chunked file works now
This commit is contained in:
152
weed/filesys/page_writer/chunked_file_writer.go
Normal file
152
weed/filesys/page_writer/chunked_file_writer.go
Normal file
@@ -0,0 +1,152 @@
|
||||
package page_writer
|
||||
|
||||
import (
|
||||
"github.com/chrislusf/seaweedfs/weed/glog"
|
||||
"io"
|
||||
"os"
|
||||
"sync"
|
||||
)
|
||||
|
||||
// ChunkedFileWriter assumes the write requests will come in within chunks
|
||||
type ChunkedFileWriter struct {
|
||||
dir string
|
||||
file *os.File
|
||||
logicToActualChunkIndex map[int]int
|
||||
chunkUsages []*PageChunkWrittenIntervalList
|
||||
ChunkSize int64
|
||||
sync.Mutex
|
||||
}
|
||||
|
||||
var _ = io.WriterAt(&ChunkedFileWriter{})
|
||||
|
||||
func NewChunkedFileWriter(dir string, chunkSize int64) *ChunkedFileWriter {
|
||||
return &ChunkedFileWriter{
|
||||
dir: dir,
|
||||
file: nil,
|
||||
logicToActualChunkIndex: make(map[int]int),
|
||||
ChunkSize: chunkSize,
|
||||
}
|
||||
}
|
||||
|
||||
func (cw *ChunkedFileWriter) WriteAt(p []byte, off int64) (n int, err error) {
|
||||
cw.Lock()
|
||||
defer cw.Unlock()
|
||||
|
||||
if cw.file == nil {
|
||||
cw.file, err = os.CreateTemp(cw.dir, "")
|
||||
if err != nil {
|
||||
glog.Errorf("create temp file: %v", err)
|
||||
return
|
||||
}
|
||||
}
|
||||
|
||||
actualOffset, chunkUsage := cw.toActualWriteOffset(off)
|
||||
n, err = cw.file.WriteAt(p, actualOffset)
|
||||
if err == nil {
|
||||
startOffset := off % cw.ChunkSize
|
||||
chunkUsage.MarkWritten(startOffset, startOffset+int64(n))
|
||||
}
|
||||
return
|
||||
}
|
||||
|
||||
func (cw *ChunkedFileWriter) ReadDataAt(p []byte, off int64) (maxStop int64) {
|
||||
cw.Lock()
|
||||
defer cw.Unlock()
|
||||
|
||||
if cw.file == nil {
|
||||
return
|
||||
}
|
||||
|
||||
logicChunkIndex := off / cw.ChunkSize
|
||||
actualChunkIndex, chunkUsage := cw.toActualReadOffset(off)
|
||||
if chunkUsage != nil {
|
||||
for t := chunkUsage.head.next; t != chunkUsage.tail; t = t.next {
|
||||
logicStart := max(off, logicChunkIndex*cw.ChunkSize+t.startOffset)
|
||||
logicStop := min(off+int64(len(p)), logicChunkIndex*cw.ChunkSize+t.stopOffset)
|
||||
if logicStart < logicStop {
|
||||
actualStart := logicStart - logicChunkIndex*cw.ChunkSize + int64(actualChunkIndex)*cw.ChunkSize
|
||||
_, err := cw.file.ReadAt(p[logicStart-off:logicStop-off], actualStart)
|
||||
if err != nil {
|
||||
glog.Errorf("reading temp file: %v", err)
|
||||
break
|
||||
}
|
||||
maxStop = max(maxStop, logicStop)
|
||||
}
|
||||
}
|
||||
}
|
||||
return
|
||||
}
|
||||
|
||||
func (cw *ChunkedFileWriter) toActualWriteOffset(logicOffset int64) (actualOffset int64, chunkUsage *PageChunkWrittenIntervalList) {
|
||||
logicChunkIndex := int(logicOffset / cw.ChunkSize)
|
||||
offsetRemainder := logicOffset % cw.ChunkSize
|
||||
existingActualChunkIndex, found := cw.logicToActualChunkIndex[logicChunkIndex]
|
||||
if found {
|
||||
return int64(existingActualChunkIndex)*cw.ChunkSize + offsetRemainder, cw.chunkUsages[existingActualChunkIndex]
|
||||
}
|
||||
cw.logicToActualChunkIndex[logicChunkIndex] = len(cw.chunkUsages)
|
||||
chunkUsage = newPageChunkWrittenIntervalList()
|
||||
cw.chunkUsages = append(cw.chunkUsages, chunkUsage)
|
||||
return int64(len(cw.chunkUsages)-1)*cw.ChunkSize + offsetRemainder, chunkUsage
|
||||
}
|
||||
|
||||
func (cw *ChunkedFileWriter) toActualReadOffset(logicOffset int64) (actualChunkIndex int, chunkUsage *PageChunkWrittenIntervalList) {
|
||||
logicChunkIndex := int(logicOffset / cw.ChunkSize)
|
||||
existingActualChunkIndex, found := cw.logicToActualChunkIndex[logicChunkIndex]
|
||||
if found {
|
||||
return existingActualChunkIndex, cw.chunkUsages[existingActualChunkIndex]
|
||||
}
|
||||
return 0, nil
|
||||
}
|
||||
|
||||
func (cw *ChunkedFileWriter) ProcessEachInterval(process func(file *os.File, logicChunkIndex int, interval *PageChunkWrittenInterval)) {
|
||||
for logicChunkIndex, actualChunkIndex := range cw.logicToActualChunkIndex {
|
||||
chunkUsage := cw.chunkUsages[actualChunkIndex]
|
||||
for t := chunkUsage.head.next; t != chunkUsage.tail; t = t.next {
|
||||
process(cw.file, logicChunkIndex, t)
|
||||
}
|
||||
}
|
||||
}
|
||||
func (cw *ChunkedFileWriter) Destroy() {
|
||||
if cw.file != nil {
|
||||
cw.file.Close()
|
||||
os.Remove(cw.file.Name())
|
||||
}
|
||||
}
|
||||
|
||||
type FileIntervalReader struct {
|
||||
f *os.File
|
||||
startOffset int64
|
||||
stopOffset int64
|
||||
position int64
|
||||
}
|
||||
|
||||
var _ = io.Reader(&FileIntervalReader{})
|
||||
|
||||
func NewFileIntervalReader(cw *ChunkedFileWriter, logicChunkIndex int, interval *PageChunkWrittenInterval) *FileIntervalReader {
|
||||
actualChunkIndex, found := cw.logicToActualChunkIndex[logicChunkIndex]
|
||||
if !found {
|
||||
// this should never happen
|
||||
return nil
|
||||
}
|
||||
return &FileIntervalReader{
|
||||
f: cw.file,
|
||||
startOffset: int64(actualChunkIndex)*cw.ChunkSize + interval.startOffset,
|
||||
stopOffset: int64(actualChunkIndex)*cw.ChunkSize + interval.stopOffset,
|
||||
position: 0,
|
||||
}
|
||||
}
|
||||
|
||||
func (fr *FileIntervalReader) Read(p []byte) (n int, err error) {
|
||||
readSize := minInt(len(p), int(fr.stopOffset-fr.startOffset-fr.position))
|
||||
n, err = fr.f.ReadAt(p[:readSize], fr.startOffset+fr.position)
|
||||
if err == nil || err == io.EOF {
|
||||
fr.position += int64(n)
|
||||
if fr.stopOffset-fr.startOffset-fr.position == 0 {
|
||||
// return a tiny bit faster
|
||||
err = io.EOF
|
||||
return
|
||||
}
|
||||
}
|
||||
return
|
||||
}
|
||||
Reference in New Issue
Block a user