// flush flush writer buffer.
func (b *SuperBlock) flush(force bool) (err error) {
var (
fd uintptr
offset int64
size int64
)
if b.write++; !force && b.write < b.conf.Block.SyncWrite {
return
}
b.write = 0
offset = needle.BlockOffset(b.syncOffset)
size = needle.BlockOffset(b.Offset - b.syncOffset)
fd = b.w.Fd()
if b.conf.Block.Syncfilerange {
if err = myos.Syncfilerange(fd, offset, size, myos.SYNC_FILE_RANGE_WRITE); err != nil {
log.Errorf("block: %s Syncfilerange() error(%v)", b.File, err)
b.LastErr = err
return
}
} else {
if err = myos.Fdatasync(fd); err != nil {
log.Errorf("block: %s Fdatasync() error(%v)", b.File, err)
b.LastErr = err
return
}
}
if err = myos.Fadvise(fd, offset, size, myos.POSIX_FADV_DONTNEED); err == nil {
b.syncOffset = b.Offset
} else {
log.Errorf("block: %s Fadvise() error(%v)", b.File, err)
b.LastErr = err
}
return
}
// init recovery super block from index or super block.
func (v *Volume) init() (err error) {
var (
size int64
offset uint32
lastOffset uint32
)
// recovery from index
if err = v.Indexer.Recovery(func(ix *index.Index) error {
// must no less than last offset
if ix.Offset < lastOffset {
log.Error("recovery index: %s lastoffset: %d error(%v)", ix, lastOffset, errors.ErrIndexOffset)
return errors.ErrIndexOffset
}
// WARN if index's offset more than the block, discard it.
if size = int64(ix.Size) + needle.BlockOffset(ix.Offset); size > v.Block.Size {
log.Error("recovery index: %s EOF", ix)
return errors.ErrIndexEOF
}
v.needles[ix.Key] = needle.NewCache(ix.Offset, ix.Size)
offset = ix.Offset + needle.NeedleOffset(int64(ix.Size))
lastOffset = ix.Offset
return nil
}); err != nil && err != errors.ErrIndexEOF {
return
}
// recovery from super block
if err = v.Block.Recovery(offset, func(n *needle.Needle, so, eo uint32) (err1 error) {
if n.Flag == needle.FlagOK {
if err1 = v.Indexer.Write(n.Key, so, n.TotalSize); err1 != nil {
return
}
} else {
so = needle.CacheDelOffset
}
v.needles[n.Key] = needle.NewCache(so, n.TotalSize)
return
}); err != nil {
return
}
// recheck offset, keep size and offset consistency
if v.Block.Size != needle.BlockOffset(v.Block.Offset) {
log.Error("block: %s size: %d, offset: %d (%d size) not consistency", v.Block.File, v.Block.Size, v.Block.Offset, needle.BlockOffset(v.Block.Offset))
return errors.ErrSuperBlockOffset
}
// flush index
err = v.Indexer.Flush()
return
}
// Recovery recovery needles map from super block.
func (b *SuperBlock) Recovery(offset uint32, fn func(*needle.Needle, uint32, uint32) error) (err error) {
// WARN block may be no left data, must update block offset first
if offset == 0 {
offset = needle.NeedleOffset(_headerOffset)
}
b.Offset = offset
if err = b.Scan(b.r, offset, func(n *needle.Needle, so, eo uint32) (err1 error) {
if err1 = fn(n, so, eo); err1 == nil {
b.Offset = eo
}
return
}); err != nil {
return
}
// advise random read
// POSIX_FADV_RANDOM disables file readahead entirely.
// These changes affect the entire file, not just the specified region
// (but other open file handles to the same file are unaffected).
if err = myos.Fadvise(b.r.Fd(), 0, 0, myos.POSIX_FADV_RANDOM); err != nil {
log.Errorf("block: %s Fadvise() error(%v)", b.File)
return
}
// reset b.w offset, discard left space which can't parse to a needle
if _, err = b.w.Seek(needle.BlockOffset(b.Offset), os.SEEK_SET); err != nil {
log.Errorf("block: %s Seek() error(%v)", b.File, err)
}
return
}
// WriteAt write a needle by specified offset;
func (b *SuperBlock) WriteAt(offset uint32, n *needle.Needle) (err error) {
if b.LastErr != nil {
return b.LastErr
}
if _, err = b.w.WriteAt(n.Buffer(), needle.BlockOffset(offset)); err != nil {
b.LastErr = err
}
return
}
// Delete logical del a needls, only update the flag to it.
func (b *SuperBlock) Delete(offset uint32) (err error) {
if b.LastErr != nil {
return b.LastErr
}
// WriteAt won't update the file offset.
_, err = b.w.WriteAt(needle.FlagDelBytes, needle.BlockOffset(offset)+needle.FlagOffset)
b.LastErr = err
return
}
// ReadAt read a needle by specified offset, before call it, must set needle
// TotalSize.
func (b *SuperBlock) ReadAt(n *needle.Needle) (err error) {
if b.LastErr != nil {
return b.LastErr
}
if _, err = b.r.ReadAt(n.Buffer(), needle.BlockOffset(n.Offset)); err == nil {
err = n.Parse()
} else {
b.LastErr = err
}
return
}
func compareTestOffset(b *SuperBlock, n *needle.Needle, offset uint32) (err error) {
var v int64
if b.Offset != offset+needle.NeedleOffset(int64(n.TotalSize)) {
err = fmt.Errorf("b.Offset: %d not match %d", b.Offset, offset)
return
}
if v, err = b.w.Seek(0, os.SEEK_CUR); err != nil {
err = fmt.Errorf("b.Seek() error(%v)", err)
return
} else {
if v != needle.BlockOffset(b.Offset) {
err = fmt.Errorf("offset: %d not match", v)
return
}
}
return
}
// Recovery recovery needles map from super block.
func (b *SuperBlock) Recovery(offset uint32, fn func(*needle.Needle, uint32, uint32) error) (err error) {
var rsize int64
// WARN block may be no left data, must update block offset first
if offset == 0 {
offset = needle.NeedleOffset(_headerOffset)
}
b.Offset = offset
if err = b.Scan(b.r, offset, func(n *needle.Needle, so, eo uint32) (err1 error) {
if err1 = fn(n, so, eo); err1 == nil {
b.Offset = eo
}
return
}); err != nil {
return
}
// advise random read
// POSIX_FADV_RANDOM disables file readahead entirely.
// These changes affect the entire file, not just the specified region
// (but other open file handles to the same file are unaffected).
if err = myos.Fadvise(b.r.Fd(), 0, 0, myos.POSIX_FADV_RANDOM); err != nil {
log.Errorf("block: %s Fadvise() error(%v)", b.File)
return
}
rsize = needle.BlockOffset(b.Offset)
// reset b.w offset, discard left space which can't parse to a needle
if _, err = b.w.Seek(rsize, os.SEEK_SET); err != nil {
log.Errorf("block: %s Seek() error(%v)", b.File, err)
return
}
// recheck offset, keep size and offset consistency
if b.Size != rsize {
log.Warningf("block: %s [real size: %d, offset: %d] but [size: %d, offset: %d] not consistency, truncate file for force recovery, this may lost data",
b.File, b.Size, needle.NeedleOffset(b.Size),
rsize, b.Offset)
// truncate file
if err = b.w.Truncate(rsize); err != nil {
log.Errorf("block: %s Truncate() error(%v)", b.File, err)
}
}
return
}
// Scan scan a block file.
func (b *SuperBlock) Scan(r *os.File, offset uint32, fn func(*needle.Needle, uint32, uint32) error) (err error) {
var (
so, eo uint32
bso int64
fi os.FileInfo
fd = r.Fd()
n = new(needle.Needle)
rd = bufio.NewReaderSize(r, b.conf.Block.BufferSize)
)
if offset == 0 {
offset = needle.NeedleOffset(_headerOffset)
}
so, eo = offset, offset
bso = needle.BlockOffset(so)
// advise sequential read
if fi, err = r.Stat(); err != nil {
log.Errorf("block: %s Stat() error(%v)", b.File)
return
}
if err = myos.Fadvise(fd, bso, fi.Size(), myos.POSIX_FADV_SEQUENTIAL); err != nil {
log.Errorf("block: %s Fadvise() error(%v)", b.File)
return
}
log.Infof("scan block: %s from offset: %d", b.File, offset)
if _, err = r.Seek(bso, os.SEEK_SET); err != nil {
log.Errorf("block: %s Seek() error(%v)", b.File)
return
}
for {
if err = n.ParseFrom(rd); err != nil {
if err != io.EOF {
log.Errorf("block: parse needle from offset: %d:%d error(%v)", so, eo, err)
}
break
}
if n.TotalSize > int32(b.conf.BlockMaxSize) {
log.Errorf("scan block: %s error(%v)", n, errors.ErrNeedleSize)
err = errors.ErrNeedleSize
break
}
if log.V(1) {
log.Info(n.String())
}
eo += n.IncrOffset
if err = fn(n, so, eo); err != nil {
log.Errorf("block: callback from offset: %d:%d error(%v)", so, eo, err)
break
}
so = eo
}
if err == io.EOF {
// advise no need page cache
if err = myos.Fadvise(fd, bso, needle.BlockOffset(eo-so), myos.POSIX_FADV_DONTNEED); err != nil {
log.Errorf("block: %s Fadvise() error(%v)", b.File)
return
}
log.Infof("scan block: %s to offset: %d [ok]", b.File, eo)
err = nil
} else {
log.Infof("scan block: %s to offset: %d error(%v) [failed]", b.File, eo, err)
}
return
}