// Create new memdb and froze the old one; need external synchronization.
func (d *DB) newMem() (m *memdb.DB, err error) {
s := d.s
num := s.allocFileNum()
w, err := newJournalWriter(s.getJournalFile(num))
if err != nil {
s.reuseFileNum(num)
return
}
old := d.journal
d.journal = w
if old != nil {
old.close()
d.fjournal = old
}
d.fseq = d.seq
m = memdb.New(s.cmp)
mem := &memSet{cur: m}
if old := d.getMem_NB(); old != nil {
mem.froze = old.cur
}
atomic.StorePointer(&d.mem, unsafe.Pointer(mem))
return
}
// Create new memdb and froze the old one; need external synchronization.
// newMem only called synchronously by the writer.
func (d *DB) newMem() (mem *memdb.DB, err error) {
s := d.s
num := s.allocFileNum()
file := s.getJournalFile(num)
w, err := file.Create()
if err != nil {
s.reuseFileNum(num)
return
}
d.memMu.Lock()
if d.journal == nil {
d.journal = journal.NewWriter(w)
} else {
d.journal.Reset(w)
d.journalWriter.Close()
d.frozenJournalFile = d.journalFile
}
d.journalWriter = w
d.journalFile = file
d.frozenMem = d.mem
d.mem = memdb.New(s.cmp, toPercent(d.s.o.GetWriteBuffer(), kWriteBufferPercent))
mem = d.mem
// The seq only incremented by the writer.
d.frozenSeq = d.seq
d.memMu.Unlock()
return
}
// 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.DB, 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.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
db.mem = memdb.New(db.s.icmp, maxInt(db.s.o.GetWriteBuffer(), n))
mem = db.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 (mc *stConstructor_MergedMemDB) init(t *testing.T, ho *stHarnessOpt) error {
ho.Randomize = true
mc.t = t
for i := range mc.db {
mc.db[i] = memdb.New(comparer.DefaultComparer, 0)
}
return nil
}
func (p *stConstructor_MergedMemDB) init(t *testing.T, ho *stHarnessOpt) error {
ho.Randomize = true
p.t = t
for i := range p.mem {
p.mem[i] = memdb.New(comparer.BytesComparer{})
}
return nil
}
func (db *DB) mpoolGet(n int) *memDB {
var mdb *memdb.DB
select {
case mdb = <-db.memPool:
default:
}
if mdb == nil || mdb.Capacity() < n {
mdb = memdb.New(db.s.icmp, maxInt(db.s.o.GetWriteBuffer(), n))
}
return &memDB{
db: db,
DB: mdb,
}
}
func TestBatch_Size(t *testing.T) {
b := new(Batch)
for i := 0; i < 2; i++ {
b.Put([]byte("key1"), []byte("value1"))
b.Put([]byte("key2"), []byte("value2"))
b.Delete([]byte("key1"))
b.Put([]byte("foo"), []byte("foovalue"))
b.Put([]byte("bar"), []byte("barvalue"))
mem := memdb.New(&iComparer{comparer.DefaultComparer}, 0)
b.memReplay(mem)
if b.size() != mem.Size() {
t.Errorf("invalid batch size calculation, want=%d got=%d", mem.Size(), b.size())
}
b.Reset()
}
}
// 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
mem, ok := db.memPool.Get().(*memDB)
if ok && mem.db.Capacity() >= n {
mem.db.Reset()
mem.incref()
} else {
mem = &memDB{
pool: db.memPool,
db: memdb.New(db.s.icmp, maxInt(db.s.o.GetWriteBuffer(), n)),
ref: 1,
}
}
mem.incref()
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
}
// 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) {
fd := storage.FileDesc{Type: storage.TypeJournal, Num: db.s.allocFileNum()}
w, err := db.s.stor.Create(fd)
if err != nil {
db.s.reuseFileNum(fd.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.frozenJournalFd = db.journalFd
}
db.journalWriter = w
db.journalFd = fd
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 (d *DB) recoverLog() (err error) {
s := d.s
icmp := s.cmp
s.printf("LogRecovery: started, min=%d", s.stLogNum)
var mem *memdb.DB
batch := new(Batch)
cm := newCMem(s)
logs, skip := files(s.getFiles(desc.TypeLog)), 0
logs.sort()
for _, log := range logs {
if log.Num() < s.stLogNum {
skip++
continue
}
s.markFileNum(log.Num())
}
var r, fr *logReader
for _, log := range logs[skip:] {
s.printf("LogRecovery: recovering, num=%d", log.Num())
r, err = newLogReader(log, true, s.logDropFunc("log", log.Num()))
if err != nil {
return
}
if mem != nil {
if mem.Len() > 0 {
err = cm.flush(mem, 0)
if err != nil {
return
}
}
err = cm.commit(r.file.Num(), d.seq)
if err != nil {
return
}
cm.reset()
fr.remove()
fr = nil
}
mem = memdb.New(icmp)
for r.log.Next() {
err = batch.decode(r.log.Record())
if err != nil {
return
}
err = batch.memReplay(mem)
if err != nil {
return
}
d.seq = batch.seq + uint64(batch.len())
if mem.Size() > s.o.GetWriteBuffer() {
// flush to table
err = cm.flush(mem, 0)
if err != nil {
return
}
// create new memdb
mem = memdb.New(icmp)
}
}
err = r.log.Error()
if err != nil {
return
}
r.close()
fr = r
}
// create new log
_, err = d.newMem()
if err != nil {
return
}
if mem != nil && mem.Len() > 0 {
err = cm.flush(mem, 0)
if err != nil {
return
}
}
err = cm.commit(d.log.file.Num(), d.seq)
if err != nil {
return
}
if fr != nil {
fr.remove()
}
return
}
// NewMemDbBuffer creates a new memDbBuffer.
func NewMemDbBuffer() MemBuffer {
return &memDbBuffer{db: memdb.New(comparer.DefaultComparer, 4*1024)}
}
func (mc *stConstructor_MemDB) init(t *testing.T, ho *stHarnessOpt) error {
ho.Randomize = true
mc.t = t
mc.db = memdb.New(comparer.DefaultComparer, 0)
return nil
}
func (p *stConstructor_MemDB) init(t *testing.T, ho *stHarnessOpt) error {
ho.Randomize = true
p.t = t
p.mem = memdb.New(comparer.BytesComparer{})
return nil
}
// conditionType is the type for condition consts.
type conditionType int
const (
// conditionIfNotExist means the condition doesn't exist.
conditionIfNotExist conditionType = iota + 1
// conditionIfEqual means the condition is equals.
conditionIfEqual
// conditionForceSet means the condition is force set.
conditionForceSet
)
var (
p = pool.NewCache("memdb pool", 100, func() interface{} {
return memdb.New(comparer.DefaultComparer, 1*1024*1024)
})
)
// conditionValue is a data structure used to store current stored data and data verification condition.
type conditionValue struct {
originValue []byte
condition conditionType
}
// IsErrNotFound checks if err is a kind of NotFound error.
func IsErrNotFound(err error) bool {
if errors2.ErrorEqual(err, leveldb.ErrNotFound) || errors2.ErrorEqual(err, ErrNotExist) {
return true
}
func (db *DB) recoverJournalRO() error {
// Get all journals and sort it by file number.
rawFds, err := db.s.stor.List(storage.TypeJournal)
if err != nil {
return err
}
sortFds(rawFds)
// Journals that will be recovered.
var fds []storage.FileDesc
for _, fd := range rawFds {
if fd.Num >= db.s.stJournalNum || fd.Num == db.s.stPrevJournalNum {
fds = append(fds, fd)
}
}
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(fds) > 0 {
db.logf("journal@recovery RO·Mode F·%d", len(fds))
var (
jr *journal.Reader
buf = &util.Buffer{}
batch = &Batch{}
)
for _, fd := range fds {
db.logf("journal@recovery recovering @%d", fd.Num)
fr, err := db.s.stor.Open(fd)
if err != nil {
return err
}
// Create or reset journal reader instance.
if jr == nil {
jr = journal.NewReader(fr, dropper{db.s, fd}, strict, checksum)
} else {
jr.Reset(fr, dropper{db.s, fd}, strict, checksum)
}
// Replay journal to memdb.
for {
r, err := jr.Next()
if err != nil {
if err == io.EOF {
break
}
fr.Close()
return errors.SetFd(err, fd)
}
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.SetFd(err, fd)
}
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.SetFd(err, fd)
}
// 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 (d *DB) recoverJournal() (err error) {
s := d.s
icmp := s.cmp
s.printf("JournalRecovery: started, min=%d", s.stJournalNum)
var mem *memdb.DB
batch := new(Batch)
cm := newCMem(s)
journals := files(s.getFiles(storage.TypeJournal))
journals.sort()
rJournals := make([]storage.File, 0, len(journals))
for _, journal := range journals {
if journal.Num() >= s.stJournalNum || journal.Num() == s.stPrevJournalNum {
s.markFileNum(journal.Num())
rJournals = append(rJournals, journal)
}
}
var r, fr *journalReader
for _, journal := range rJournals {
s.printf("JournalRecovery: recovering, num=%d", journal.Num())
r, err = newJournalReader(journal, true, s.journalDropFunc("journal", journal.Num()))
if err != nil {
return
}
if mem != nil {
if mem.Len() > 0 {
err = cm.flush(mem, 0)
if err != nil {
return
}
}
err = cm.commit(r.file.Num(), d.seq)
if err != nil {
return
}
cm.reset()
fr.remove()
fr = nil
}
mem = memdb.New(icmp)
for r.journal.Next() {
err = batch.decode(r.journal.Record())
if err != nil {
return
}
err = batch.memReplay(mem)
if err != nil {
return
}
d.seq = batch.seq + uint64(batch.len())
if mem.Size() > s.o.GetWriteBuffer() {
// flush to table
err = cm.flush(mem, 0)
if err != nil {
return
}
// create new memdb
mem = memdb.New(icmp)
}
}
err = r.journal.Error()
if err != nil {
return
}
r.close()
fr = r
}
// create new journal
_, err = d.newMem()
if err != nil {
return
}
if mem != nil && mem.Len() > 0 {
err = cm.flush(mem, 0)
if err != nil {
return
}
}
err = cm.commit(d.journal.file.Num(), d.seq)
if err != nil {
return
}
if fr != nil {
fr.remove()
}
return
}
// NewMemoryKeyValue returns a KeyValue implementation that's backed only
// by memory. It's mostly useful for tests and development.
func NewMemoryKeyValue() KeyValue {
db := memdb.New(comparer.DefaultComparer, 128)
return &memKeys{db: db}
}
func (d *DB) recoverJournal() error {
s := d.s
ff0, err := s.getFiles(storage.TypeJournal)
if err != nil {
return err
}
ff1 := files(ff0)
ff1.sort()
ff2 := make([]storage.File, 0, len(ff1))
for _, file := range ff1 {
if file.Num() >= s.stJournalNum || file.Num() == s.stPrevJournalNum {
s.markFileNum(file.Num())
ff2 = append(ff2, file)
}
}
var jr *journal.Reader
var of storage.File
var mem *memdb.DB
batch := new(Batch)
cm := newCMem(s)
buf := new(util.Buffer)
// Options.
strict := s.o.GetStrict(opt.StrictJournal)
checksum := s.o.GetStrict(opt.StrictJournalChecksum)
writeBuffer := s.o.GetWriteBuffer()
recoverJournal := func(file storage.File) error {
s.logf("journal@recovery recovering @%d", file.Num())
reader, err := file.Open()
if err != nil {
return err
}
defer reader.Close()
if jr == nil {
jr = journal.NewReader(reader, dropper{s, file}, strict, checksum)
} else {
jr.Reset(reader, dropper{s, file}, strict, checksum)
}
if of != nil {
if mem.Len() > 0 {
if err := cm.flush(mem, 0); err != nil {
return err
}
}
if err := cm.commit(file.Num(), d.seq); err != nil {
return err
}
cm.reset()
of.Remove()
of = nil
}
// Reset memdb.
mem.Reset()
for {
r, err := jr.Next()
if err != nil {
if err == io.EOF {
break
}
return err
}
buf.Reset()
if _, err := buf.ReadFrom(r); err != nil {
if strict {
return err
}
continue
}
if err := batch.decode(buf.Bytes()); err != nil {
return err
}
if err := batch.memReplay(mem); err != nil {
return err
}
d.seq = batch.seq + uint64(batch.len())
if mem.Size() >= writeBuffer {
// Large enough, flush it.
if err := cm.flush(mem, 0); err != nil {
return err
}
// Reset memdb.
mem.Reset()
}
}
of = file
return nil
}
// Recover all journals.
if len(ff2) > 0 {
s.logf("journal@recovery F·%d", len(ff2))
mem = memdb.New(s.icmp, writeBuffer)
for _, file := range ff2 {
if err := recoverJournal(file); err != nil {
return err
}
}
// Flush the last journal.
if mem.Len() > 0 {
if err := cm.flush(mem, 0); err != nil {
return err
}
}
}
// Create a new journal.
if _, err := d.newMem(0); err != nil {
return err
}
// Commit.
if err := cm.commit(d.journalFile.Num(), d.seq); err != nil {
return err
}
// Remove the last journal.
if of != nil {
of.Remove()
}
return nil
}
func (db *DB) recoverJournal() error {
// Get all tables and sort it by file number.
journalFiles_, err := db.s.getFiles(storage.TypeJournal)
if err != nil {
return err
}
journalFiles := files(journalFiles_)
journalFiles.sort()
// Discard older journal.
prev := -1
for i, file := range journalFiles {
if file.Num() >= db.s.stJournalNum {
if prev >= 0 {
i--
journalFiles[i] = journalFiles[prev]
}
journalFiles = journalFiles[i:]
break
} else if file.Num() == db.s.stPrevJournalNum {
prev = i
}
}
var jr *journal.Reader
var of storage.File
var mem *memdb.DB
batch := new(Batch)
cm := newCMem(db.s)
buf := new(util.Buffer)
// Options.
strict := db.s.o.GetStrict(opt.StrictJournal)
checksum := db.s.o.GetStrict(opt.StrictJournalChecksum)
writeBuffer := db.s.o.GetWriteBuffer()
recoverJournal := func(file storage.File) error {
db.logf("journal@recovery recovering @%d", file.Num())
reader, err := file.Open()
if err != nil {
return err
}
defer reader.Close()
// Create/reset journal reader instance.
if jr == nil {
jr = journal.NewReader(reader, dropper{db.s, file}, strict, checksum)
} else {
jr.Reset(reader, dropper{db.s, file}, strict, checksum)
}
// Flush memdb and remove obsolete journal file.
if of != nil {
if mem.Len() > 0 {
if err := cm.flush(mem, 0); err != nil {
return err
}
}
if err := cm.commit(file.Num(), db.seq); err != nil {
return err
}
cm.reset()
of.Remove()
of = nil
}
// Replay journal to memdb.
mem.Reset()
for {
r, err := jr.Next()
if err != nil {
if err == io.EOF {
break
}
return err
}
buf.Reset()
if _, err := buf.ReadFrom(r); err != nil {
if err == io.ErrUnexpectedEOF {
continue
} else {
return err
}
}
if err := batch.decode(buf.Bytes()); err != nil {
return err
}
if err := batch.memReplay(mem); err != nil {
return err
}
// Save sequence number.
db.seq = batch.seq + uint64(batch.len())
// Flush it if large enough.
if mem.Size() >= writeBuffer {
if err := cm.flush(mem, 0); err != nil {
return err
}
mem.Reset()
}
}
of = file
return nil
}
// Recover all journals.
if len(journalFiles) > 0 {
db.logf("journal@recovery F·%d", len(journalFiles))
// Mark file number as used.
db.s.markFileNum(journalFiles[len(journalFiles)-1].Num())
mem = memdb.New(db.s.icmp, writeBuffer)
for _, file := range journalFiles {
if err := recoverJournal(file); err != nil {
return err
}
}
// Flush the last journal.
if mem.Len() > 0 {
if err := cm.flush(mem, 0); err != nil {
return err
}
}
}
// Create a new journal.
if _, err := db.newMem(0); err != nil {
return err
}
// Commit.
if err := cm.commit(db.journalFile.Num(), db.seq); err != nil {
// Close journal.
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
}
func (db *DB) recoverJournal() error {
// Get all journals and sort it by file number.
rawFds, err := db.s.stor.List(storage.TypeJournal)
if err != nil {
return err
}
sortFds(rawFds)
// Journals that will be recovered.
var fds []storage.FileDesc
for _, fd := range rawFds {
if fd.Num >= db.s.stJournalNum || fd.Num == db.s.stPrevJournalNum {
fds = append(fds, fd)
}
}
var (
ofd storage.FileDesc // Obsolete file.
rec = &sessionRecord{}
)
// Recover journals.
if len(fds) > 0 {
db.logf("journal@recovery F·%d", len(fds))
// Mark file number as used.
db.s.markFileNum(fds[len(fds)-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 _, fd := range fds {
db.logf("journal@recovery recovering @%d", fd.Num)
fr, err := db.s.stor.Open(fd)
if err != nil {
return err
}
// Create or reset journal reader instance.
if jr == nil {
jr = journal.NewReader(fr, dropper{db.s, fd}, strict, checksum)
} else {
jr.Reset(fr, dropper{db.s, fd}, strict, checksum)
}
// Flush memdb and remove obsolete journal file.
if !ofd.Nil() {
if mdb.Len() > 0 {
if _, err := db.s.flushMemdb(rec, mdb, 0); err != nil {
fr.Close()
return err
}
}
rec.setJournalNum(fd.Num)
rec.setSeqNum(db.seq)
if err := db.s.commit(rec); err != nil {
fr.Close()
return err
}
rec.resetAddedTables()
db.s.stor.Remove(ofd)
ofd = storage.FileDesc{}
}
// Replay journal to memdb.
mdb.Reset()
for {
r, err := jr.Next()
if err != nil {
if err == io.EOF {
break
}
fr.Close()
return errors.SetFd(err, fd)
}
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.SetFd(err, fd)
}
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.SetFd(err, fd)
}
// 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()
ofd = fd
}
// 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.journalFd.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 !ofd.Nil() {
db.s.stor.Remove(ofd)
}
return nil
}