seaweedfs/weed/storage/erasure_coding/ec_encoder.go

311 lines
8.3 KiB
Go
Raw Normal View History

2019-05-15 08:02:00 +00:00
package erasure_coding
2019-05-15 17:02:44 +00:00
import (
"fmt"
2019-05-15 17:02:44 +00:00
"io"
"os"
"github.com/klauspost/reedsolomon"
"github.com/seaweedfs/seaweedfs/weed/glog"
"github.com/seaweedfs/seaweedfs/weed/storage/idx"
"github.com/seaweedfs/seaweedfs/weed/storage/needle_map"
"github.com/seaweedfs/seaweedfs/weed/storage/types"
"github.com/seaweedfs/seaweedfs/weed/util"
2019-05-15 17:02:44 +00:00
)
2019-05-15 08:02:00 +00:00
const (
DataShardsCount = 10
ParityShardsCount = 4
2019-05-25 09:02:44 +00:00
TotalShardsCount = DataShardsCount + ParityShardsCount
2019-05-15 08:02:00 +00:00
ErasureCodingLargeBlockSize = 1024 * 1024 * 1024 // 1GB
ErasureCodingSmallBlockSize = 1024 * 1024 // 1MB
)
2019-05-15 17:02:44 +00:00
// WriteSortedFileFromIdx generates .ecx file from existing .idx file
// all keys are sorted in ascending order
func WriteSortedFileFromIdx(baseFileName string, ext string) (e error) {
2020-02-05 05:16:34 +00:00
nm, err := readNeedleMap(baseFileName)
if nm != nil {
defer nm.Close()
}
if err != nil {
2020-02-05 05:16:34 +00:00
return fmt.Errorf("readNeedleMap: %v", err)
}
ecxFile, err := os.OpenFile(baseFileName+ext, os.O_TRUNC|os.O_CREATE|os.O_WRONLY, 0644)
if err != nil {
2019-06-20 05:57:14 +00:00
return fmt.Errorf("failed to open ecx file: %v", err)
}
defer ecxFile.Close()
2020-02-05 05:16:34 +00:00
err = nm.AscendingVisit(func(value needle_map.NeedleValue) error {
bytes := value.ToBytes()
_, writeErr := ecxFile.Write(bytes)
return writeErr
})
if err != nil {
return fmt.Errorf("failed to visit idx file: %v", err)
}
return nil
}
2019-12-23 20:48:20 +00:00
// WriteEcFiles generates .ec00 ~ .ec13 files
func WriteEcFiles(baseFileName string) error {
return generateEcFiles(baseFileName, 256*1024, ErasureCodingLargeBlockSize, ErasureCodingSmallBlockSize)
}
2019-06-03 09:26:31 +00:00
func RebuildEcFiles(baseFileName string) ([]uint32, error) {
return generateMissingEcFiles(baseFileName, 256*1024, ErasureCodingLargeBlockSize, ErasureCodingSmallBlockSize)
}
func ToExt(ecIndex int) string {
return fmt.Sprintf(".ec%02d", ecIndex)
}
func generateEcFiles(baseFileName string, bufferSize int, largeBlockSize int64, smallBlockSize int64) error {
file, err := os.OpenFile(baseFileName+".dat", os.O_RDONLY, 0)
if err != nil {
return fmt.Errorf("failed to open dat file: %v", err)
}
defer file.Close()
fi, err := file.Stat()
if err != nil {
return fmt.Errorf("failed to stat dat file: %v", err)
}
2020-06-25 16:43:38 +00:00
2020-06-25 17:45:34 +00:00
glog.V(0).Infof("encodeDatFile %s.dat size:%d", baseFileName, fi.Size())
err = encodeDatFile(fi.Size(), err, baseFileName, bufferSize, largeBlockSize, file, smallBlockSize)
if err != nil {
return fmt.Errorf("encodeDatFile: %v", err)
}
return nil
}
2019-06-03 09:26:31 +00:00
func generateMissingEcFiles(baseFileName string, bufferSize int, largeBlockSize int64, smallBlockSize int64) (generatedShardIds []uint32, err error) {
shardHasData := make([]bool, TotalShardsCount)
inputFiles := make([]*os.File, TotalShardsCount)
outputFiles := make([]*os.File, TotalShardsCount)
for shardId := 0; shardId < TotalShardsCount; shardId++ {
shardFileName := baseFileName + ToExt(shardId)
if util.FileExists(shardFileName) {
shardHasData[shardId] = true
inputFiles[shardId], err = os.OpenFile(shardFileName, os.O_RDONLY, 0)
if err != nil {
return nil, err
}
defer inputFiles[shardId].Close()
} else {
outputFiles[shardId], err = os.OpenFile(shardFileName, os.O_TRUNC|os.O_WRONLY|os.O_CREATE, 0644)
if err != nil {
return nil, err
}
defer outputFiles[shardId].Close()
generatedShardIds = append(generatedShardIds, uint32(shardId))
}
}
err = rebuildEcFiles(shardHasData, inputFiles, outputFiles)
if err != nil {
return nil, fmt.Errorf("rebuildEcFiles: %v", err)
}
return
}
func encodeData(file *os.File, enc reedsolomon.Encoder, startOffset, blockSize int64, buffers [][]byte, outputs []*os.File) error {
bufferSize := int64(len(buffers[0]))
if bufferSize == 0 {
glog.Fatal("unexpected zero buffer size")
}
batchCount := blockSize / bufferSize
if blockSize%bufferSize != 0 {
glog.Fatalf("unexpected block size %d buffer size %d", blockSize, bufferSize)
}
for b := int64(0); b < batchCount; b++ {
err := encodeDataOneBatch(file, enc, startOffset+b*bufferSize, blockSize, buffers, outputs)
if err != nil {
return err
}
}
return nil
}
func openEcFiles(baseFileName string, forRead bool) (files []*os.File, err error) {
2019-05-25 09:02:44 +00:00
for i := 0; i < TotalShardsCount; i++ {
2019-05-24 18:52:23 +00:00
fname := baseFileName + ToExt(i)
openOption := os.O_TRUNC | os.O_CREATE | os.O_WRONLY
2019-05-19 10:01:58 +00:00
if forRead {
openOption = os.O_RDONLY
}
f, err := os.OpenFile(fname, openOption, 0644)
if err != nil {
return files, fmt.Errorf("failed to open file %s: %v", fname, err)
}
files = append(files, f)
}
return
}
func closeEcFiles(files []*os.File) {
for _, f := range files {
if f != nil {
f.Close()
}
}
}
func encodeDataOneBatch(file *os.File, enc reedsolomon.Encoder, startOffset, blockSize int64, buffers [][]byte, outputs []*os.File) error {
2019-05-15 17:02:44 +00:00
// read data into buffers
for i := 0; i < DataShardsCount; i++ {
n, err := file.ReadAt(buffers[i], startOffset+blockSize*int64(i))
if err != nil {
if err != io.EOF {
return err
}
}
if n < len(buffers[i]) {
for t := len(buffers[i]) - 1; t >= n; t-- {
buffers[i][t] = 0
}
}
}
err := enc.Encode(buffers)
if err != nil {
return err
}
2019-05-25 09:02:44 +00:00
for i := 0; i < TotalShardsCount; i++ {
_, err := outputs[i].Write(buffers[i])
if err != nil {
return err
}
}
2019-05-15 17:02:44 +00:00
return nil
}
func encodeDatFile(remainingSize int64, err error, baseFileName string, bufferSize int, largeBlockSize int64, file *os.File, smallBlockSize int64) error {
2019-06-03 09:26:31 +00:00
var processedSize int64
2019-06-03 09:26:31 +00:00
enc, err := reedsolomon.New(DataShardsCount, ParityShardsCount)
if err != nil {
return fmt.Errorf("failed to create encoder: %v", err)
}
2019-06-03 09:26:31 +00:00
2019-05-25 09:02:44 +00:00
buffers := make([][]byte, TotalShardsCount)
for i := range buffers {
2019-06-03 09:26:31 +00:00
buffers[i] = make([]byte, bufferSize)
}
outputs, err := openEcFiles(baseFileName, false)
defer closeEcFiles(outputs)
if err != nil {
2019-06-20 05:57:14 +00:00
return fmt.Errorf("failed to open ec files %s: %v", baseFileName, err)
}
2019-06-03 09:26:31 +00:00
for remainingSize > largeBlockSize*DataShardsCount {
err = encodeData(file, enc, processedSize, largeBlockSize, buffers, outputs)
if err != nil {
return fmt.Errorf("failed to encode large chunk data: %v", err)
}
remainingSize -= largeBlockSize * DataShardsCount
processedSize += largeBlockSize * DataShardsCount
}
for remainingSize > 0 {
err = encodeData(file, enc, processedSize, smallBlockSize, buffers, outputs)
if err != nil {
return fmt.Errorf("failed to encode small chunk data: %v", err)
}
remainingSize -= smallBlockSize * DataShardsCount
processedSize += smallBlockSize * DataShardsCount
}
return nil
}
2019-06-03 09:26:31 +00:00
func rebuildEcFiles(shardHasData []bool, inputFiles []*os.File, outputFiles []*os.File) error {
enc, err := reedsolomon.New(DataShardsCount, ParityShardsCount)
if err != nil {
return fmt.Errorf("failed to create encoder: %v", err)
}
buffers := make([][]byte, TotalShardsCount)
for i := range buffers {
2019-06-03 09:26:31 +00:00
if shardHasData[i] {
buffers[i] = make([]byte, ErasureCodingSmallBlockSize)
}
}
var startOffset int64
var inputBufferDataSize int
for {
// read the input data from files
for i := 0; i < TotalShardsCount; i++ {
if shardHasData[i] {
n, _ := inputFiles[i].ReadAt(buffers[i], startOffset)
if n == 0 {
return nil
}
2019-06-03 18:50:54 +00:00
if inputBufferDataSize == 0 {
2019-06-03 09:26:31 +00:00
inputBufferDataSize = n
}
if inputBufferDataSize != n {
2019-06-03 18:50:54 +00:00
return fmt.Errorf("ec shard size expected %d actual %d", inputBufferDataSize, n)
2019-06-03 09:26:31 +00:00
}
} else {
buffers[i] = nil
}
}
// encode the data
err = enc.Reconstruct(buffers)
if err != nil {
return fmt.Errorf("reconstruct: %v", err)
}
// write the data to output files
for i := 0; i < TotalShardsCount; i++ {
if !shardHasData[i] {
n, _ := outputFiles[i].WriteAt(buffers[i][:inputBufferDataSize], startOffset)
if inputBufferDataSize != n {
return fmt.Errorf("fail to write to %s", outputFiles[i].Name())
}
}
}
startOffset += int64(inputBufferDataSize)
}
}
2020-02-05 05:16:34 +00:00
func readNeedleMap(baseFileName string) (*needle_map.MemDb, error) {
indexFile, err := os.OpenFile(baseFileName+".idx", os.O_RDONLY, 0644)
if err != nil {
return nil, fmt.Errorf("cannot read Volume Index %s.idx: %v", baseFileName, err)
}
defer indexFile.Close()
cm := needle_map.NewMemDb()
err = idx.WalkIndexFile(indexFile, 0, func(key types.NeedleId, offset types.Offset, size types.Size) error {
if !offset.IsZero() && size != types.TombstoneFileSize {
cm.Set(key, offset, size)
} else {
cm.Delete(key)
}
return nil
})
return cm, err
}