// Creates an empty table and returns table writer.
func (t *tOps) create() (*tWriter, error) {
file := t.s.getTableFile(t.s.allocFileNum())
fw, err := file.Create()
if err != nil {
return nil, err
}
return &tWriter{
t: t,
file: file,
w: fw,
tw: table.NewWriter(fw, t.s.o),
}, nil
}
func recoverTable(s *session, o *opt.Options) error {
// Get all tables and sort it by file number.
tableFiles_, err := s.getFiles(storage.TypeTable)
if err != nil {
return err
}
tableFiles := files(tableFiles_)
tableFiles.sort()
var mSeq uint64
var good, corrupted int
rec := new(sessionRecord)
bpool := util.NewBufferPool(o.GetBlockSize() + 5)
buildTable := func(iter iterator.Iterator) (tmp storage.File, size int64, err error) {
tmp = s.newTemp()
writer, err := tmp.Create()
if err != nil {
return
}
defer func() {
writer.Close()
if err != nil {
tmp.Remove()
tmp = nil
}
}()
// Copy entries.
tw := table.NewWriter(writer, o)
for iter.Next() {
key := iter.Key()
if validIkey(key) {
err = tw.Append(key, iter.Value())
if err != nil {
return
}
}
}
err = iter.Error()
if err != nil {
return
}
err = tw.Close()
if err != nil {
return
}
err = writer.Sync()
if err != nil {
return
}
size = int64(tw.BytesLen())
return
}
recoverTable := func(file storage.File) error {
s.logf("table@recovery recovering @%d", file.Num())
reader, err := file.Open()
if err != nil {
return err
}
defer reader.Close()
// Get file size.
size, err := reader.Seek(0, 2)
if err != nil {
return err
}
var tSeq uint64
var tgood, tcorrupted, blockerr int
var imin, imax []byte
tr := table.NewReader(reader, size, nil, bpool, o)
iter := tr.NewIterator(nil, nil)
iter.(iterator.ErrorCallbackSetter).SetErrorCallback(func(err error) {
s.logf("table@recovery found error @%d %q", file.Num(), err)
blockerr++
})
// Scan the table.
for iter.Next() {
key := iter.Key()
_, seq, _, ok := parseIkey(key)
if !ok {
tcorrupted++
continue
}
tgood++
if seq > tSeq {
tSeq = seq
}
if imin == nil {
imin = append([]byte{}, key...)
}
imax = append(imax[:0], key...)
}
if err := iter.Error(); err != nil {
iter.Release()
return err
}
iter.Release()
if tgood > 0 {
if tcorrupted > 0 || blockerr > 0 {
// Rebuild the table.
s.logf("table@recovery rebuilding @%d", file.Num())
iter := tr.NewIterator(nil, nil)
tmp, newSize, err := buildTable(iter)
iter.Release()
if err != nil {
return err
}
reader.Close()
if err := file.Replace(tmp); err != nil {
return err
}
size = newSize
}
if tSeq > mSeq {
mSeq = tSeq
}
// Add table to level 0.
rec.addTable(0, file.Num(), uint64(size), imin, imax)
s.logf("table@recovery recovered @%d N·%d C·%d B·%d S·%d Q·%d", file.Num(), tgood, tcorrupted, blockerr, size, tSeq)
} else {
s.logf("table@recovery unrecoverable @%d C·%d B·%d S·%d", file.Num(), tcorrupted, blockerr, size)
}
good += tgood
corrupted += tcorrupted
return nil
}
// Recover all tables.
if len(tableFiles) > 0 {
s.logf("table@recovery F·%d", len(tableFiles))
// Mark file number as used.
s.markFileNum(tableFiles[len(tableFiles)-1].Num())
for _, file := range tableFiles {
if err := recoverTable(file); err != nil {
return err
}
}
s.logf("table@recovery recovered F·%d N·%d C·%d Q·%d", len(tableFiles), good, corrupted, mSeq)
}
// Set sequence number.
rec.setSeq(mSeq + 1)
// Create new manifest.
if err := s.create(); err != nil {
return err
}
// Commit.
return s.commit(rec)
}