// Create new memdb and froze the old one; need external synchronization.
// newMem only called synchronously by the writer.
func (db *DB) newMem(n int) (mem *memDB, err error) {
num := db.s.allocFileNum()
file := db.s.getJournalFile(num)
w, err := file.Create()
if err != nil {
db.s.reuseFileNum(num)
return
}
db.memMu.Lock()
defer db.memMu.Unlock()
if db.frozenMem != nil {
panic("still has frozen mem")
}
if db.journal == nil {
db.journal = journal.NewWriter(w)
} else {
db.journal.Reset(w)
db.journalWriter.Close()
db.frozenJournalFile = db.journalFile
}
db.journalWriter = w
db.journalFile = file
db.frozenMem = db.mem
mdb := db.mpoolGet()
if mdb == nil || mdb.Capacity() < n {
mdb = memdb.New(db.s.icmp, maxInt(db.s.o.GetWriteBuffer(), n))
}
mem = &memDB{
db: db,
DB: mdb,
ref: 2,
}
db.mem = mem
// The seq only incremented by the writer. And whoever called newMem
// should hold write lock, so no need additional synchronization here.
db.frozenSeq = db.seq
return
}
func (db *DB) recoverJournalRO() error {
// Get all journals and sort it by file number.
allJournalFiles, err := db.s.getFiles(storage.TypeJournal)
if err != nil {
return err
}
files(allJournalFiles).sort()
// Journals that will be recovered.
var recJournalFiles []storage.File
for _, jf := range allJournalFiles {
if jf.Num() >= db.s.stJournalNum || jf.Num() == db.s.stPrevJournalNum {
recJournalFiles = append(recJournalFiles, jf)
}
}
var (
// Options.
strict = db.s.o.GetStrict(opt.StrictJournal)
checksum = db.s.o.GetStrict(opt.StrictJournalChecksum)
writeBuffer = db.s.o.GetWriteBuffer()
mdb = memdb.New(db.s.icmp, writeBuffer)
)
// Recover journals.
if len(recJournalFiles) > 0 {
db.logf("journal@recovery RO·Mode F·%d", len(recJournalFiles))
var (
jr *journal.Reader
buf = &util.Buffer{}
batch = &Batch{}
)
for _, jf := range recJournalFiles {
db.logf("journal@recovery recovering @%d", jf.Num())
fr, err := jf.Open()
if err != nil {
return err
}
// Create or reset journal reader instance.
if jr == nil {
jr = journal.NewReader(fr, dropper{db.s, jf}, strict, checksum)
} else {
jr.Reset(fr, dropper{db.s, jf}, strict, checksum)
}
// Replay journal to memdb.
for {
r, err := jr.Next()
if err != nil {
if err == io.EOF {
break
}
fr.Close()
return errors.SetFile(err, jf)
}
buf.Reset()
if _, err := buf.ReadFrom(r); err != nil {
if err == io.ErrUnexpectedEOF {
// This is error returned due to corruption, with strict == false.
continue
}
fr.Close()
return errors.SetFile(err, jf)
}
if err := batch.memDecodeAndReplay(db.seq, buf.Bytes(), mdb); err != nil {
if !strict && errors.IsCorrupted(err) {
db.s.logf("journal error: %v (skipped)", err)
// We won't apply sequence number as it might be corrupted.
continue
}
fr.Close()
return errors.SetFile(err, jf)
}
// Save sequence number.
db.seq = batch.seq + uint64(batch.Len())
}
fr.Close()
}
}
// Set memDB.
db.mem = &memDB{db: db, DB: mdb, ref: 1}
return nil
}
func (db *DB) recoverJournal() error {
// Get all journals and sort it by file number.
allJournalFiles, err := db.s.getFiles(storage.TypeJournal)
if err != nil {
return err
}
files(allJournalFiles).sort()
// Journals that will be recovered.
var recJournalFiles []storage.File
for _, jf := range allJournalFiles {
if jf.Num() >= db.s.stJournalNum || jf.Num() == db.s.stPrevJournalNum {
recJournalFiles = append(recJournalFiles, jf)
}
}
var (
of storage.File // Obsolete file.
rec = &sessionRecord{}
)
// Recover journals.
if len(recJournalFiles) > 0 {
db.logf("journal@recovery F·%d", len(recJournalFiles))
// Mark file number as used.
db.s.markFileNum(recJournalFiles[len(recJournalFiles)-1].Num())
var (
// Options.
strict = db.s.o.GetStrict(opt.StrictJournal)
checksum = db.s.o.GetStrict(opt.StrictJournalChecksum)
writeBuffer = db.s.o.GetWriteBuffer()
jr *journal.Reader
mdb = memdb.New(db.s.icmp, writeBuffer)
buf = &util.Buffer{}
batch = &Batch{}
)
for _, jf := range recJournalFiles {
db.logf("journal@recovery recovering @%d", jf.Num())
fr, err := jf.Open()
if err != nil {
return err
}
// Create or reset journal reader instance.
if jr == nil {
jr = journal.NewReader(fr, dropper{db.s, jf}, strict, checksum)
} else {
jr.Reset(fr, dropper{db.s, jf}, strict, checksum)
}
// Flush memdb and remove obsolete journal file.
if of != nil {
if mdb.Len() > 0 {
if _, err := db.s.flushMemdb(rec, mdb, -1); err != nil {
fr.Close()
return err
}
}
rec.setJournalNum(jf.Num())
rec.setSeqNum(db.seq)
if err := db.s.commit(rec); err != nil {
fr.Close()
return err
}
rec.resetAddedTables()
of.Remove()
of = nil
}
// Replay journal to memdb.
mdb.Reset()
for {
r, err := jr.Next()
if err != nil {
if err == io.EOF {
break
}
fr.Close()
return errors.SetFile(err, jf)
}
buf.Reset()
if _, err := buf.ReadFrom(r); err != nil {
if err == io.ErrUnexpectedEOF {
// This is error returned due to corruption, with strict == false.
continue
}
fr.Close()
return errors.SetFile(err, jf)
}
if err := batch.memDecodeAndReplay(db.seq, buf.Bytes(), mdb); err != nil {
if !strict && errors.IsCorrupted(err) {
db.s.logf("journal error: %v (skipped)", err)
// We won't apply sequence number as it might be corrupted.
continue
}
fr.Close()
return errors.SetFile(err, jf)
}
// Save sequence number.
db.seq = batch.seq + uint64(batch.Len())
// Flush it if large enough.
if mdb.Size() >= writeBuffer {
if _, err := db.s.flushMemdb(rec, mdb, 0); err != nil {
fr.Close()
return err
}
mdb.Reset()
}
}
fr.Close()
of = jf
}
// Flush the last memdb.
if mdb.Len() > 0 {
if _, err := db.s.flushMemdb(rec, mdb, 0); err != nil {
return err
}
}
}
// Create a new journal.
if _, err := db.newMem(0); err != nil {
return err
}
// Commit.
rec.setJournalNum(db.journalFile.Num())
rec.setSeqNum(db.seq)
if err := db.s.commit(rec); err != nil {
// Close journal on error.
if db.journal != nil {
db.journal.Close()
db.journalWriter.Close()
}
return err
}
// Remove the last obsolete journal file.
if of != nil {
of.Remove()
}
return nil
}