||
|
|||
return data, nil |
|||
} |
|||
|
|||
func (v *ChunkCacheVolume) WriteNeedle(key types.NeedleId, data []byte) error { |
|||
|
|||
offset := v.fileSize |
|||
|
|||
written, err := v.DataBackend.WriteAt(data, offset) |
|||
if err != nil { |
|||
return err |
|||
} else if written != len(data) { |
|||
return fmt.Errorf("partial written %d, expected %d", written, len(data)) |
|||
} |
|||
|
|||
v.fileSize += int64(written) |
|||
extraSize := written % types.NeedlePaddingSize |
|||
if extraSize != 0 { |
|||
v.DataBackend.WriteAt(v.smallBuffer[:types.NeedlePaddingSize-extraSize], offset+int64(written)) |
|||
v.fileSize += int64(types.NeedlePaddingSize - extraSize) |
|||
} |
|||
|
|||
if err := v.nm.Put(key, types.ToOffset(offset), uint32(len(data))); err != nil { |
|||
glog.V(4).Infof("failed to save in needle map %d: %v", key, err) |
|||
} |
|||
|
|||
return nil |
|||
} |
@ -0,0 +1,58 @@ |
|||
package chunk_cache |
|||
|
|||
import ( |
|||
"bytes" |
|||
"fmt" |
|||
"io/ioutil" |
|||
"math/rand" |
|||
"os" |
|||
"testing" |
|||
) |
|||
|
|||
func TestOnDisk(t *testing.T) { |
|||
|
|||
tmpDir, _ := ioutil.TempDir("", "c") |
|||
defer os.RemoveAll(tmpDir) |
|||
|
|||
totalDiskSizeMb := int64(6) |
|||
segmentCount := 2 |
|||
|
|||
cache := NewChunkCache(0, tmpDir, totalDiskSizeMb, segmentCount) |
|||
|
|||
writeCount := 5 |
|||
type test_data struct { |
|||
data []byte |
|||
fileId string |
|||
} |
|||
testData := make([]*test_data, writeCount) |
|||
for i:=0;i<writeCount;i++{ |
|||
buff := make([]byte, 1024*1024) |
|||
rand.Read(buff) |
|||
testData[i] = &test_data{ |
|||
data: buff, |
|||
fileId: fmt.Sprintf("1,%daabbccdd", i+1), |
|||
} |
|||
cache.SetChunk(testData[i].fileId, testData[i].data) |
|||
} |
|||
|
|||
for i:=0;i<writeCount;i++{ |
|||
data := cache.GetChunk(testData[i].fileId) |
|||
if bytes.Compare(data, testData[i].data) != 0 { |
|||
t.Errorf("failed to write to and read from cache: %d", i) |
|||
} |
|||
} |
|||
|
|||
cache.Shutdown() |
|||
|
|||
cache = NewChunkCache(0, tmpDir, totalDiskSizeMb, segmentCount) |
|||
|
|||
for i:=0;i<writeCount;i++{ |
|||
data := cache.GetChunk(testData[i].fileId) |
|||
if bytes.Compare(data, testData[i].data) != 0 { |
|||
t.Errorf("failed to write to and read from cache: %d", i) |
|||
} |
|||
} |
|||
|
|||
cache.Shutdown() |
|||
|
|||
} |
Write
Preview
Loading…
Cancel
Save
Reference in new issue