lsm/seaweedfs
1
0
Fork 0
mirror of https://github.com/seaweedfs/seaweedfs.git synced 2026-02-09 09:17:28 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

properly working local write buffer

Browse Source
This commit is contained in:
Chris Lu
2018-05-29 01:21:21 -07:00
parent c4b92e17d0
commit 4e3ea49cff
5 changed files with 54 additions and 91 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
weed/command/mount.go
View File

@@ -18,7 +18,7 @@ func init() {
mountOptions.dir = cmdMount.Flag.String("dir", ".", "mount weed filer to this directory")
mountOptions.collection = cmdMount.Flag.String("collection", "", "collection to create the files")
mountOptions.replication = cmdMount.Flag.String("replication", "000", "replication to create to files")
mountOptions.chunkSizeLimitMB = cmdMount.Flag.Int("chunkSizeLimitMB", 0, "if set, limit the chunk size in MB")
mountOptions.chunkSizeLimitMB = cmdMount.Flag.Int("chunkSizeLimitMB", 16, "local write buffer size, also chunk large files")
}
var cmdMount = &Command{

4
weed/command/mount_std.go
View File

@@ -19,6 +19,10 @@ func runMount(cmd *Command, args []string) bool {
fmt.Printf("Please specify the mount directory via \"-dir\"")
return false
}
if *mountOptions.chunkSizeLimitMB <= 0 {
fmt.Printf("Please specify a reasonable buffer size.")
return false
}
fuse.Unmount(*mountOptions.dir)

2
weed/filesys/dir.go
View File

@@ -124,7 +124,7 @@ func (dir *Dir) Create(ctx context.Context, req *fuse.CreateRequest,
file.isOpen = true
return file, &FileHandle{
f: file,
dirtyPages: &ContinuousDirtyPages{f: file},
dirtyPages: newDirtyPages(file),
RequestId: req.Header.ID,
NodeId: req.Header.Node,
Uid: req.Uid,

135
weed/filesys/dirty_page.go
View File

@@ -1,10 +1,8 @@
package filesys
import (
"sync"
"fmt"
"bytes"
"io"
"time"
"context"
@@ -13,98 +11,64 @@ import (
"github.com/chrislusf/seaweedfs/weed/glog"
)
type DirtyPage struct {
Offset int64
Data []byte
type ContinuousDirtyPages struct {
hasData bool
Offset int64
Size int64
Data []byte
f *File
}
type ContinuousDirtyPages struct {
sync.Mutex
pages []*DirtyPage
f *File
func newDirtyPages(file *File) *ContinuousDirtyPages {
return &ContinuousDirtyPages{
Data: make([]byte, file.wfs.chunkSizeLimit),
f: file,
}
}
func (pages *ContinuousDirtyPages) AddPage(ctx context.Context, offset int64, data []byte) (chunk *filer_pb.FileChunk, err error) {
pages.Lock()
defer pages.Unlock()
isPerfectOverwrite := false
isPerfectAppend := false
if len(pages.pages) > 0 {
lastPage := pages.pages[len(pages.pages)-1]
if lastPage.Offset+int64(len(lastPage.Data)) == offset {
// write continuous pages
glog.V(4).Infof("%s/%s append [%d,%d)", pages.f.dir.Path, pages.f.Name, offset, offset+int64(len(data)))
isPerfectAppend = true
}
if pages.pages[0].Offset == offset && pages.totalSize() == int64(len(data)) {
glog.V(4).Infof("%s/%s overwrite [%d,%d)", pages.f.dir.Path, pages.f.Name, offset, offset+int64(len(data)))
isPerfectOverwrite = true
}
} else {
glog.V(4).Infof("%s/%s append [%d,%d)", pages.f.dir.Path, pages.f.Name, offset, offset+int64(len(data)))
isPerfectAppend = true
if len(data) > len(pages.Data) {
// this is more than what we can hold.
panic("not prepared if buffer is smaller than each system write!")
}
isPerfectReplace := false
for _, page := range pages.pages {
if page.Offset == offset && len(page.Data) == len(data) {
// perfect replace
glog.V(4).Infof("%s/%s replace [%d,%d)", pages.f.dir.Path, pages.f.Name, offset, offset+int64(len(data)))
page.Data = data
isPerfectReplace = true
}
}
if offset < pages.Offset || offset >= pages.Offset+int64(len(pages.Data)) ||
pages.Offset+int64(len(pages.Data)) < offset+int64(len(data)) {
// if the data is out of range,
// or buffer is full if adding new data,
// flush current buffer and add new data
if isPerfectReplace {
return nil, nil
}
// println("offset", offset, "size", len(data), "existing offset", pages.Offset, "size", pages.Size)
if isPerfectAppend || isPerfectOverwrite {
if isPerfectAppend {
glog.V(4).Infof("%s/%s append2 [%d,%d)", pages.f.dir.Path, pages.f.Name, offset, offset+int64(len(data)))
pages.pages = append(pages.pages, &DirtyPage{
Offset: offset,
Data: data,
})
}
if isPerfectOverwrite {
glog.V(4).Infof("%s/%s overwrite2 [%d,%d)", pages.f.dir.Path, pages.f.Name, offset, offset+int64(len(data)))
pages.pages = []*DirtyPage{&DirtyPage{
Offset: offset,
Data: data,
}}
}
if pages.f.wfs.chunkSizeLimit > 0 && pages.totalSize() >= pages.f.wfs.chunkSizeLimit {
chunk, err = pages.saveToStorage(ctx)
pages.pages = nil
glog.V(4).Infof("%s/%s over size limit [%d,%d)", pages.f.dir.Path, pages.f.Name, chunk.Offset, chunk.Offset+int64(chunk.Size))
if chunk, err = pages.saveToStorage(ctx); err == nil {
if chunk != nil {
glog.V(4).Infof("%s/%s add save [%d,%d)", pages.f.dir.Path, pages.f.Name, chunk.Offset, chunk.Offset+int64(chunk.Size))
}
} else {
glog.V(0).Infof("%s/%s add save [%d,%d): %v", pages.f.dir.Path, pages.f.Name, chunk.Offset, chunk.Offset+int64(chunk.Size), err)
return
}
pages.Offset = offset
pages.Size = int64(len(data))
copy(pages.Data, data)
return
}
chunk, err = pages.saveToStorage(ctx)
glog.V(4).Infof("%s/%s saved [%d,%d)", pages.f.dir.Path, pages.f.Name, chunk.Offset, chunk.Offset+int64(chunk.Size))
pages.pages = []*DirtyPage{&DirtyPage{
Offset: offset,
Data: data,
}}
copy(pages.Data[offset-pages.Offset:], data)
pages.Size = max(pages.Size, offset+int64(len(data))-pages.Offset)
return
}
func (pages *ContinuousDirtyPages) FlushToStorage(ctx context.Context) (chunk *filer_pb.FileChunk, err error) {
pages.Lock()
defer pages.Unlock()
if pages.Size == 0 {
return nil, nil
}
if chunk, err = pages.saveToStorage(ctx); err == nil {
pages.pages = nil
pages.Size = 0
if chunk != nil {
glog.V(4).Infof("%s/%s flush [%d,%d)", pages.f.dir.Path, pages.f.Name, chunk.Offset, chunk.Offset+int64(chunk.Size))
}
@@ -112,16 +76,9 @@ func (pages *ContinuousDirtyPages) FlushToStorage(ctx context.Context) (chunk *f
return
}
func (pages *ContinuousDirtyPages) totalSize() (total int64) {
for _, page := range pages.pages {
total += int64(len(page.Data))
}
return
}
func (pages *ContinuousDirtyPages) saveToStorage(ctx context.Context) (*filer_pb.FileChunk, error) {
if len(pages.pages) == 0 {
if pages.Size == 0 {
return nil, nil
}
@@ -148,13 +105,8 @@ func (pages *ContinuousDirtyPages) saveToStorage(ctx context.Context) (*filer_pb
return nil, fmt.Errorf("filer assign volume: %v", err)
}
var readers []io.Reader
for _, page := range pages.pages {
readers = append(readers, bytes.NewReader(page.Data))
}
fileUrl := fmt.Sprintf("http://%s/%s", host, fileId)
bufReader := io.MultiReader(readers...)
bufReader := bytes.NewReader(pages.Data[:pages.Size])
uploadResult, err := operation.Upload(fileUrl, pages.f.Name, bufReader, false, "application/octet-stream", nil, "")
if err != nil {
glog.V(0).Infof("upload data %v to %s: %v", pages.f.Name, fileUrl, err)
@@ -167,9 +119,16 @@ func (pages *ContinuousDirtyPages) saveToStorage(ctx context.Context) (*filer_pb
return &filer_pb.FileChunk{
FileId: fileId,
Offset: pages.pages[0].Offset,
Size: uint64(pages.totalSize()),
Offset: pages.Offset,
Size: uint64(pages.Size),
Mtime: time.Now().UnixNano(),
}, nil
}
func max(x, y int64) int64 {
if x > y {
return x
}
return y
}

2
weed/filesys/file.go
View File

@@ -85,7 +85,7 @@ func (file *File) Open(ctx context.Context, req *fuse.OpenRequest, resp *fuse.Op
return &FileHandle{
f: file,
dirtyPages: &ContinuousDirtyPages{f: file},
dirtyPages: newDirtyPages(file),
RequestId: req.Header.ID,
NodeId: req.Header.Node,
Uid: req.Uid,