seaweedfs/weed/storage/volume_vacuum.go

340 lines
10 KiB
Go
Raw Normal View History

2014-05-20 02:24:35 +00:00
package storage
import (
2014-05-20 03:54:39 +00:00
"fmt"
2014-05-20 02:24:35 +00:00
"os"
"time"
"github.com/chrislusf/seaweedfs/weed/glog"
2018-07-22 00:39:10 +00:00
. "github.com/chrislusf/seaweedfs/weed/storage/types"
"github.com/chrislusf/seaweedfs/weed/util"
2014-05-20 02:24:35 +00:00
)
func (v *Volume) garbageLevel() float64 {
if v.ContentSize() == 0 {
return 0
}
2014-05-20 02:24:35 +00:00
return float64(v.nm.DeletedSize()) / float64(v.ContentSize())
}
func (v *Volume) Compact(preallocate int64) error {
2018-10-19 03:47:30 +00:00
glog.V(3).Infof("Compacting volume %d ...", v.Id)
//no need to lock for copy on write
//v.accessLock.Lock()
//defer v.accessLock.Unlock()
//glog.V(3).Infof("Got Compaction lock...")
2014-05-20 02:24:35 +00:00
filePath := v.FileName()
2016-10-07 08:22:24 +00:00
v.lastCompactIndexOffset = v.nm.IndexFileSize()
v.lastCompactRevision = v.SuperBlock.CompactRevision
glog.V(3).Infof("creating copies for volume %d ,last offset %d...", v.Id, v.lastCompactIndexOffset)
return v.copyDataAndGenerateIndexFile(filePath+".cpd", filePath+".cpx", preallocate)
2014-05-20 02:24:35 +00:00
}
func (v *Volume) Compact2() error {
2018-10-19 03:47:30 +00:00
glog.V(3).Infof("Compact2 volume %d ...", v.Id)
filePath := v.FileName()
glog.V(3).Infof("creating copies for volume %d ...", v.Id)
return v.copyDataBasedOnIndexFile(filePath+".cpd", filePath+".cpx")
}
2014-05-20 02:24:35 +00:00
func (v *Volume) commitCompact() error {
2018-10-19 03:47:30 +00:00
glog.V(0).Infof("Committing volume %d vacuuming...", v.Id)
2015-05-23 17:16:01 +00:00
v.dataFileAccessLock.Lock()
defer v.dataFileAccessLock.Unlock()
2018-10-19 03:47:30 +00:00
glog.V(3).Infof("Got volume %d committing lock...", v.Id)
v.nm.Close()
2014-05-20 02:24:35 +00:00
_ = v.dataFile.Close()
2014-05-20 02:24:35 +00:00
var e error
if e = v.makeupDiff(v.FileName()+".cpd", v.FileName()+".cpx", v.FileName()+".dat", v.FileName()+".idx"); e != nil {
2018-10-19 03:47:30 +00:00
glog.V(0).Infof("makeupDiff in commitCompact volume %d failed %v", v.Id, e)
e = os.Remove(v.FileName() + ".cpd")
if e != nil {
return e
}
e = os.Remove(v.FileName() + ".cpx")
if e != nil {
return e
}
} else {
var e error
if e = os.Rename(v.FileName()+".cpd", v.FileName()+".dat"); e != nil {
return e
}
if e = os.Rename(v.FileName()+".cpx", v.FileName()+".idx"); e != nil {
return e
}
2014-05-20 02:24:35 +00:00
}
2014-05-27 00:34:54 +00:00
//glog.V(3).Infof("Pretending to be vacuuming...")
//time.Sleep(20 * time.Second)
os.RemoveAll(v.FileName() + ".ldb")
os.RemoveAll(v.FileName() + ".bdb")
2018-10-19 03:47:30 +00:00
glog.V(3).Infof("Loading volume %d commit file...", v.Id)
2017-01-08 19:01:46 +00:00
if e = v.load(true, false, v.needleMapKind, 0); e != nil {
2014-05-20 02:24:35 +00:00
return e
}
return nil
}
2014-05-20 03:54:39 +00:00
func (v *Volume) cleanupCompact() error {
2018-10-19 03:47:30 +00:00
glog.V(0).Infof("Cleaning up volume %d vacuuming...", v.Id)
e1 := os.Remove(v.FileName() + ".cpd")
e2 := os.Remove(v.FileName() + ".cpx")
if e1 != nil {
return e1
}
if e2 != nil {
return e2
}
return nil
}
2016-10-07 08:22:24 +00:00
func fetchCompactRevisionFromDatFile(file *os.File) (compactRevision uint16, err error) {
2018-06-24 18:37:08 +00:00
superBlock, err := ReadSuperBlock(file)
2016-10-07 08:22:24 +00:00
if err != nil {
return 0, err
}
return superBlock.CompactRevision, nil
}
func (v *Volume) makeupDiff(newDatFileName, newIdxFileName, oldDatFileName, oldIdxFileName string) (err error) {
var indexSize int64
oldIdxFile, err := os.Open(oldIdxFileName)
defer oldIdxFile.Close()
oldDatFile, err := os.Open(oldDatFileName)
defer oldDatFile.Close()
if indexSize, err = verifyIndexFileIntegrity(oldIdxFile); err != nil {
return fmt.Errorf("verifyIndexFileIntegrity %s failed: %v", oldIdxFileName, err)
}
2016-10-07 08:22:24 +00:00
if indexSize == 0 || uint64(indexSize) <= v.lastCompactIndexOffset {
return nil
}
2016-10-07 08:22:24 +00:00
oldDatCompactRevision, err := fetchCompactRevisionFromDatFile(oldDatFile)
if err != nil {
return fmt.Errorf("fetchCompactRevisionFromDatFile src %s failed: %v", oldDatFile.Name(), err)
2016-10-07 08:22:24 +00:00
}
if oldDatCompactRevision != v.lastCompactRevision {
return fmt.Errorf("current old dat file's compact revision %d is not the expected one %d", oldDatCompactRevision, v.lastCompactRevision)
}
type keyField struct {
offset Offset
2016-10-07 08:22:24 +00:00
size uint32
}
incrementedHasUpdatedIndexEntry := make(map[NeedleId]keyField)
2016-10-07 08:22:24 +00:00
for idx_offset := indexSize - NeedleEntrySize; uint64(idx_offset) >= v.lastCompactIndexOffset; idx_offset -= NeedleEntrySize {
var IdxEntry []byte
if IdxEntry, err = readIndexEntryAtOffset(oldIdxFile, idx_offset); err != nil {
return fmt.Errorf("readIndexEntry %s at offset %d failed: %v", oldIdxFileName, idx_offset, err)
}
key, offset, size := IdxFileEntry(IdxEntry)
2018-07-07 07:48:58 +00:00
glog.V(4).Infof("key %d offset %d size %d", key, offset, size)
2016-10-07 08:22:24 +00:00
if _, found := incrementedHasUpdatedIndexEntry[key]; !found {
incrementedHasUpdatedIndexEntry[key] = keyField{
offset: offset,
size: size,
}
}
}
// no updates during commit step
if len(incrementedHasUpdatedIndexEntry) == 0 {
return nil
}
// deal with updates during commit step
var (
dst, idx *os.File
)
if dst, err = os.OpenFile(newDatFileName, os.O_RDWR, 0644); err != nil {
return fmt.Errorf("open dat file %s failed: %v", newDatFileName, err)
}
defer dst.Close()
if idx, err = os.OpenFile(newIdxFileName, os.O_RDWR, 0644); err != nil {
return fmt.Errorf("open idx file %s failed: %v", newIdxFileName, err)
}
defer idx.Close()
var newDatCompactRevision uint16
newDatCompactRevision, err = fetchCompactRevisionFromDatFile(dst)
if err != nil {
return fmt.Errorf("fetchCompactRevisionFromDatFile dst %s failed: %v", dst.Name(), err)
}
if oldDatCompactRevision+1 != newDatCompactRevision {
return fmt.Errorf("oldDatFile %s 's compact revision is %d while newDatFile %s 's compact revision is %d", oldDatFileName, oldDatCompactRevision, newDatFileName, newDatCompactRevision)
}
2016-10-07 08:22:24 +00:00
idx_entry_bytes := make([]byte, NeedleIdSize+OffsetSize+SizeSize)
for key, incre_idx_entry := range incrementedHasUpdatedIndexEntry {
NeedleIdToBytes(idx_entry_bytes[0:NeedleIdSize], key)
OffsetToBytes(idx_entry_bytes[NeedleIdSize:NeedleIdSize+OffsetSize], incre_idx_entry.offset)
util.Uint32toBytes(idx_entry_bytes[NeedleIdSize+OffsetSize:NeedleIdSize+OffsetSize+SizeSize], incre_idx_entry.size)
var offset int64
if offset, err = dst.Seek(0, 2); err != nil {
glog.V(0).Infof("failed to seek the end of file: %v", err)
return
}
//ensure file writing starting from aligned positions
if offset%NeedlePaddingSize != 0 {
offset = offset + (NeedlePaddingSize - offset%NeedlePaddingSize)
if offset, err = v.dataFile.Seek(offset, 0); err != nil {
glog.V(0).Infof("failed to align in datafile %s: %v", v.dataFile.Name(), err)
return
}
}
2016-10-07 08:22:24 +00:00
//updated needle
if incre_idx_entry.offset != 0 && incre_idx_entry.size != 0 && incre_idx_entry.size != TombstoneFileSize {
//even the needle cache in memory is hit, the need_bytes is correct
2018-07-07 07:48:58 +00:00
glog.V(4).Infof("file %d offset %d size %d", key, int64(incre_idx_entry.offset)*NeedlePaddingSize, incre_idx_entry.size)
var needle_bytes []byte
2018-07-24 08:36:04 +00:00
needle_bytes, err = ReadNeedleBlob(oldDatFile, int64(incre_idx_entry.offset)*NeedlePaddingSize, incre_idx_entry.size, v.Version())
if err != nil {
return fmt.Errorf("ReadNeedleBlob %s key %d offset %d size %d failed: %v", oldDatFile.Name(), key, int64(incre_idx_entry.offset)*NeedlePaddingSize, incre_idx_entry.size, err)
2016-10-07 08:22:24 +00:00
}
dst.Write(needle_bytes)
util.Uint32toBytes(idx_entry_bytes[8:12], uint32(offset/NeedlePaddingSize))
} else { //deleted needle
//fakeDelNeedle 's default Data field is nil
fakeDelNeedle := new(Needle)
fakeDelNeedle.Id = key
fakeDelNeedle.Cookie = 0x12345678
fakeDelNeedle.AppendAtNs = uint64(time.Now().UnixNano())
_, _, err = fakeDelNeedle.Append(dst, v.Version())
if err != nil {
return fmt.Errorf("append deleted %d failed: %v", key, err)
2016-10-07 08:22:24 +00:00
}
util.Uint32toBytes(idx_entry_bytes[8:12], uint32(0))
}
if _, err := idx.Seek(0, 2); err != nil {
return fmt.Errorf("cannot seek end of indexfile %s: %v",
newIdxFileName, err)
}
_, err = idx.Write(idx_entry_bytes)
}
2016-09-27 05:30:44 +00:00
return nil
2016-09-27 05:26:39 +00:00
}
2017-01-08 19:01:46 +00:00
func (v *Volume) copyDataAndGenerateIndexFile(dstName, idxName string, preallocate int64) (err error) {
2014-05-20 03:54:39 +00:00
var (
dst, idx *os.File
)
2017-01-08 19:01:46 +00:00
if dst, err = createVolumeFile(dstName, preallocate); err != nil {
2014-05-20 03:54:39 +00:00
return
}
defer dst.Close()
if idx, err = os.OpenFile(idxName, os.O_WRONLY|os.O_CREATE|os.O_TRUNC, 0644); err != nil {
return
}
defer idx.Close()
2017-05-27 05:51:25 +00:00
nm := NewBtreeNeedleMap(idx)
2018-06-24 18:37:08 +00:00
new_offset := int64(0)
2014-05-20 03:54:39 +00:00
now := uint64(time.Now().Unix())
2018-07-24 08:36:04 +00:00
var version Version
err = ScanVolumeFile(v.dir, v.Collection, v.Id, v.needleMapKind,
func(superBlock SuperBlock) error {
2018-07-24 08:36:04 +00:00
version = superBlock.Version()
superBlock.CompactRevision++
_, err = dst.Write(superBlock.Bytes())
2018-06-24 18:37:08 +00:00
new_offset = int64(superBlock.BlockSize())
return err
}, true, func(n *Needle, offset int64) error {
if n.HasTtl() && now >= n.LastModified+uint64(v.Ttl.Minutes()*60) {
return nil
2014-05-20 03:54:39 +00:00
}
nv, ok := v.nm.Get(n.Id)
glog.V(4).Infoln("needle expected offset ", offset, "ok", ok, "nv", nv)
if ok && int64(nv.Offset)*NeedlePaddingSize == offset && nv.Size > 0 {
if err = nm.Put(n.Id, Offset(new_offset/NeedlePaddingSize), n.Size); err != nil {
return fmt.Errorf("cannot put needle: %s", err)
}
2017-01-06 18:22:20 +00:00
if _, _, err := n.Append(dst, v.Version()); err != nil {
return fmt.Errorf("cannot append needle: %s", err)
}
2018-07-24 08:36:04 +00:00
new_offset += n.DiskSize(version)
2018-10-15 07:40:12 +00:00
glog.V(4).Infoln("saving key", n.Id, "volume offset", offset, "=>", new_offset, "data_size", n.Size)
2014-05-20 03:54:39 +00:00
}
return nil
})
2014-05-20 03:54:39 +00:00
return
}
func (v *Volume) copyDataBasedOnIndexFile(dstName, idxName string) (err error) {
var (
dst, idx, oldIndexFile *os.File
)
if dst, err = os.OpenFile(dstName, os.O_WRONLY|os.O_CREATE|os.O_TRUNC, 0644); err != nil {
return
}
defer dst.Close()
if idx, err = os.OpenFile(idxName, os.O_WRONLY|os.O_CREATE|os.O_TRUNC, 0644); err != nil {
return
}
defer idx.Close()
if oldIndexFile, err = os.OpenFile(v.FileName()+".idx", os.O_RDONLY, 0644); err != nil {
return
}
defer oldIndexFile.Close()
2017-05-27 05:51:25 +00:00
nm := NewBtreeNeedleMap(idx)
now := uint64(time.Now().Unix())
v.SuperBlock.CompactRevision++
dst.Write(v.SuperBlock.Bytes())
2018-06-24 18:37:08 +00:00
new_offset := int64(v.SuperBlock.BlockSize())
WalkIndexFile(oldIndexFile, func(key NeedleId, offset Offset, size uint32) error {
2017-01-06 18:22:20 +00:00
if offset == 0 || size == TombstoneFileSize {
return nil
}
nv, ok := v.nm.Get(key)
if !ok {
return nil
}
n := new(Needle)
n.ReadData(v.dataFile, int64(offset)*NeedlePaddingSize, size, v.Version())
if n.HasTtl() && now >= n.LastModified+uint64(v.Ttl.Minutes()*60) {
return nil
}
glog.V(4).Infoln("needle expected offset ", offset, "ok", ok, "nv", nv)
if nv.Offset == offset && nv.Size > 0 {
if err = nm.Put(n.Id, Offset(new_offset/NeedlePaddingSize), n.Size); err != nil {
return fmt.Errorf("cannot put needle: %s", err)
}
2017-01-06 18:22:20 +00:00
if _, _, err = n.Append(dst, v.Version()); err != nil {
return fmt.Errorf("cannot append needle: %s", err)
}
2018-07-24 08:36:04 +00:00
new_offset += n.DiskSize(v.Version())
glog.V(3).Infoln("saving key", n.Id, "volume offset", offset, "=>", new_offset, "data_size", n.Size)
}
return nil
})
return
}