client-go/integration_tests/2pc_test.go

1536 lines
48 KiB
Go

// Copyright 2021 TiKV Authors
//
// 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.
// NOTE: The code in this file is based on code from the
// TiDB project, licensed under the Apache License v 2.0
//
// https://github.com/pingcap/tidb/tree/cc5e161ac06827589c4966674597c137cc9e809c/store/tikv/tests/2pc_test.go
// https://github.com/pingcap/tidb/tree/cc5e161ac06827589c4966674597c137cc9e809c/store/tikv/tests/2pc_fail_test.go
// https://github.com/pingcap/tidb/tree/cc5e161ac06827589c4966674597c137cc9e809c/store/tikv/tests/2pc_slow_test.go
//
// Copyright 2016 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 tikv_test
import (
"bytes"
"context"
"fmt"
"math"
"math/rand"
"sync"
"sync/atomic"
"testing"
"time"
"github.com/ninedraft/israce"
"github.com/pingcap/errors"
"github.com/pingcap/failpoint"
"github.com/pingcap/kvproto/pkg/kvrpcpb"
"github.com/pingcap/parser/terror"
drivertxn "github.com/pingcap/tidb/store/driver/txn"
"github.com/stretchr/testify/suite"
"github.com/tikv/client-go/v2/config"
tikverr "github.com/tikv/client-go/v2/error"
"github.com/tikv/client-go/v2/kv"
"github.com/tikv/client-go/v2/oracle"
"github.com/tikv/client-go/v2/testutils"
"github.com/tikv/client-go/v2/tikv"
"github.com/tikv/client-go/v2/tikvrpc"
"github.com/tikv/client-go/v2/txnkv"
"github.com/tikv/client-go/v2/txnkv/transaction"
"github.com/tikv/client-go/v2/txnkv/txnlock"
)
var (
txnCommitBatchSize = tikv.ConfigProbe{}.GetTxnCommitBatchSize()
bigTxnThreshold = tikv.ConfigProbe{}.GetBigTxnThreshold()
)
func TestCommitter(t *testing.T) {
suite.Run(t, new(testCommitterSuite))
}
type testCommitterSuite struct {
suite.Suite
cluster testutils.Cluster
store tikv.StoreProbe
}
func (s *testCommitterSuite) SetupSuite() {
atomic.StoreUint64(&transaction.ManagedLockTTL, 3000) // 3s
atomic.StoreUint64(&transaction.CommitMaxBackoff, 1000)
atomic.StoreUint64(&transaction.VeryLongMaxBackoff, 1000)
}
func (s *testCommitterSuite) TearDownSuite() {
atomic.StoreUint64(&transaction.CommitMaxBackoff, 20000)
atomic.StoreUint64(&transaction.VeryLongMaxBackoff, 600000)
}
func (s *testCommitterSuite) SetupTest() {
client, cluster, pdClient, err := testutils.NewMockTiKV("", nil)
s.Require().Nil(err)
testutils.BootstrapWithMultiRegions(cluster, []byte("a"), []byte("b"), []byte("c"))
s.cluster = cluster
pdCli := &tikv.CodecPDClient{Client: pdClient}
spkv := tikv.NewMockSafePointKV()
store, err := tikv.NewKVStore("mocktikv-store", pdCli, spkv, client)
store.EnableTxnLocalLatches(8096)
s.Require().Nil(err)
s.store = tikv.StoreProbe{KVStore: store}
}
func (s *testCommitterSuite) TearDownTest() {
s.store.Close()
}
func (s *testCommitterSuite) begin() transaction.TxnProbe {
txn, err := s.store.Begin()
s.Require().Nil(err)
return txn
}
func (s *testCommitterSuite) beginAsyncCommit() transaction.TxnProbe {
txn, err := s.store.Begin()
s.Require().Nil(err)
txn.SetEnableAsyncCommit(true)
return txn
}
func (s *testCommitterSuite) checkValues(m map[string]string) {
txn := s.begin()
for k, v := range m {
val, err := txn.Get(context.TODO(), []byte(k))
s.Nil(err)
s.Equal(string(val), v)
}
}
func (s *testCommitterSuite) mustCommit(m map[string]string) {
txn := s.begin()
for k, v := range m {
err := txn.Set([]byte(k), []byte(v))
s.Nil(err)
}
err := txn.Commit(context.Background())
s.Nil(err)
s.checkValues(m)
}
func randKV(keyLen, valLen int) (string, string) {
const letters = "abc"
k, v := make([]byte, keyLen), make([]byte, valLen)
for i := range k {
k[i] = letters[rand.Intn(len(letters))]
}
for i := range v {
v[i] = letters[rand.Intn(len(letters))]
}
return string(k), string(v)
}
func (s *testCommitterSuite) TestDeleteYourWritesTTL() {
conf := *config.GetGlobalConfig()
oldConf := conf
defer config.StoreGlobalConfig(&oldConf)
conf.TiKVClient.TTLRefreshedTxnSize = 0
config.StoreGlobalConfig(&conf)
{
txn := s.begin()
err := txn.GetMemBuffer().SetWithFlags([]byte("bb"), []byte{0}, kv.SetPresumeKeyNotExists)
s.Nil(err)
err = txn.Set([]byte("ba"), []byte{1})
s.Nil(err)
err = txn.Delete([]byte("bb"))
s.Nil(err)
committer, err := txn.NewCommitter(0)
s.Nil(err)
err = committer.PrewriteAllMutations(context.Background())
s.Nil(err)
s.True(committer.IsTTLRunning())
}
{
txn := s.begin()
err := txn.GetMemBuffer().SetWithFlags([]byte("dd"), []byte{0}, kv.SetPresumeKeyNotExists)
s.Nil(err)
err = txn.Set([]byte("de"), []byte{1})
s.Nil(err)
err = txn.Delete([]byte("dd"))
s.Nil(err)
committer, err := txn.NewCommitter(0)
s.Nil(err)
err = committer.PrewriteAllMutations(context.Background())
s.Nil(err)
s.True(committer.IsTTLRunning())
}
}
func (s *testCommitterSuite) TestCommitRollback() {
s.mustCommit(map[string]string{
"a": "a",
"b": "b",
"c": "c",
})
txn := s.begin()
txn.Set([]byte("a"), []byte("a1"))
txn.Set([]byte("b"), []byte("b1"))
txn.Set([]byte("c"), []byte("c1"))
s.mustCommit(map[string]string{
"c": "c2",
})
err := txn.Commit(context.Background())
s.NotNil(err)
s.checkValues(map[string]string{
"a": "a",
"b": "b",
"c": "c2",
})
}
func (s *testCommitterSuite) TestPrewriteRollback() {
s.mustCommit(map[string]string{
"a": "a0",
"b": "b0",
})
ctx := context.Background()
txn1 := s.begin()
err := txn1.Set([]byte("a"), []byte("a1"))
s.Nil(err)
err = txn1.Set([]byte("b"), []byte("b1"))
s.Nil(err)
committer, err := txn1.NewCommitter(0)
s.Nil(err)
err = committer.PrewriteAllMutations(ctx)
s.Nil(err)
txn2 := s.begin()
v, err := txn2.Get(context.TODO(), []byte("a"))
s.Nil(err)
s.Equal(v, []byte("a0"))
err = committer.PrewriteAllMutations(ctx)
if err != nil {
// Retry.
txn1 = s.begin()
err = txn1.Set([]byte("a"), []byte("a1"))
s.Nil(err)
err = txn1.Set([]byte("b"), []byte("b1"))
s.Nil(err)
committer, err = txn1.NewCommitter(0)
s.Nil(err)
err = committer.PrewriteAllMutations(ctx)
s.Nil(err)
}
commitTS, err := s.store.GetOracle().GetTimestamp(ctx, &oracle.Option{TxnScope: oracle.GlobalTxnScope})
s.Nil(err)
committer.SetCommitTS(commitTS)
err = committer.CommitMutations(ctx)
s.Nil(err)
txn3 := s.begin()
v, err = txn3.Get(context.TODO(), []byte("b"))
s.Nil(err)
s.Equal(v, []byte("b1"))
}
func (s *testCommitterSuite) TestContextCancel() {
txn1 := s.begin()
err := txn1.Set([]byte("a"), []byte("a1"))
s.Nil(err)
err = txn1.Set([]byte("b"), []byte("b1"))
s.Nil(err)
committer, err := txn1.NewCommitter(0)
s.Nil(err)
ctx, cancel := context.WithCancel(context.Background())
cancel() // cancel the context
err = committer.PrewriteAllMutations(ctx)
s.Equal(errors.Cause(err), context.Canceled)
}
func (s *testCommitterSuite) TestContextCancel2() {
txn := s.begin()
err := txn.Set([]byte("a"), []byte("a"))
s.Nil(err)
err = txn.Set([]byte("b"), []byte("b"))
s.Nil(err)
ctx, cancel := context.WithCancel(context.Background())
err = txn.Commit(ctx)
s.Nil(err)
cancel()
// Secondary keys should not be canceled.
s.Eventually(func() bool {
return !s.isKeyLocked([]byte("b"))
}, 2*time.Second, 20*time.Millisecond, "Secondary locks are not committed after 2 seconds")
}
func (s *testCommitterSuite) TestContextCancelRetryable() {
txn1, txn2, txn3 := s.begin(), s.begin(), s.begin()
// txn1 locks "b"
err := txn1.Set([]byte("b"), []byte("b1"))
s.Nil(err)
committer, err := txn1.NewCommitter(0)
s.Nil(err)
err = committer.PrewriteAllMutations(context.Background())
s.Nil(err)
// txn3 writes "c"
err = txn3.Set([]byte("c"), []byte("c3"))
s.Nil(err)
err = txn3.Commit(context.Background())
s.Nil(err)
// txn2 writes "a"(PK), "b", "c" on different regions.
// "c" will return a retryable error.
// "b" will get a Locked error first, then the context must be canceled after backoff for lock.
err = txn2.Set([]byte("a"), []byte("a2"))
s.Nil(err)
err = txn2.Set([]byte("b"), []byte("b2"))
s.Nil(err)
err = txn2.Set([]byte("c"), []byte("c2"))
s.Nil(err)
err = txn2.Commit(context.Background())
s.NotNil(err)
_, ok := err.(*tikverr.ErrWriteConflictInLatch)
s.True(ok, fmt.Sprintf("err: %s", err))
}
func (s *testCommitterSuite) TestContextCancelCausingUndetermined() {
// For a normal transaction, if RPC returns context.Canceled error while sending commit
// requests, the transaction should go to the undetermined state.
txn := s.begin()
err := txn.Set([]byte("a"), []byte("va"))
s.Nil(err)
committer, err := txn.NewCommitter(0)
s.Nil(err)
committer.PrewriteAllMutations(context.Background())
s.Nil(err)
s.Nil(failpoint.Enable("tikvclient/rpcContextCancelErr", `return(true)`))
defer func() {
s.Nil(failpoint.Disable("tikvclient/rpcContextCancelErr"))
}()
err = committer.CommitMutations(context.Background())
s.NotNil(committer.GetUndeterminedErr())
s.Equal(errors.Cause(err), context.Canceled)
}
func (s *testCommitterSuite) mustGetRegionID(key []byte) uint64 {
loc, err := s.store.GetRegionCache().LocateKey(tikv.NewBackofferWithVars(context.Background(), 500, nil), key)
s.Nil(err)
return loc.Region.GetID()
}
func (s *testCommitterSuite) isKeyLocked(key []byte) bool {
ver, err := s.store.CurrentTimestamp(oracle.GlobalTxnScope)
s.Nil(err)
bo := tikv.NewBackofferWithVars(context.Background(), 500, nil)
req := tikvrpc.NewRequest(tikvrpc.CmdGet, &kvrpcpb.GetRequest{
Key: key,
Version: ver,
})
loc, err := s.store.GetRegionCache().LocateKey(bo, key)
s.Nil(err)
resp, err := s.store.SendReq(bo, req, loc.Region, 5000)
s.Nil(err)
s.NotNil(resp.Resp)
keyErr := (resp.Resp.(*kvrpcpb.GetResponse)).GetError()
return keyErr.GetLocked() != nil
}
func (s *testCommitterSuite) TestPrewriteCancel() {
// Setup region delays for key "b" and "c".
delays := map[uint64]time.Duration{
s.mustGetRegionID([]byte("b")): time.Millisecond * 10,
s.mustGetRegionID([]byte("c")): time.Millisecond * 20,
}
s.store.SetTiKVClient(&slowClient{
Client: s.store.GetTiKVClient(),
regionDelays: delays,
})
txn1, txn2 := s.begin(), s.begin()
// txn2 writes "b"
err := txn2.Set([]byte("b"), []byte("b2"))
s.Nil(err)
err = txn2.Commit(context.Background())
s.Nil(err)
// txn1 writes "a"(PK), "b", "c" on different regions.
// "b" will return an error and cancel commit.
err = txn1.Set([]byte("a"), []byte("a1"))
s.Nil(err)
err = txn1.Set([]byte("b"), []byte("b1"))
s.Nil(err)
err = txn1.Set([]byte("c"), []byte("c1"))
s.Nil(err)
err = txn1.Commit(context.Background())
s.NotNil(err)
// "c" should be cleaned up in reasonable time.
s.Eventually(func() bool {
return !s.isKeyLocked([]byte("c"))
}, 500*time.Millisecond, 10*time.Millisecond)
}
// slowClient wraps rpcClient and makes some regions respond with delay.
type slowClient struct {
tikv.Client
regionDelays map[uint64]time.Duration
}
func (c *slowClient) SendReq(ctx context.Context, addr string, req *tikvrpc.Request, timeout time.Duration) (*tikvrpc.Response, error) {
for id, delay := range c.regionDelays {
reqCtx := &req.Context
if reqCtx.GetRegionId() == id {
time.Sleep(delay)
}
}
return c.Client.SendRequest(ctx, addr, req, timeout)
}
func (s *testCommitterSuite) TestIllegalTso() {
txn := s.begin()
data := map[string]string{
"name": "aa",
"age": "12",
}
for k, v := range data {
err := txn.Set([]byte(k), []byte(v))
s.Nil(err)
}
// make start ts bigger.
txn.SetStartTS(math.MaxUint64)
err := txn.Commit(context.Background())
s.NotNil(err)
s.errMsgMustContain(err, "invalid txnStartTS")
}
func (s *testCommitterSuite) errMsgMustContain(err error, msg string) {
s.Contains(err.Error(), msg)
}
func (s *testCommitterSuite) TestCommitBeforePrewrite() {
txn := s.begin()
err := txn.Set([]byte("a"), []byte("a1"))
s.Nil(err)
committer, err := txn.NewCommitter(0)
s.Nil(err)
ctx := context.Background()
committer.Cleanup(ctx)
err = committer.PrewriteAllMutations(ctx)
s.NotNil(err)
s.errMsgMustContain(err, "already rolled back")
}
func (s *testCommitterSuite) TestPrewritePrimaryKeyFailed() {
// commit (a,a1)
txn1 := s.begin()
err := txn1.Set([]byte("a"), []byte("a1"))
s.Nil(err)
err = txn1.Commit(context.Background())
s.Nil(err)
// check a
txn := s.begin()
v, err := txn.Get(context.TODO(), []byte("a"))
s.Nil(err)
s.Equal(v, []byte("a1"))
// set txn2's startTs before txn1's
txn2 := s.begin()
txn2.SetStartTS(txn1.StartTS() - 1)
err = txn2.Set([]byte("a"), []byte("a2"))
s.Nil(err)
err = txn2.Set([]byte("b"), []byte("b2"))
s.Nil(err)
// prewrite:primary a failed, b success
err = txn2.Commit(context.Background())
s.NotNil(err)
// txn2 failed with a rollback for record a.
txn = s.begin()
v, err = txn.Get(context.TODO(), []byte("a"))
s.Nil(err)
s.Equal(v, []byte("a1"))
_, err = txn.Get(context.TODO(), []byte("b"))
s.True(tikverr.IsErrNotFound(err))
// clean again, shouldn't be failed when a rollback already exist.
ctx := context.Background()
committer, err := txn2.NewCommitter(0)
s.Nil(err)
committer.Cleanup(ctx)
// check the data after rollback twice.
txn = s.begin()
v, err = txn.Get(context.TODO(), []byte("a"))
s.Nil(err)
s.Equal(v, []byte("a1"))
// update data in a new txn, should be success.
err = txn.Set([]byte("a"), []byte("a3"))
s.Nil(err)
err = txn.Commit(context.Background())
s.Nil(err)
// check value
txn = s.begin()
v, err = txn.Get(context.TODO(), []byte("a"))
s.Nil(err)
s.Equal(v, []byte("a3"))
}
func (s *testCommitterSuite) TestWrittenKeysOnConflict() {
// This test checks that when there is a write conflict, written keys is collected,
// so we can use it to clean up keys.
region, _ := s.cluster.GetRegionByKey([]byte("x"))
newRegionID := s.cluster.AllocID()
newPeerID := s.cluster.AllocID()
s.cluster.Split(region.Id, newRegionID, []byte("y"), []uint64{newPeerID}, newPeerID)
var totalTime time.Duration
for i := 0; i < 10; i++ {
txn1 := s.begin()
txn2 := s.begin()
txn2.Set([]byte("x1"), []byte("1"))
committer2, err := txn2.NewCommitter(2)
s.Nil(err)
err = committer2.Execute(context.Background())
s.Nil(err)
txn1.Set([]byte("x1"), []byte("1"))
txn1.Set([]byte("y1"), []byte("2"))
committer1, err := txn1.NewCommitter(2)
s.Nil(err)
err = committer1.Execute(context.Background())
s.NotNil(err)
committer1.WaitCleanup()
txn3 := s.begin()
start := time.Now()
txn3.Get(context.TODO(), []byte("y1"))
totalTime += time.Since(start)
txn3.Commit(context.Background())
}
s.Less(totalTime, time.Millisecond*200)
}
func (s *testCommitterSuite) TestPrewriteTxnSize() {
// Prepare two regions first: (, 100) and [100, )
region, _ := s.cluster.GetRegionByKey([]byte{50})
newRegionID := s.cluster.AllocID()
newPeerID := s.cluster.AllocID()
s.cluster.Split(region.Id, newRegionID, []byte{100}, []uint64{newPeerID}, newPeerID)
txn := s.begin()
var val [1024]byte
for i := byte(50); i < 120; i++ {
err := txn.Set([]byte{i}, val[:])
s.Nil(err)
}
committer, err := txn.NewCommitter(1)
s.Nil(err)
ctx := context.Background()
err = committer.PrewriteAllMutations(ctx)
s.Nil(err)
// Check the written locks in the first region (50 keys)
for i := byte(50); i < 100; i++ {
lock := s.getLockInfo([]byte{i})
s.Equal(int(lock.TxnSize), 50)
}
// Check the written locks in the second region (20 keys)
for i := byte(100); i < 120; i++ {
lock := s.getLockInfo([]byte{i})
s.Equal(int(lock.TxnSize), 20)
}
}
func (s *testCommitterSuite) TestRejectCommitTS() {
txn := s.begin()
s.Nil(txn.Set([]byte("x"), []byte("v")))
committer, err := txn.NewCommitter(1)
s.Nil(err)
bo := tikv.NewBackofferWithVars(context.Background(), 5000, nil)
loc, err := s.store.GetRegionCache().LocateKey(bo, []byte("x"))
s.Nil(err)
mutations := []*kvrpcpb.Mutation{
{
Op: committer.GetMutations().GetOp(0),
Key: committer.GetMutations().GetKey(0),
Value: committer.GetMutations().GetValue(0),
},
}
prewrite := &kvrpcpb.PrewriteRequest{
Mutations: mutations,
PrimaryLock: committer.GetPrimaryKey(),
StartVersion: committer.GetStartTS(),
LockTtl: committer.GetLockTTL(),
MinCommitTs: committer.GetStartTS() + 100, // Set minCommitTS
}
req := tikvrpc.NewRequest(tikvrpc.CmdPrewrite, prewrite)
_, err = s.store.SendReq(bo, req, loc.Region, 5000)
s.Nil(err)
// Make commitTS less than minCommitTS.
committer.SetCommitTS(committer.GetStartTS() + 1)
// Ensure that the new commit ts is greater than minCommitTS when retry
time.Sleep(3 * time.Millisecond)
err = committer.CommitMutations(context.Background())
s.Nil(err)
// Use startTS+2 to read the data and get nothing.
// Use max.Uint64 to read the data and success.
// That means the final commitTS > startTS+2, it's not the one we provide.
// So we cover the rety commitTS logic.
txn1, err := s.store.BeginWithOption(tikv.DefaultStartTSOption().SetStartTS(committer.GetStartTS() + 2))
s.Nil(err)
_, err = txn1.Get(bo.GetCtx(), []byte("x"))
s.True(tikverr.IsErrNotFound(err))
txn2, err := s.store.BeginWithOption(tikv.DefaultStartTSOption().SetStartTS(math.MaxUint64))
s.Nil(err)
val, err := txn2.Get(bo.GetCtx(), []byte("x"))
s.Nil(err)
s.True(bytes.Equal(val, []byte("v")))
}
func (s *testCommitterSuite) TestPessimisticPrewriteRequest() {
// This test checks that the isPessimisticLock field is set in the request even when no keys are pessimistic lock.
txn := s.begin()
txn.SetPessimistic(true)
err := txn.Set([]byte("t1"), []byte("v1"))
s.Nil(err)
committer, err := txn.NewCommitter(0)
s.Nil(err)
committer.SetForUpdateTS(100)
req := committer.BuildPrewriteRequest(1, 1, 1, committer.GetMutations().Slice(0, 1), 1)
s.Greater(len(req.Prewrite().IsPessimisticLock), 0)
s.Equal(req.Prewrite().ForUpdateTs, uint64(100))
}
func (s *testCommitterSuite) TestUnsetPrimaryKey() {
// This test checks that the isPessimisticLock field is set in the request even when no keys are pessimistic lock.
key := []byte("key")
txn := s.begin()
s.Nil(txn.Set(key, key))
s.Nil(txn.Commit(context.Background()))
txn = s.begin()
txn.SetPessimistic(true)
_, _ = txn.GetUnionStore().Get(context.TODO(), key)
s.Nil(txn.GetMemBuffer().SetWithFlags(key, key, kv.SetPresumeKeyNotExists))
lockCtx := &kv.LockCtx{ForUpdateTS: txn.StartTS(), WaitStartTime: time.Now()}
err := txn.LockKeys(context.Background(), lockCtx, key)
s.NotNil(err)
s.Nil(txn.Delete(key))
key2 := []byte("key2")
s.Nil(txn.Set(key2, key2))
err = txn.Commit(context.Background())
s.Nil(err)
}
func (s *testCommitterSuite) TestPessimisticLockedKeysDedup() {
txn := s.begin()
txn.SetPessimistic(true)
lockCtx := &kv.LockCtx{ForUpdateTS: 100, WaitStartTime: time.Now()}
err := txn.LockKeys(context.Background(), lockCtx, []byte("abc"), []byte("def"))
s.Nil(err)
lockCtx = &kv.LockCtx{ForUpdateTS: 100, WaitStartTime: time.Now()}
err = txn.LockKeys(context.Background(), lockCtx, []byte("abc"), []byte("def"))
s.Nil(err)
s.Len(txn.CollectLockedKeys(), 2)
}
func (s *testCommitterSuite) TestPessimisticTTL() {
key := []byte("key")
txn := s.begin()
txn.SetPessimistic(true)
time.Sleep(time.Millisecond * 100)
lockCtx := &kv.LockCtx{ForUpdateTS: txn.StartTS(), WaitStartTime: time.Now()}
err := txn.LockKeys(context.Background(), lockCtx, key)
s.Nil(err)
time.Sleep(time.Millisecond * 100)
key2 := []byte("key2")
lockCtx = &kv.LockCtx{ForUpdateTS: txn.StartTS(), WaitStartTime: time.Now()}
err = txn.LockKeys(context.Background(), lockCtx, key2)
s.Nil(err)
lockInfo := s.getLockInfo(key)
msBeforeLockExpired := s.store.GetOracle().UntilExpired(txn.StartTS(), lockInfo.LockTtl, &oracle.Option{TxnScope: oracle.GlobalTxnScope})
s.GreaterOrEqual(msBeforeLockExpired, int64(100))
lr := s.store.NewLockResolver()
bo := tikv.NewBackofferWithVars(context.Background(), 5000, nil)
status, err := lr.GetTxnStatus(bo, txn.StartTS(), key2, 0, txn.StartTS(), true, false, nil)
s.Nil(err)
s.GreaterOrEqual(status.TTL(), lockInfo.LockTtl)
// Check primary lock TTL is auto increasing while the pessimistic txn is ongoing.
check := func() bool {
lockInfoNew := s.getLockInfo(key)
if lockInfoNew.LockTtl > lockInfo.LockTtl {
currentTS, err := s.store.GetOracle().GetTimestamp(bo.GetCtx(), &oracle.Option{TxnScope: oracle.GlobalTxnScope})
s.Nil(err)
// Check that the TTL is update to a reasonable range.
expire := oracle.ExtractPhysical(txn.StartTS()) + int64(lockInfoNew.LockTtl)
now := oracle.ExtractPhysical(currentTS)
s.True(expire > now)
s.True(uint64(expire-now) <= atomic.LoadUint64(&transaction.ManagedLockTTL))
return true
}
return false
}
s.Eventually(check, 5*time.Second, 100*time.Millisecond)
}
func (s *testCommitterSuite) TestPessimisticLockReturnValues() {
key := []byte("key")
key2 := []byte("key2")
txn := s.begin()
s.Nil(txn.Set(key, key))
s.Nil(txn.Set(key2, key2))
s.Nil(txn.Commit(context.Background()))
txn = s.begin()
txn.SetPessimistic(true)
lockCtx := &kv.LockCtx{ForUpdateTS: txn.StartTS(), WaitStartTime: time.Now()}
lockCtx.InitReturnValues(2)
s.Nil(txn.LockKeys(context.Background(), lockCtx, key, key2))
s.Len(lockCtx.Values, 2)
s.Equal(lockCtx.Values[string(key)].Value, key)
s.Equal(lockCtx.Values[string(key2)].Value, key2)
}
// TestElapsedTTL tests that elapsed time is correct even if ts physical time is greater than local time.
func (s *testCommitterSuite) TestElapsedTTL() {
key := []byte("key")
txn := s.begin()
txn.SetStartTS(oracle.GoTimeToTS(time.Now().Add(time.Second*10)) + 1)
txn.SetPessimistic(true)
time.Sleep(time.Millisecond * 100)
lockCtx := &kv.LockCtx{
ForUpdateTS: oracle.ComposeTS(oracle.ExtractPhysical(txn.StartTS())+100, 1),
WaitStartTime: time.Now(),
}
err := txn.LockKeys(context.Background(), lockCtx, key)
s.Nil(err)
lockInfo := s.getLockInfo(key)
s.GreaterOrEqual(lockInfo.LockTtl-atomic.LoadUint64(&transaction.ManagedLockTTL), uint64(100))
s.Less(lockInfo.LockTtl-atomic.LoadUint64(&transaction.ManagedLockTTL), uint64(150))
}
func (s *testCommitterSuite) TestDeleteYourWriteCauseGhostPrimary() {
s.cluster.SplitKeys([]byte("d"), []byte("a"), 4)
k1 := []byte("a") // insert but deleted key at first pos in txn1
k2 := []byte("b") // insert key at second pos in txn1
k3 := []byte("c") // insert key in txn1 and will be conflict read by txn2
// insert k1, k2, k3 and delete k1
txn1 := s.begin()
txn1.SetPessimistic(false)
s.store.ClearTxnLatches()
txn1.Get(context.Background(), k1)
txn1.GetMemBuffer().SetWithFlags(k1, []byte{0}, kv.SetPresumeKeyNotExists)
txn1.Set(k2, []byte{1})
txn1.Set(k3, []byte{2})
txn1.Delete(k1)
committer1, err := txn1.NewCommitter(0)
s.Nil(err)
// setup test knob in txn's committer
ac, bk := make(chan struct{}), make(chan struct{})
committer1.SetPrimaryKeyBlocker(ac, bk)
txn1.SetCommitter(committer1)
var txn1Done sync.WaitGroup
txn1Done.Add(1)
go func() {
err1 := txn1.Commit(context.Background())
s.Nil(err1)
txn1Done.Done()
}()
// resume after after primary key be committed
<-ac
// start txn2 to read k3(prewrite success and primary should be committed)
txn2 := s.begin()
txn2.SetPessimistic(false)
s.store.ClearTxnLatches()
v, err := txn2.Get(context.Background(), k3)
s.Nil(err) // should resolve lock and read txn1 k3 result instead of rollback it.
s.Equal(v[0], byte(2))
bk <- struct{}{}
txn1Done.Wait()
}
func (s *testCommitterSuite) TestDeleteAllYourWrites() {
s.cluster.SplitKeys([]byte("d"), []byte("a"), 4)
k1 := []byte("a")
k2 := []byte("b")
k3 := []byte("c")
// insert k1, k2, k3 and delete k1, k2, k3
txn1 := s.begin()
txn1.SetPessimistic(false)
s.store.ClearTxnLatches()
txn1.GetMemBuffer().SetWithFlags(k1, []byte{0}, kv.SetPresumeKeyNotExists)
txn1.Delete(k1)
txn1.GetMemBuffer().SetWithFlags(k2, []byte{1}, kv.SetPresumeKeyNotExists)
txn1.Delete(k2)
txn1.GetMemBuffer().SetWithFlags(k3, []byte{2}, kv.SetPresumeKeyNotExists)
txn1.Delete(k3)
err1 := txn1.Commit(context.Background())
s.Nil(err1)
}
func (s *testCommitterSuite) TestDeleteAllYourWritesWithSFU() {
s.cluster.SplitKeys([]byte("d"), []byte("a"), 4)
k1 := []byte("a")
k2 := []byte("b")
k3 := []byte("c")
// insert k1, k2, k2 and delete k1
txn1 := s.begin()
txn1.SetPessimistic(false)
s.store.ClearTxnLatches()
txn1.GetMemBuffer().SetWithFlags(k1, []byte{0}, kv.SetPresumeKeyNotExists)
txn1.Delete(k1)
err := txn1.LockKeys(context.Background(), &kv.LockCtx{}, k2, k3) // select * from t where x in (k2, k3) for update
s.Nil(err)
committer1, err := txn1.NewCommitter(0)
s.Nil(err)
// setup test knob in txn's committer
ac, bk := make(chan struct{}), make(chan struct{})
committer1.SetPrimaryKeyBlocker(ac, bk)
txn1.SetCommitter(committer1)
var txn1Done sync.WaitGroup
txn1Done.Add(1)
go func() {
err1 := txn1.Commit(context.Background())
s.Nil(err1)
txn1Done.Done()
}()
// resume after after primary key be committed
<-ac
// start txn2 to read k3
txn2 := s.begin()
txn2.SetPessimistic(false)
s.store.ClearTxnLatches()
err = txn2.Set(k3, []byte{33})
s.Nil(err)
var meetLocks []*txnkv.Lock
resolver := tikv.NewLockResolverProb(s.store.GetLockResolver())
resolver.SetMeetLockCallback(func(locks []*txnkv.Lock) {
meetLocks = append(meetLocks, locks...)
})
err = txn2.Commit(context.Background())
s.Nil(err)
bk <- struct{}{}
txn1Done.Wait()
s.Equal(meetLocks[0].Primary[0], k2[0])
}
// TestAcquireFalseTimeoutLock tests acquiring a key which is a secondary key of another transaction.
// The lock's own TTL is expired but the primary key is still alive due to heartbeats.
func (s *testCommitterSuite) TestAcquireFalseTimeoutLock() {
atomic.StoreUint64(&transaction.ManagedLockTTL, 1000) // 1s
defer atomic.StoreUint64(&transaction.ManagedLockTTL, 3000) // restore default test value
// k1 is the primary lock of txn1
k1 := []byte("k1")
// k2 is a secondary lock of txn1 and a key txn2 wants to lock
k2 := []byte("k2")
txn1 := s.begin()
txn1.SetPessimistic(true)
// lock the primary key
lockCtx := &kv.LockCtx{ForUpdateTS: txn1.StartTS(), WaitStartTime: time.Now()}
err := txn1.LockKeys(context.Background(), lockCtx, k1)
s.Nil(err)
// lock the secondary key
lockCtx = &kv.LockCtx{ForUpdateTS: txn1.StartTS(), WaitStartTime: time.Now()}
err = txn1.LockKeys(context.Background(), lockCtx, k2)
s.Nil(err)
// Heartbeats will increase the TTL of the primary key
// wait until secondary key exceeds its own TTL
time.Sleep(time.Duration(atomic.LoadUint64(&transaction.ManagedLockTTL)) * time.Millisecond)
txn2 := s.begin()
txn2.SetPessimistic(true)
// test no wait
lockCtx = kv.NewLockCtx(txn2.StartTS(), kv.LockNoWait, time.Now())
err = txn2.LockKeys(context.Background(), lockCtx, k2)
// cannot acquire lock immediately thus error
s.Equal(err.Error(), tikverr.ErrLockAcquireFailAndNoWaitSet.Error())
// test for wait limited time (200ms)
lockCtx = kv.NewLockCtx(txn2.StartTS(), 200, time.Now())
err = txn2.LockKeys(context.Background(), lockCtx, k2)
// cannot acquire lock in time thus error
s.Equal(err.Error(), tikverr.ErrLockWaitTimeout.Error())
}
func (s *testCommitterSuite) getLockInfo(key []byte) *kvrpcpb.LockInfo {
txn := s.begin()
err := txn.Set(key, key)
s.Nil(err)
committer, err := txn.NewCommitter(1)
s.Nil(err)
bo := tikv.NewBackofferWithVars(context.Background(), 5000, nil)
loc, err := s.store.GetRegionCache().LocateKey(bo, key)
s.Nil(err)
req := committer.BuildPrewriteRequest(loc.Region.GetID(), loc.Region.GetConfVer(), loc.Region.GetVer(), committer.GetMutations().Slice(0, 1), 1)
resp, err := s.store.SendReq(bo, req, loc.Region, 5000)
s.Nil(err)
s.NotNil(resp.Resp)
keyErrs := (resp.Resp.(*kvrpcpb.PrewriteResponse)).Errors
s.Len(keyErrs, 1)
locked := keyErrs[0].Locked
s.NotNil(locked)
return locked
}
func (s *testCommitterSuite) TestPkNotFound() {
atomic.StoreUint64(&transaction.ManagedLockTTL, 100) // 100ms
defer atomic.StoreUint64(&transaction.ManagedLockTTL, 3000) // restore default value
ctx := context.Background()
// k1 is the primary lock of txn1.
k1 := []byte("k1")
// k2 is a secondary lock of txn1 and a key txn2 wants to lock.
k2 := []byte("k2")
k3 := []byte("k3")
txn1 := s.begin()
txn1.SetPessimistic(true)
// lock the primary key.
lockCtx := &kv.LockCtx{ForUpdateTS: txn1.StartTS(), WaitStartTime: time.Now()}
err := txn1.LockKeys(ctx, lockCtx, k1)
s.Nil(err)
// lock the secondary key.
lockCtx = &kv.LockCtx{ForUpdateTS: txn1.StartTS(), WaitStartTime: time.Now()}
err = txn1.LockKeys(ctx, lockCtx, k2, k3)
s.Nil(err)
// Stop txn ttl manager and remove primary key, like tidb server crashes and the priamry key lock does not exists actually,
// while the secondary lock operation succeeded.
txn1.GetCommitter().CloseTTLManager()
var status txnkv.TxnStatus
bo := tikv.NewBackofferWithVars(ctx, 5000, nil)
lockKey2 := &txnkv.Lock{
Key: k2,
Primary: k1,
TxnID: txn1.StartTS(),
TTL: 0, // let the primary lock k1 expire doing check.
TxnSize: txnCommitBatchSize,
LockType: kvrpcpb.Op_PessimisticLock,
LockForUpdateTS: txn1.StartTS(),
}
resolver := tikv.NewLockResolverProb(s.store.GetLockResolver())
status, err = resolver.GetTxnStatusFromLock(bo, lockKey2, oracle.GoTimeToTS(time.Now().Add(200*time.Millisecond)), false)
s.Nil(err)
s.Equal(status.Action(), kvrpcpb.Action_TTLExpirePessimisticRollback)
// Txn2 tries to lock the secondary key k2, there should be no dead loop.
// Since the resolving key k2 is a pessimistic lock, no rollback record should be written, and later lock
// and the other secondary key k3 should succeed if there is no fail point enabled.
status, err = resolver.GetTxnStatusFromLock(bo, lockKey2, oracle.GoTimeToTS(time.Now().Add(200*time.Millisecond)), false)
s.Nil(err)
s.Equal(status.Action(), kvrpcpb.Action_LockNotExistDoNothing)
txn2 := s.begin()
txn2.SetPessimistic(true)
lockCtx = &kv.LockCtx{ForUpdateTS: txn2.StartTS(), WaitStartTime: time.Now()}
err = txn2.LockKeys(ctx, lockCtx, k2)
s.Nil(err)
// Pessimistic rollback using smaller forUpdateTS does not take effect.
lockKey3 := &txnkv.Lock{
Key: k3,
Primary: k1,
TxnID: txn1.StartTS(),
TTL: transaction.ManagedLockTTL,
TxnSize: txnCommitBatchSize,
LockType: kvrpcpb.Op_PessimisticLock,
LockForUpdateTS: txn1.StartTS() - 1,
}
err = resolver.ResolvePessimisticLock(ctx, lockKey3)
s.Nil(err)
lockCtx = &kv.LockCtx{ForUpdateTS: txn1.StartTS(), WaitStartTime: time.Now()}
err = txn1.LockKeys(ctx, lockCtx, k3)
s.Nil(err)
// After disable fail point, the rollbackIfNotExist flag will be set, and the resolve should succeed. In this
// case, the returned action of TxnStatus should be LockNotExistDoNothing, and lock on k3 could be resolved.
txn3 := s.begin()
txn3.SetPessimistic(true)
lockCtx = kv.NewLockCtx(txn3.StartTS(), kv.LockNoWait, time.Now())
err = txn3.LockKeys(ctx, lockCtx, k3)
s.Nil(err)
status, err = resolver.GetTxnStatusFromLock(bo, lockKey3, oracle.GoTimeToTS(time.Now().Add(200*time.Millisecond)), false)
s.Nil(err)
s.Equal(status.Action(), kvrpcpb.Action_LockNotExistDoNothing)
}
func (s *testCommitterSuite) TestPessimisticLockPrimary() {
// a is the primary lock of txn1
k1 := []byte("a")
// b is a secondary lock of txn1 and a key txn2 wants to lock, b is on another region
k2 := []byte("b")
txn1 := s.begin()
txn1.SetPessimistic(true)
// txn1 lock k1
lockCtx := &kv.LockCtx{ForUpdateTS: txn1.StartTS(), WaitStartTime: time.Now()}
err := txn1.LockKeys(context.Background(), lockCtx, k1)
s.Nil(err)
// txn2 wants to lock k1, k2, k1(pk) is blocked by txn1, pessimisticLockKeys has been changed to
// lock primary key first and then secondary keys concurrently, k2 should not be locked by txn2
doneCh := make(chan error)
go func() {
txn2 := s.begin()
txn2.SetPessimistic(true)
lockCtx2 := kv.NewLockCtx(txn2.StartTS(), 200, time.Now())
waitErr := txn2.LockKeys(context.Background(), lockCtx2, k1, k2)
doneCh <- waitErr
}()
time.Sleep(50 * time.Millisecond)
// txn3 should locks k2 successfully using no wait
txn3 := s.begin()
txn3.SetPessimistic(true)
lockCtx3 := kv.NewLockCtx(txn3.StartTS(), kv.LockNoWait, time.Now())
s.Nil(failpoint.Enable("tikvclient/txnNotFoundRetTTL", "return"))
err = txn3.LockKeys(context.Background(), lockCtx3, k2)
s.Nil(failpoint.Disable("tikvclient/txnNotFoundRetTTL"))
s.Nil(err)
waitErr := <-doneCh
s.Equal(tikverr.ErrLockWaitTimeout, waitErr)
}
func (s *testCommitterSuite) TestResolvePessimisticLock() {
untouchedIndexKey := []byte("t00000001_i000000001")
untouchedIndexValue := []byte{0, 0, 0, 0, 0, 0, 0, 1, 49}
noValueIndexKey := []byte("t00000001_i000000002")
txn := s.begin()
txn.SetKVFilter(drivertxn.TiDBKVFilter{})
err := txn.Set(untouchedIndexKey, untouchedIndexValue)
s.Nil(err)
lockCtx := kv.NewLockCtx(txn.StartTS(), kv.LockNoWait, time.Now())
err = txn.LockKeys(context.Background(), lockCtx, untouchedIndexKey, noValueIndexKey)
s.Nil(err)
commit, err := txn.NewCommitter(1)
s.Nil(err)
mutation := commit.MutationsOfKeys([][]byte{untouchedIndexKey, noValueIndexKey})
s.Equal(mutation.Len(), 2)
s.Equal(mutation.GetOp(0), kvrpcpb.Op_Lock)
s.Equal(mutation.GetKey(0), untouchedIndexKey)
s.Equal(mutation.GetValue(0), untouchedIndexValue)
s.Equal(mutation.GetOp(1), kvrpcpb.Op_Lock)
s.Equal(mutation.GetKey(1), noValueIndexKey)
s.Len(mutation.GetValue(1), 0)
}
func (s *testCommitterSuite) TestCommitDeadLock() {
// Split into two region and let k1 k2 in different regions.
s.cluster.SplitKeys([]byte("z"), []byte("a"), 2)
k1 := []byte("a_deadlock_k1")
k2 := []byte("y_deadlock_k2")
region1, _ := s.cluster.GetRegionByKey(k1)
region2, _ := s.cluster.GetRegionByKey(k2)
s.True(region1.Id != region2.Id)
txn1 := s.begin()
txn1.Set(k1, []byte("t1"))
txn1.Set(k2, []byte("t1"))
commit1, err := txn1.NewCommitter(1)
s.Nil(err)
commit1.SetPrimaryKey(k1)
commit1.SetTxnSize(1000 * 1024 * 1024)
txn2 := s.begin()
txn2.Set(k1, []byte("t2"))
txn2.Set(k2, []byte("t2"))
commit2, err := txn2.NewCommitter(2)
s.Nil(err)
commit2.SetPrimaryKey(k2)
commit2.SetTxnSize(1000 * 1024 * 1024)
s.cluster.ScheduleDelay(txn2.StartTS(), region1.Id, 5*time.Millisecond)
s.cluster.ScheduleDelay(txn1.StartTS(), region2.Id, 5*time.Millisecond)
// Txn1 prewrites k1, k2 and txn2 prewrites k2, k1, the large txn
// protocol run ttlManager and update their TTL, cause dead lock.
ch := make(chan error, 2)
var wg sync.WaitGroup
wg.Add(1)
go func() {
ch <- commit2.Execute(context.Background())
wg.Done()
}()
ch <- commit1.Execute(context.Background())
wg.Wait()
close(ch)
res := 0
for e := range ch {
if e != nil {
res++
}
}
s.Equal(res, 1)
}
// TestPushPessimisticLock tests that push forward the minCommiTS of pessimistic locks.
func (s *testCommitterSuite) TestPushPessimisticLock() {
// k1 is the primary key.
k1, k2 := []byte("a"), []byte("b")
ctx := context.Background()
txn1 := s.begin()
txn1.SetPessimistic(true)
lockCtx := &kv.LockCtx{ForUpdateTS: txn1.StartTS(), WaitStartTime: time.Now()}
err := txn1.LockKeys(context.Background(), lockCtx, k1, k2)
s.Nil(err)
txn1.Set(k2, []byte("v2"))
committer := txn1.GetCommitter()
err = committer.InitKeysAndMutations()
s.Nil(err)
// Strip the prewrite of the primary key.
committer.SetMutations(committer.GetMutations().Slice(1, 2))
s.Nil(err)
err = committer.PrewriteAllMutations(ctx)
s.Nil(err)
// The primary lock is a pessimistic lock and the secondary lock is a optimistic lock.
lock1 := s.getLockInfo(k1)
s.Equal(lock1.LockType, kvrpcpb.Op_PessimisticLock)
s.Equal(lock1.PrimaryLock, k1)
lock2 := s.getLockInfo(k2)
s.Equal(lock2.LockType, kvrpcpb.Op_Put)
s.Equal(lock2.PrimaryLock, k1)
txn2 := s.begin()
start := time.Now()
_, err = txn2.Get(ctx, k2)
elapsed := time.Since(start)
// The optimistic lock shouldn't block reads.
s.Less(elapsed, 500*time.Millisecond)
s.True(tikverr.IsErrNotFound(err))
txn1.Rollback()
txn2.Rollback()
}
// TestResolveMixed tests mixed resolve with left behind optimistic locks and pessimistic locks,
// using clean whole region resolve path
func (s *testCommitterSuite) TestResolveMixed() {
atomic.StoreUint64(&transaction.ManagedLockTTL, 100) // 100ms
defer atomic.StoreUint64(&transaction.ManagedLockTTL, 3000) // restore default value
ctx := context.Background()
// pk is the primary lock of txn1
pk := []byte("pk")
secondaryLockkeys := make([][]byte, 0, bigTxnThreshold)
for i := 0; i < bigTxnThreshold; i++ {
optimisticLock := []byte(fmt.Sprintf("optimisticLockKey%d", i))
secondaryLockkeys = append(secondaryLockkeys, optimisticLock)
}
pessimisticLockKey := []byte("pessimisticLockKey")
// make the optimistic and pessimistic lock left with primary lock not found
txn1 := s.begin()
txn1.SetPessimistic(true)
// lock the primary key
lockCtx := &kv.LockCtx{ForUpdateTS: txn1.StartTS(), WaitStartTime: time.Now()}
err := txn1.LockKeys(context.Background(), lockCtx, pk)
s.Nil(err)
// lock the optimistic keys
for i := 0; i < bigTxnThreshold; i++ {
txn1.Set(secondaryLockkeys[i], []byte(fmt.Sprintf("v%d", i)))
}
committer := txn1.GetCommitter()
err = committer.InitKeysAndMutations()
s.Nil(err)
err = committer.PrewriteAllMutations(ctx)
s.Nil(err)
// lock the pessimistic keys
err = txn1.LockKeys(context.Background(), lockCtx, pessimisticLockKey)
s.Nil(err)
lock1 := s.getLockInfo(pessimisticLockKey)
s.Equal(lock1.LockType, kvrpcpb.Op_PessimisticLock)
s.Equal(lock1.PrimaryLock, pk)
optimisticLockKey := secondaryLockkeys[0]
lock2 := s.getLockInfo(optimisticLockKey)
s.Equal(lock2.LockType, kvrpcpb.Op_Put)
s.Equal(lock2.PrimaryLock, pk)
// stop txn ttl manager and remove primary key, make the other keys left behind
committer.CloseTTLManager()
muts := transaction.NewPlainMutations(1)
muts.Push(kvrpcpb.Op_Lock, pk, nil, true)
err = committer.PessimisticRollbackMutations(context.Background(), &muts)
s.Nil(err)
// try to resolve the left optimistic locks, use clean whole region
time.Sleep(time.Duration(atomic.LoadUint64(&transaction.ManagedLockTTL)) * time.Millisecond)
optimisticLockInfo := s.getLockInfo(optimisticLockKey)
lock := txnlock.NewLock(optimisticLockInfo)
resolver := tikv.NewLockResolverProb(s.store.GetLockResolver())
err = resolver.ResolveLock(ctx, lock)
s.Nil(err)
// txn2 tries to lock the pessimisticLockKey, the lock should has been resolved in clean whole region resolve
txn2 := s.begin()
txn2.SetPessimistic(true)
lockCtx = kv.NewLockCtx(txn2.StartTS(), kv.LockNoWait, time.Now())
err = txn2.LockKeys(context.Background(), lockCtx, pessimisticLockKey)
s.Nil(err)
err = txn1.Rollback()
s.Nil(err)
err = txn2.Rollback()
s.Nil(err)
}
// TestSecondaryKeys tests that when async commit is enabled, each prewrite message includes an
// accurate list of secondary keys.
func (s *testCommitterSuite) TestPrewriteSecondaryKeys() {
// Prepare two regions first: (, 100) and [100, )
region, _ := s.cluster.GetRegionByKey([]byte{50})
newRegionID := s.cluster.AllocID()
newPeerID := s.cluster.AllocID()
s.cluster.Split(region.Id, newRegionID, []byte{100}, []uint64{newPeerID}, newPeerID)
txn := s.beginAsyncCommit()
var val [1024]byte
for i := byte(50); i < 120; i++ {
err := txn.Set([]byte{i}, val[:])
s.Nil(err)
}
// Some duplicates.
for i := byte(50); i < 120; i += 10 {
err := txn.Set([]byte{i}, val[512:700])
s.Nil(err)
}
committer, err := txn.NewCommitter(1)
s.Nil(err)
mock := mockClient{inner: s.store.GetTiKVClient()}
s.store.SetTiKVClient(&mock)
ctx := context.Background()
// TODO remove this when minCommitTS is returned from mockStore prewrite response.
committer.SetMinCommitTS(committer.GetStartTS() + 10)
committer.SetNoFallBack()
err = committer.Execute(ctx)
s.Nil(err)
s.True(mock.seenPrimaryReq > 0)
s.True(mock.seenSecondaryReq > 0)
}
func (s *testCommitterSuite) TestAsyncCommit() {
ctx := context.Background()
pk := []byte("tpk")
pkVal := []byte("pkVal")
k1 := []byte("tk1")
k1Val := []byte("k1Val")
txn1 := s.beginAsyncCommit()
err := txn1.Set(pk, pkVal)
s.Nil(err)
err = txn1.Set(k1, k1Val)
s.Nil(err)
committer, err := txn1.NewCommitter(0)
s.Nil(err)
committer.SetMinCommitTS(txn1.StartTS() + 10)
err = committer.Execute(ctx)
s.Nil(err)
s.checkValues(map[string]string{
string(pk): string(pkVal),
string(k1): string(k1Val),
})
}
func updateGlobalConfig(f func(conf *config.Config)) {
g := config.GetGlobalConfig()
newConf := *g
f(&newConf)
config.StoreGlobalConfig(&newConf)
}
// restoreFunc gets a function that restore the config to the current value.
func restoreGlobalConfFunc() (restore func()) {
g := config.GetGlobalConfig()
return func() {
config.StoreGlobalConfig(g)
}
}
func (s *testCommitterSuite) TestAsyncCommitCheck() {
defer restoreGlobalConfFunc()()
updateGlobalConfig(func(conf *config.Config) {
conf.TiKVClient.AsyncCommit.KeysLimit = 16
conf.TiKVClient.AsyncCommit.TotalKeySizeLimit = 64
})
txn := s.beginAsyncCommit()
buf := []byte{0, 0, 0, 0}
// Set 16 keys, each key is 4 bytes long. So the total size of keys is 64 bytes.
for i := 0; i < 16; i++ {
buf[0] = byte(i)
err := txn.Set(buf, []byte("v"))
s.Nil(err)
}
committer, err := txn.NewCommitter(1)
s.Nil(err)
s.True(committer.CheckAsyncCommit())
updateGlobalConfig(func(conf *config.Config) {
conf.TiKVClient.AsyncCommit.KeysLimit = 15
})
s.False(committer.CheckAsyncCommit())
updateGlobalConfig(func(conf *config.Config) {
conf.TiKVClient.AsyncCommit.KeysLimit = 20
conf.TiKVClient.AsyncCommit.TotalKeySizeLimit = 63
})
s.False(committer.CheckAsyncCommit())
}
type mockClient struct {
inner tikv.Client
seenPrimaryReq uint32
seenSecondaryReq uint32
}
func (m *mockClient) SendRequest(ctx context.Context, addr string, req *tikvrpc.Request, timeout time.Duration) (*tikvrpc.Response, error) {
// If we find a prewrite request, check if it satisfies our constraints.
if pr, ok := req.Req.(*kvrpcpb.PrewriteRequest); ok {
if pr.UseAsyncCommit {
if isPrimary(pr) {
// The primary key should not be included, nor should there be any duplicates. All keys should be present.
if !includesPrimary(pr) && allKeysNoDups(pr) {
atomic.StoreUint32(&m.seenPrimaryReq, 1)
}
} else {
// Secondaries should only be sent with the primary key
if len(pr.Secondaries) == 0 {
atomic.StoreUint32(&m.seenSecondaryReq, 1)
}
}
}
}
return m.inner.SendRequest(ctx, addr, req, timeout)
}
func (m *mockClient) Close() error {
return m.inner.Close()
}
func isPrimary(req *kvrpcpb.PrewriteRequest) bool {
for _, m := range req.Mutations {
if bytes.Equal(req.PrimaryLock, m.Key) {
return true
}
}
return false
}
func includesPrimary(req *kvrpcpb.PrewriteRequest) bool {
for _, k := range req.Secondaries {
if bytes.Equal(req.PrimaryLock, k) {
return true
}
}
return false
}
func allKeysNoDups(req *kvrpcpb.PrewriteRequest) bool {
check := make(map[string]bool)
// Create the check map and check for duplicates.
for _, k := range req.Secondaries {
s := string(k)
if check[s] {
return false
}
check[s] = true
}
// Check every key is present.
for i := byte(50); i < 120; i++ {
k := []byte{i}
if !bytes.Equal(req.PrimaryLock, k) && !check[string(k)] {
return false
}
}
return true
}
// TestFailCommitPrimaryRpcErrors tests rpc errors are handled properly when
// committing primary region task.
func (s *testCommitterSuite) TestFailCommitPrimaryRpcErrors() {
s.Nil(failpoint.Enable("tikvclient/rpcCommitResult", `return("timeout")`))
defer func() {
s.Nil(failpoint.Disable("tikvclient/rpcCommitResult"))
}()
// The rpc error will be wrapped to ErrResultUndetermined.
t1 := s.begin()
err := t1.Set([]byte("a"), []byte("a1"))
s.Nil(err)
err = t1.Commit(context.Background())
s.NotNil(err)
s.True(terror.ErrorEqual(err, terror.ErrResultUndetermined), errors.ErrorStack(err))
// We don't need to call "Rollback" after "Commit" fails.
err = t1.Rollback()
s.Equal(err, tikverr.ErrInvalidTxn)
}
// TestFailCommitPrimaryRegionError tests RegionError is handled properly when
// committing primary region task.
func (s *testCommitterSuite) TestFailCommitPrimaryRegionError() {
s.Nil(failpoint.Enable("tikvclient/rpcCommitResult", `return("notLeader")`))
defer func() {
s.Nil(failpoint.Disable("tikvclient/rpcCommitResult"))
}()
// Ensure it returns the original error without wrapped to ErrResultUndetermined
// if it exceeds max retry timeout on RegionError.
t2 := s.begin()
err := t2.Set([]byte("b"), []byte("b1"))
s.Nil(err)
err = t2.Commit(context.Background())
s.NotNil(err)
s.True(terror.ErrorNotEqual(err, terror.ErrResultUndetermined))
}
// TestFailCommitPrimaryRPCErrorThenRegionError tests the case when commit first
// receive a rpc timeout, then region errors afterwrards.
func (s *testCommitterSuite) TestFailCommitPrimaryRPCErrorThenRegionError() {
s.Nil(failpoint.Enable("tikvclient/rpcCommitResult", `1*return("timeout")->return("notLeader")`))
defer func() {
s.Nil(failpoint.Disable("tikvclient/rpcCommitResult"))
}()
// The region error will be wrapped to ErrResultUndetermined.
t1 := s.begin()
err := t1.Set([]byte("a"), []byte("a1"))
s.Nil(err)
err = t1.Commit(context.Background())
s.NotNil(err)
s.True(terror.ErrorEqual(err, terror.ErrResultUndetermined), errors.ErrorStack(err))
}
// TestFailCommitPrimaryKeyError tests KeyError is handled properly when
// committing primary region task.
func (s *testCommitterSuite) TestFailCommitPrimaryKeyError() {
s.Nil(failpoint.Enable("tikvclient/rpcCommitResult", `return("keyError")`))
defer func() {
s.Nil(failpoint.Disable("tikvclient/rpcCommitResult"))
}()
// Ensure it returns the original error without wrapped to ErrResultUndetermined
// if it meets KeyError.
t3 := s.begin()
err := t3.Set([]byte("c"), []byte("c1"))
s.Nil(err)
err = t3.Commit(context.Background())
s.NotNil(err)
s.True(terror.ErrorNotEqual(err, terror.ErrResultUndetermined))
}
// TestFailCommitPrimaryRPCErrorThenKeyError tests KeyError overwrites the undeterminedErr.
func (s *testCommitterSuite) TestFailCommitPrimaryRPCErrorThenKeyError() {
s.Nil(failpoint.Enable("tikvclient/rpcCommitResult", `1*return("timeout")->return("keyError")`))
defer func() {
s.Nil(failpoint.Disable("tikvclient/rpcCommitResult"))
}()
// Ensure it returns the original error without wrapped to ErrResultUndetermined
// if it meets KeyError.
t3 := s.begin()
err := t3.Set([]byte("c"), []byte("c1"))
s.Nil(err)
err = t3.Commit(context.Background())
s.NotNil(err)
s.False(terror.ErrorEqual(err, terror.ErrResultUndetermined))
}
func (s *testCommitterSuite) TestFailCommitTimeout() {
s.Nil(failpoint.Enable("tikvclient/rpcCommitTimeout", `return(true)`))
defer func() {
s.Nil(failpoint.Disable("tikvclient/rpcCommitTimeout"))
}()
txn := s.begin()
err := txn.Set([]byte("a"), []byte("a1"))
s.Nil(err)
err = txn.Set([]byte("b"), []byte("b1"))
s.Nil(err)
err = txn.Set([]byte("c"), []byte("c1"))
s.Nil(err)
err = txn.Commit(context.Background())
s.NotNil(err)
txn2 := s.begin()
value, err := txn2.Get(context.TODO(), []byte("a"))
s.Nil(err)
s.Greater(len(value), 0)
_, err = txn2.Get(context.TODO(), []byte("b"))
s.Nil(err)
s.Greater(len(value), 0)
}
// TestCommitMultipleRegions tests commit multiple regions.
// The test takes too long under the race detector.
func (s *testCommitterSuite) TestCommitMultipleRegions() {
if israce.Race {
s.T().Skip("skip slow tests")
}
m := make(map[string]string)
for i := 0; i < 100; i++ {
k, v := randKV(10, 10)
m[k] = v
}
s.mustCommit(m)
// Test big values.
m = make(map[string]string)
for i := 0; i < 50; i++ {
k, v := randKV(11, int(txnCommitBatchSize)/7)
m[k] = v
}
s.mustCommit(m)
}