Files
tidb/session/txn.go
tiancaiamao d587c63062 *: some tiny improvements for the large transaction (#13815)
1. Set the default txn size limit to 1G
2. Reduce memory footprint during insert for the large txn
3. Recycle the session buffer memory to avoid leak
4. Set rate limit to avoid 2PC prewrite overloads the TiKV
5. Start ttlManager for the large txn
2019-12-03 00:40:40 +08:00

491 lines
13 KiB
Go
Executable File

// Copyright 2018 PingCAP, Inc.
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// See the License for the specific language governing permissions and
// limitations under the License.
package session
import (
"bytes"
"context"
"fmt"
"strings"
"sync/atomic"
"github.com/opentracing/opentracing-go"
"github.com/pingcap/errors"
"github.com/pingcap/failpoint"
"github.com/pingcap/tidb/executor"
"github.com/pingcap/tidb/kv"
"github.com/pingcap/tidb/sessionctx"
"github.com/pingcap/tidb/sessionctx/binloginfo"
"github.com/pingcap/tidb/store/tikv/oracle"
"github.com/pingcap/tidb/table"
"github.com/pingcap/tidb/tablecodec"
"github.com/pingcap/tidb/util/logutil"
"github.com/pingcap/tipb/go-binlog"
"go.uber.org/zap"
)
// TxnState wraps kv.Transaction to provide a new kv.Transaction.
// 1. It holds all statement related modification in the buffer before flush to the txn,
// so if execute statement meets error, the txn won't be made dirty.
// 2. It's a lazy transaction, that means it's a txnFuture before StartTS() is really need.
type TxnState struct {
// States of a TxnState should be one of the followings:
// Invalid: kv.Transaction == nil && txnFuture == nil
// Pending: kv.Transaction == nil && txnFuture != nil
// Valid: kv.Transaction != nil && txnFuture == nil
kv.Transaction
txnFuture *txnFuture
buf kv.MemBuffer
mutations map[int64]*binlog.TableMutation
dirtyTableOP []dirtyTableOperation
// If doNotCommit is not nil, Commit() will not commit the transaction.
// doNotCommit flag may be set when StmtCommit fail.
doNotCommit error
}
func (st *TxnState) init() {
st.buf = kv.NewMemDbBuffer(kv.DefaultTxnMembufCap)
st.mutations = make(map[int64]*binlog.TableMutation)
}
// Valid implements the kv.Transaction interface.
func (st *TxnState) Valid() bool {
return st.Transaction != nil && st.Transaction.Valid()
}
func (st *TxnState) pending() bool {
return st.Transaction == nil && st.txnFuture != nil
}
func (st *TxnState) validOrPending() bool {
return st.txnFuture != nil || st.Valid()
}
func (st *TxnState) String() string {
if st.Transaction != nil {
return st.Transaction.String()
}
if st.txnFuture != nil {
return "txnFuture"
}
return "invalid transaction"
}
// GoString implements the "%#v" format for fmt.Printf.
func (st *TxnState) GoString() string {
var s strings.Builder
s.WriteString("Txn{")
if st.pending() {
s.WriteString("state=pending")
} else if st.Valid() {
s.WriteString("state=valid")
fmt.Fprintf(&s, ", txnStartTS=%d", st.Transaction.StartTS())
if len(st.dirtyTableOP) > 0 {
fmt.Fprintf(&s, ", len(dirtyTable)=%d, %#v", len(st.dirtyTableOP), st.dirtyTableOP)
}
if len(st.mutations) > 0 {
fmt.Fprintf(&s, ", len(mutations)=%d, %#v", len(st.mutations), st.mutations)
}
if st.buf != nil && st.buf.Len() != 0 {
fmt.Fprintf(&s, ", buf.length: %d, buf.size: %d", st.buf.Len(), st.buf.Size())
}
} else {
s.WriteString("state=invalid")
}
s.WriteString("}")
return s.String()
}
func (st *TxnState) changeInvalidToValid(txn kv.Transaction) {
st.Transaction = txn
st.txnFuture = nil
}
func (st *TxnState) changeInvalidToPending(future *txnFuture) {
st.Transaction = nil
st.txnFuture = future
}
func (st *TxnState) changePendingToValid(txnCap int) error {
if st.txnFuture == nil {
return errors.New("transaction future is not set")
}
future := st.txnFuture
st.txnFuture = nil
txn, err := future.wait()
if err != nil {
st.Transaction = nil
return err
}
txn.SetCap(txnCap)
st.Transaction = txn
return nil
}
func (st *TxnState) changeToInvalid() {
st.Transaction = nil
st.txnFuture = nil
}
// dirtyTableOperation represents an operation to dirtyTable, we log the operation
// first and apply the operation log when statement commit.
type dirtyTableOperation struct {
kind int
tid int64
handle int64
}
var hasMockAutoIDRetry = int64(0)
func enableMockAutoIDRetry() {
atomic.StoreInt64(&hasMockAutoIDRetry, 1)
}
func mockAutoIDRetry() bool {
return atomic.LoadInt64(&hasMockAutoIDRetry) == 1
}
// Commit overrides the Transaction interface.
func (st *TxnState) Commit(ctx context.Context) error {
defer st.reset()
if len(st.mutations) != 0 || len(st.dirtyTableOP) != 0 || st.buf.Len() != 0 {
logutil.BgLogger().Error("the code should never run here",
zap.String("TxnState", st.GoString()),
zap.Stack("something must be wrong"))
return errors.New("invalid transaction")
}
if st.doNotCommit != nil {
if err1 := st.Transaction.Rollback(); err1 != nil {
logutil.BgLogger().Error("rollback error", zap.Error(err1))
}
return errors.Trace(st.doNotCommit)
}
// mockCommitError8942 is used for PR #8942.
failpoint.Inject("mockCommitError8942", func(val failpoint.Value) {
if val.(bool) {
failpoint.Return(kv.ErrTxnRetryable)
}
})
// mockCommitRetryForAutoID is used to mock an commit retry for adjustAutoIncrementDatum.
failpoint.Inject("mockCommitRetryForAutoID", func(val failpoint.Value) {
if val.(bool) && !mockAutoIDRetry() {
enableMockAutoIDRetry()
failpoint.Return(kv.ErrTxnRetryable)
}
})
return st.Transaction.Commit(ctx)
}
// Rollback overrides the Transaction interface.
func (st *TxnState) Rollback() error {
defer st.reset()
return st.Transaction.Rollback()
}
func (st *TxnState) reset() {
st.doNotCommit = nil
st.cleanup()
st.changeToInvalid()
}
// Get overrides the Transaction interface.
func (st *TxnState) Get(ctx context.Context, k kv.Key) ([]byte, error) {
val, err := st.buf.Get(ctx, k)
if kv.IsErrNotFound(err) {
val, err = st.Transaction.Get(ctx, k)
if kv.IsErrNotFound(err) {
return nil, err
}
}
if err != nil {
return nil, err
}
if len(val) == 0 {
return nil, kv.ErrNotExist
}
return val, nil
}
// BatchGet overrides the Transaction interface.
func (st *TxnState) BatchGet(ctx context.Context, keys []kv.Key) (map[string][]byte, error) {
bufferValues := make([][]byte, len(keys))
shrinkKeys := make([]kv.Key, 0, len(keys))
for i, key := range keys {
val, err := st.buf.Get(ctx, key)
if kv.IsErrNotFound(err) {
shrinkKeys = append(shrinkKeys, key)
continue
}
if err != nil {
return nil, err
}
if len(val) != 0 {
bufferValues[i] = val
}
}
storageValues, err := st.Transaction.BatchGet(ctx, shrinkKeys)
if err != nil {
return nil, err
}
for i, key := range keys {
if bufferValues[i] == nil {
continue
}
storageValues[string(key)] = bufferValues[i]
}
return storageValues, nil
}
// Set overrides the Transaction interface.
func (st *TxnState) Set(k kv.Key, v []byte) error {
return st.buf.Set(k, v)
}
// Delete overrides the Transaction interface.
func (st *TxnState) Delete(k kv.Key) error {
return st.buf.Delete(k)
}
// Iter overrides the Transaction interface.
func (st *TxnState) Iter(k kv.Key, upperBound kv.Key) (kv.Iterator, error) {
bufferIt, err := st.buf.Iter(k, upperBound)
if err != nil {
return nil, err
}
retrieverIt, err := st.Transaction.Iter(k, upperBound)
if err != nil {
return nil, err
}
return kv.NewUnionIter(bufferIt, retrieverIt, false)
}
// IterReverse overrides the Transaction interface.
func (st *TxnState) IterReverse(k kv.Key) (kv.Iterator, error) {
bufferIt, err := st.buf.IterReverse(k)
if err != nil {
return nil, err
}
retrieverIt, err := st.Transaction.IterReverse(k)
if err != nil {
return nil, err
}
return kv.NewUnionIter(bufferIt, retrieverIt, true)
}
func (st *TxnState) cleanup() {
const sz4M = 4 << 20
if st.buf.Size() > sz4M {
// The memory footprint for the large transaction could be huge here.
// Each active session has its own buffer, we should free the buffer to
// avoid memory leak.
st.buf = kv.NewMemDbBuffer(kv.DefaultTxnMembufCap)
} else {
st.buf.Reset()
}
for key := range st.mutations {
delete(st.mutations, key)
}
if st.dirtyTableOP != nil {
empty := dirtyTableOperation{}
for i := 0; i < len(st.dirtyTableOP); i++ {
st.dirtyTableOP[i] = empty
}
if len(st.dirtyTableOP) > 256 {
// Reduce memory footprint for the large transaction.
st.dirtyTableOP = nil
} else {
st.dirtyTableOP = st.dirtyTableOP[:0]
}
}
}
// KeysNeedToLock returns the keys need to be locked.
func (st *TxnState) KeysNeedToLock() ([]kv.Key, error) {
keys := make([]kv.Key, 0, st.buf.Len())
if err := kv.WalkMemBuffer(st.buf, func(k kv.Key, v []byte) error {
if !keyNeedToLock(k, v) {
return nil
}
// If the key is already locked, it will be deduplicated in LockKeys method later.
// The statement MemBuffer will be reused, so we must copy the key here.
keys = append(keys, append([]byte{}, k...))
return nil
}); err != nil {
return nil, err
}
return keys, nil
}
func keyNeedToLock(k, v []byte) bool {
isTableKey := bytes.HasPrefix(k, tablecodec.TablePrefix())
if !isTableKey {
// meta key always need to lock.
return true
}
isDelete := len(v) == 0
if isDelete {
// only need to delete row key.
return k[10] == 'r'
}
if tablecodec.IsUntouchedIndexKValue(k, v) {
return false
}
isNonUniqueIndex := len(v) == 1
// Put row key and unique index need to lock.
return !isNonUniqueIndex
}
func getBinlogMutation(ctx sessionctx.Context, tableID int64) *binlog.TableMutation {
bin := binloginfo.GetPrewriteValue(ctx, true)
for i := range bin.Mutations {
if bin.Mutations[i].TableId == tableID {
return &bin.Mutations[i]
}
}
idx := len(bin.Mutations)
bin.Mutations = append(bin.Mutations, binlog.TableMutation{TableId: tableID})
return &bin.Mutations[idx]
}
func mergeToMutation(m1, m2 *binlog.TableMutation) {
m1.InsertedRows = append(m1.InsertedRows, m2.InsertedRows...)
m1.UpdatedRows = append(m1.UpdatedRows, m2.UpdatedRows...)
m1.DeletedIds = append(m1.DeletedIds, m2.DeletedIds...)
m1.DeletedPks = append(m1.DeletedPks, m2.DeletedPks...)
m1.DeletedRows = append(m1.DeletedRows, m2.DeletedRows...)
m1.Sequence = append(m1.Sequence, m2.Sequence...)
}
func mergeToDirtyDB(dirtyDB *executor.DirtyDB, op dirtyTableOperation) {
dt := dirtyDB.GetDirtyTable(op.tid)
switch op.kind {
case table.DirtyTableAddRow:
dt.AddRow(op.handle)
case table.DirtyTableDeleteRow:
dt.DeleteRow(op.handle)
}
}
type txnFailFuture struct{}
func (txnFailFuture) Wait() (uint64, error) {
return 0, errors.New("mock get timestamp fail")
}
// txnFuture is a promise, which promises to return a txn in future.
type txnFuture struct {
future oracle.Future
store kv.Storage
}
func (tf *txnFuture) wait() (kv.Transaction, error) {
startTS, err := tf.future.Wait()
if err == nil {
return tf.store.BeginWithStartTS(startTS)
} else if _, ok := tf.future.(txnFailFuture); ok {
return nil, err
}
// It would retry get timestamp.
return tf.store.Begin()
}
func (s *session) getTxnFuture(ctx context.Context) *txnFuture {
if span := opentracing.SpanFromContext(ctx); span != nil && span.Tracer() != nil {
span1 := span.Tracer().StartSpan("session.getTxnFuture", opentracing.ChildOf(span.Context()))
defer span1.Finish()
ctx = opentracing.ContextWithSpan(ctx, span1)
}
oracleStore := s.store.GetOracle()
var tsFuture oracle.Future
if s.sessionVars.LowResolutionTSO {
tsFuture = oracleStore.GetLowResolutionTimestampAsync(ctx)
} else {
tsFuture = oracleStore.GetTimestampAsync(ctx)
}
ret := &txnFuture{future: tsFuture, store: s.store}
failpoint.InjectContext(ctx, "mockGetTSFail", func() {
ret.future = txnFailFuture{}
})
return ret
}
// StmtCommit implements the sessionctx.Context interface.
func (s *session) StmtCommit() error {
defer s.txn.cleanup()
st := &s.txn
var count int
err := kv.WalkMemBuffer(st.buf, func(k kv.Key, v []byte) error {
failpoint.Inject("mockStmtCommitError", func(val failpoint.Value) {
if val.(bool) {
count++
}
})
if count > 3 {
return errors.New("mock stmt commit error")
}
if len(v) == 0 {
return st.Transaction.Delete(k)
}
return st.Transaction.Set(k, v)
})
if err != nil {
st.doNotCommit = err
return err
}
// Need to flush binlog.
for tableID, delta := range st.mutations {
mutation := getBinlogMutation(s, tableID)
mergeToMutation(mutation, delta)
}
if len(st.dirtyTableOP) > 0 {
dirtyDB := executor.GetDirtyDB(s)
for _, op := range st.dirtyTableOP {
mergeToDirtyDB(dirtyDB, op)
}
}
return nil
}
// StmtRollback implements the sessionctx.Context interface.
func (s *session) StmtRollback() {
s.txn.cleanup()
}
// StmtGetMutation implements the sessionctx.Context interface.
func (s *session) StmtGetMutation(tableID int64) *binlog.TableMutation {
st := &s.txn
if _, ok := st.mutations[tableID]; !ok {
st.mutations[tableID] = &binlog.TableMutation{TableId: tableID}
}
return st.mutations[tableID]
}
func (s *session) StmtAddDirtyTableOP(op int, tid int64, handle int64) {
s.txn.dirtyTableOP = append(s.txn.dirtyTableOP, dirtyTableOperation{op, tid, handle})
}