seaweedfs/weed/filer/reader_at.go

217 lines
6.6 KiB
Go
Raw Normal View History

2020-09-01 07:21:19 +00:00
package filer
import (
"context"
"fmt"
2020-11-01 10:36:43 +00:00
"io"
"math/rand"
"sync"
"github.com/seaweedfs/seaweedfs/weed/glog"
"github.com/seaweedfs/seaweedfs/weed/pb/filer_pb"
"github.com/seaweedfs/seaweedfs/weed/util"
"github.com/seaweedfs/seaweedfs/weed/util/chunk_cache"
"github.com/seaweedfs/seaweedfs/weed/wdclient"
2020-10-11 03:09:43 +00:00
)
type ChunkReadAt struct {
2022-02-26 10:16:47 +00:00
masterClient *wdclient.MasterClient
chunkViews []*ChunkView
readerLock sync.Mutex
fileSize int64
readerCache *ReaderCache
readerPattern *ReaderPattern
lastChunkFid string
}
var _ = io.ReaderAt(&ChunkReadAt{})
var _ = io.Closer(&ChunkReadAt{})
func LookupFn(filerClient filer_pb.FilerClient) wdclient.LookupFileIdFunctionType {
2020-10-04 03:16:42 +00:00
vidCache := make(map[string]*filer_pb.Locations)
2020-10-22 02:28:59 +00:00
var vicCacheLock sync.RWMutex
return func(fileId string) (targetUrls []string, err error) {
2020-10-04 03:16:42 +00:00
vid := VolumeId(fileId)
2020-10-22 02:28:59 +00:00
vicCacheLock.RLock()
2020-10-04 03:16:42 +00:00
locations, found := vidCache[vid]
2020-10-22 02:28:59 +00:00
vicCacheLock.RUnlock()
2020-10-04 03:16:42 +00:00
2020-11-01 08:58:48 +00:00
if !found {
2020-11-01 10:36:43 +00:00
util.Retry("lookup volume "+vid, func() error {
err = filerClient.WithFilerClient(false, func(client filer_pb.SeaweedFilerClient) error {
2020-11-01 08:58:48 +00:00
resp, err := client.LookupVolume(context.Background(), &filer_pb.LookupVolumeRequest{
VolumeIds: []string{vid},
})
if err != nil {
return err
}
locations = resp.LocationsMap[vid]
if locations == nil || len(locations.Locations) == 0 {
glog.V(0).Infof("failed to locate %s", fileId)
return fmt.Errorf("failed to locate %s", fileId)
}
vicCacheLock.Lock()
vidCache[vid] = locations
vicCacheLock.Unlock()
return nil
2020-10-04 03:16:42 +00:00
})
2020-11-01 08:58:48 +00:00
return err
})
2020-10-04 03:16:42 +00:00
}
2020-10-10 22:43:22 +00:00
if err != nil {
return nil, err
}
fcDataCenter := filerClient.GetDataCenter()
var sameDcTargetUrls, otherTargetUrls []string
for _, loc := range locations.Locations {
2021-01-28 22:36:29 +00:00
volumeServerAddress := filerClient.AdjustedUrl(loc)
targetUrl := fmt.Sprintf("http://%s/%s", volumeServerAddress, fileId)
if fcDataCenter == "" || fcDataCenter != loc.DataCenter {
otherTargetUrls = append(otherTargetUrls, targetUrl)
} else {
sameDcTargetUrls = append(sameDcTargetUrls, targetUrl)
}
}
rand.Shuffle(len(sameDcTargetUrls), func(i, j int) {
sameDcTargetUrls[i], sameDcTargetUrls[j] = sameDcTargetUrls[j], sameDcTargetUrls[i]
})
rand.Shuffle(len(otherTargetUrls), func(i, j int) {
otherTargetUrls[i], otherTargetUrls[j] = otherTargetUrls[j], otherTargetUrls[i]
})
// Prefer same data center
targetUrls = append(sameDcTargetUrls, otherTargetUrls...)
return
}
}
func NewChunkReaderAtFromClient(lookupFn wdclient.LookupFileIdFunctionType, chunkViews []*ChunkView, chunkCache chunk_cache.ChunkCache, fileSize int64) *ChunkReadAt {
return &ChunkReadAt{
chunkViews: chunkViews,
fileSize: fileSize,
readerCache: newReaderCache(32, chunkCache, lookupFn),
readerPattern: NewReaderPattern(),
}
}
func (c *ChunkReadAt) Close() error {
2022-02-26 10:16:47 +00:00
c.readerCache.destroy()
return nil
}
func (c *ChunkReadAt) ReadAt(p []byte, offset int64) (n int, err error) {
c.readerPattern.MonitorReadAt(offset, len(p))
c.readerLock.Lock()
defer c.readerLock.Unlock()
2021-05-11 04:47:51 +00:00
// glog.V(4).Infof("ReadAt [%d,%d) of total file size %d bytes %d chunk views", offset, offset+int64(len(p)), c.fileSize, len(c.chunkViews))
2021-01-18 09:14:27 +00:00
return c.doReadAt(p, offset)
}
func (c *ChunkReadAt) doReadAt(p []byte, offset int64) (n int, err error) {
2020-08-18 03:14:40 +00:00
startOffset, remaining := offset, int64(len(p))
2022-02-26 11:06:17 +00:00
var nextChunks []*ChunkView
2020-08-17 23:05:40 +00:00
for i, chunk := range c.chunkViews {
2020-08-18 03:14:40 +00:00
if remaining <= 0 {
break
}
2020-10-04 08:31:04 +00:00
if i+1 < len(c.chunkViews) {
2022-02-26 11:06:17 +00:00
nextChunks = c.chunkViews[i+1:]
2020-10-04 08:31:04 +00:00
}
2020-08-18 03:14:40 +00:00
if startOffset < chunk.LogicOffset {
gap := chunk.LogicOffset - startOffset
2022-01-13 14:20:33 +00:00
glog.V(4).Infof("zero [%d,%d)", startOffset, chunk.LogicOffset)
2022-08-21 18:49:29 +00:00
n += zero(p, startOffset-offset, gap)
startOffset, remaining = chunk.LogicOffset, remaining-gap
2020-08-18 03:14:40 +00:00
if remaining <= 0 {
break
}
2020-08-17 04:07:46 +00:00
}
2020-08-16 22:16:46 +00:00
// fmt.Printf(">>> doReadAt [%d,%d), chunk[%d,%d)\n", offset, offset+int64(len(p)), chunk.LogicOffset, chunk.LogicOffset+int64(chunk.Size))
2020-08-18 03:14:40 +00:00
chunkStart, chunkStop := max(chunk.LogicOffset, startOffset), min(chunk.LogicOffset+int64(chunk.Size), startOffset+remaining)
2020-08-17 04:07:46 +00:00
if chunkStart >= chunkStop {
continue
}
2021-05-11 04:47:51 +00:00
// glog.V(4).Infof("read [%d,%d), %d/%d chunk %s [%d,%d)", chunkStart, chunkStop, i, len(c.chunkViews), chunk.FileId, chunk.LogicOffset-chunk.Offset, chunk.LogicOffset-chunk.Offset+int64(chunk.Size))
bufferOffset := chunkStart - chunk.LogicOffset + chunk.Offset
2022-02-26 11:06:17 +00:00
copied, err := c.readChunkSliceAt(p[startOffset-offset:chunkStop-chunkStart+startOffset-offset], chunk, nextChunks, uint64(bufferOffset))
2020-08-17 04:07:46 +00:00
if err != nil {
glog.Errorf("fetching chunk %+v: %v\n", chunk, err)
2022-02-26 10:16:47 +00:00
return copied, err
}
2020-08-17 04:07:46 +00:00
n += copied
2020-08-18 03:20:08 +00:00
startOffset, remaining = startOffset+int64(copied), remaining-int64(copied)
}
2020-08-16 07:49:08 +00:00
2021-05-11 04:47:51 +00:00
// glog.V(4).Infof("doReadAt [%d,%d), n:%v, err:%v", offset, offset+int64(len(p)), n, err)
2020-08-16 07:49:08 +00:00
// zero the remaining bytes if a gap exists at the end of the last chunk (or a fully sparse file)
if err == nil && remaining > 0 {
var delta int64
if c.fileSize > startOffset {
delta = min(remaining, c.fileSize-startOffset)
startOffset -= offset
} else {
delta = remaining
startOffset = max(startOffset-offset, startOffset-remaining-offset)
}
glog.V(4).Infof("zero2 [%d,%d) of file size %d bytes", startOffset, startOffset+delta, c.fileSize)
2022-08-21 18:49:29 +00:00
n += zero(p, startOffset, delta)
}
2020-08-18 03:14:40 +00:00
if err == nil && offset+int64(len(p)) >= c.fileSize {
2020-08-16 07:49:08 +00:00
err = io.EOF
}
2020-08-16 22:16:46 +00:00
// fmt.Printf("~~~ filled %d, err: %v\n\n", n, err)
return
}
2022-02-26 11:06:17 +00:00
func (c *ChunkReadAt) readChunkSliceAt(buffer []byte, chunkView *ChunkView, nextChunkViews []*ChunkView, offset uint64) (n int, err error) {
2021-12-23 00:17:30 +00:00
if c.readerPattern.IsRandomMode() {
n, err := c.readerCache.chunkCache.ReadChunkAt(buffer, chunkView.FileId, offset)
if n > 0 {
return n, err
}
return fetchChunkRange(buffer, c.readerCache.lookupFileIdFn, chunkView.FileId, chunkView.CipherKey, chunkView.IsGzipped, int64(offset))
}
2022-02-26 10:16:47 +00:00
n, err = c.readerCache.ReadChunkAt(buffer, chunkView.FileId, chunkView.CipherKey, chunkView.IsGzipped, int64(offset), int(chunkView.ChunkSize), chunkView.LogicOffset == 0)
if c.lastChunkFid != chunkView.FileId {
if chunkView.Offset == 0 { // start of a new chunk
if c.lastChunkFid != "" {
c.readerCache.UnCache(c.lastChunkFid)
c.readerCache.MaybeCache(nextChunkViews)
} else {
if len(nextChunkViews) >= 1 {
c.readerCache.MaybeCache(nextChunkViews[:1]) // just read the next chunk if at the very beginning
}
}
2022-02-26 11:06:17 +00:00
}
}
c.lastChunkFid = chunkView.FileId
2020-08-16 22:16:46 +00:00
return
}
2022-08-21 18:49:29 +00:00
func zero(buffer []byte, start, length int64) int {
end := min(start+length, int64(len(buffer)))
2022-08-21 18:49:29 +00:00
start = max(start, 0)
// zero the bytes
for o := start; o < end; o++ {
buffer[o] = 0
}
2022-08-21 18:49:29 +00:00
return int(end - start)
}