txnkv: prevent some actions from being interrupted by kill (tikv#1665)

fix pingcap/tidb#61454

Signed-off-by: zyguan <zhongyangguan@gmail.com>

Author: zyguan

integration_tests/2pc_test.go

@@ -2707,6 +2707,50 @@ func (s *testCommitterSuite) TestKillSignal() {
 	s.ErrorContains(err, "query interrupted")
 }
 
+func (s *testCommitterSuite) TestUninterruptibleAction() {
+	s.Run("Cleanup", func() {
+		var killed uint32 = 0
+		txn := s.begin()
+		txn.SetVars(kv.NewVariables(&killed))
+		err := txn.Set([]byte("k1"), []byte("v1"))
+		s.NoError(err)
+		committer, err := txn.NewCommitter(0)
+		s.NoError(err)
+		err = committer.PrewriteAllMutations(context.Background())
+		s.NoError(err)
+		atomic.StoreUint32(&killed, 2)
+		s.NoError(committer.CleanupMutations(context.Background()))
+	})
+	s.Run("PessimisticRollback", func() {
+		var killed uint32 = 0
+		txn := s.begin()
+		txn.SetVars(kv.NewVariables(&killed))
+		txn.SetPessimistic(true)
+		err := txn.LockKeys(context.Background(), kv.NewLockCtx(txn.StartTS(), kv.LockNoWait, time.Now()), []byte("k2"))
+		s.NoError(err)
+		atomic.StoreUint32(&killed, 2)
+		committer, err := txn.NewCommitter(0)
+		s.NoError(err)
+		s.NoError(committer.PessimisticRollbackMutations(context.Background(), committer.GetMutations()))
+	})
+	s.Run("Commit", func() {
+		var killed uint32 = 0
+		txn := s.begin()
+		txn.SetVars(kv.NewVariables(&killed))
+		err := txn.Set([]byte("k1"), []byte("v1"))
+		s.NoError(err)
+		committer, err := txn.NewCommitter(0)
+		s.NoError(err)
+		err = committer.PrewriteAllMutations(context.Background())
+		s.NoError(err)
+		atomic.StoreUint32(&killed, 2)
+		commitTS, err := s.store.GetOracle().GetTimestamp(context.Background(), &oracle.Option{})
+		s.NoError(err)
+		committer.SetCommitTS(commitTS)
+		s.NoError(committer.CommitMutations(context.Background()))
+	})
+}
+
 func (s *testCommitterSuite) Test2PCLifecycleHooks() {
 	reachedPre := atomic.Bool{}
 	reachedPost := atomic.Bool{}

internal/locate/region_request.go

@@ -912,9 +912,11 @@ func (s *sendReqState) next() (done bool) {
 	bo, req := s.args.bo, s.args.req
 
 	// check whether the session/query is killed during the Next()
-	if err := bo.CheckKilled(); err != nil {
-		s.vars.resp, s.vars.err = nil, err
-		return true
+	if req.IsInterruptible() {
+		if err := bo.CheckKilled(); err != nil {
+			s.vars.resp, s.vars.err = nil, err
+			return true
+		}
 	}
 
 	// handle send error

tikvrpc/tikvrpc.go

@@ -340,6 +340,16 @@ func (req *Request) IsDebugReq() bool {
 	return false
 }
 
+// IsInterruptible checks if the request can be interrupted when the query is killed.
+func (req *Request) IsInterruptible() bool {
+	switch req.Type {
+	case CmdPessimisticRollback, CmdBatchRollback, CmdCommit:
+		return false
+	default:
+		return true
+	}
+}
+
 // Get returns GetRequest in request.
 func (req *Request) Get() *kvrpcpb.GetRequest {
 	return req.Req.(*kvrpcpb.GetRequest)

txnkv/transaction/2pc.go

@@ -75,6 +75,7 @@ const slowRequestThreshold = time.Minute
 type twoPhaseCommitAction interface {
 	handleSingleBatch(*twoPhaseCommitter, *retry.Backoffer, batchMutations) error
 	tiKVTxnRegionsNumHistogram() prometheus.Observer
+	isInterruptible() bool
 	String() string
 }
 
@@ -1069,9 +1070,10 @@ func (c *twoPhaseCommitter) doActionOnBatches(
 	// killSignal should never be nil for TiDB
 	if c.txn != nil && c.txn.vars != nil && c.txn.vars.Killed != nil {
 		// Do not reset the killed flag here. Let the upper layer reset the flag.
-		// Before it resets, any request is considered valid to be killed.
+		// Before it resets, any request is considered valid to be killed if the
+		// corresponding action is interruptible.
 		status := atomic.LoadUint32(c.txn.vars.Killed)
-		if status != 0 {
+		if status != 0 && action.isInterruptible() {
 			logutil.BgLogger().Info(
 				"query is killed", zap.Uint32(
 					"signal",

txnkv/transaction/cleanup.go

@@ -103,6 +103,10 @@ func (action actionCleanup) handleSingleBatch(c *twoPhaseCommitter, bo *retry.Ba
 	return nil
 }
 
+func (actionCleanup) isInterruptible() bool {
+	return false
+}
+
 func (c *twoPhaseCommitter) cleanupMutations(bo *retry.Backoffer, mutations CommitterMutations) error {
 	return c.doActionOnMutations(bo, actionCleanup{isInternal: c.txn.isInternal()}, mutations)
 }

txnkv/transaction/commit.go

@@ -247,6 +247,10 @@ func (action actionCommit) handleSingleBatch(c *twoPhaseCommitter, bo *retry.Bac
 	return nil
 }
 
+func (actionCommit) isInterruptible() bool {
+	return false
+}
+
 func (c *twoPhaseCommitter) commitMutations(bo *retry.Backoffer, mutations CommitterMutations) error {
 	if span := opentracing.SpanFromContext(bo.GetCtx()); span != nil && span.Tracer() != nil {
 		span1 := span.Tracer().StartSpan("twoPhaseCommitter.commitMutations", opentracing.ChildOf(span.Context()))

txnkv/transaction/pessimistic.go

@@ -525,6 +525,10 @@ func (action actionPessimisticLock) handlePessimisticLockResponseForceLockMode(
 	return true, nil
 }
 
+func (actionPessimisticLock) isInterruptible() bool {
+	return true
+}
+
 func (actionPessimisticRollback) handleSingleBatch(
 	c *twoPhaseCommitter, bo *retry.Backoffer, batch batchMutations,
 ) error {
@@ -560,6 +564,10 @@ func (actionPessimisticRollback) handleSingleBatch(
 	return nil
 }
 
+func (actionPessimisticRollback) isInterruptible() bool {
+	return false
+}
+
 func (c *twoPhaseCommitter) pessimisticLockMutations(
 	bo *retry.Backoffer, lockCtx *kv.LockCtx, lockWaitMode kvrpcpb.PessimisticLockWakeUpMode,
 	mutations CommitterMutations,

txnkv/transaction/pipelined_flush.go

@@ -284,6 +284,10 @@ func (action actionPipelinedFlush) handleSingleBatch(
 	}
 }
 
+func (actionPipelinedFlush) isInterruptible() bool {
+	return true
+}
+
 func (c *twoPhaseCommitter) pipelinedFlushMutations(bo *retry.Backoffer, mutations CommitterMutations, generation uint64) error {
 	if span := opentracing.SpanFromContext(bo.GetCtx()); span != nil && span.Tracer() != nil {
 		span1 := span.Tracer().StartSpan("twoPhaseCommitter.pipelinedFlushMutations", opentracing.ChildOf(span.Context()))

txnkv/transaction/prewrite.go

@@ -282,6 +282,10 @@ func (action actionPrewrite) handleSingleBatchFailpoint(c *twoPhaseCommitter, bo
 	return nil
 }
 
+func (actionPrewrite) isInterruptible() bool {
+	return true
+}
+
 func (c *twoPhaseCommitter) prewriteMutations(bo *retry.Backoffer, mutations CommitterMutations) error {
 	if span := opentracing.SpanFromContext(bo.GetCtx()); span != nil && span.Tracer() != nil {
 		span1 := span.Tracer().StartSpan("twoPhaseCommitter.prewriteMutations", opentracing.ChildOf(span.Context()))