executor: kill tidb [session id] can't stop executors and release r…

…esources quickly (pingcap#9844)

server/conn.go

@@ -107,6 +107,7 @@ type clientConn struct {
 	mu struct {
 		sync.RWMutex
 		cancelFunc context.CancelFunc
+		resultSets []ResultSet
 	}
 }
 
@@ -883,6 +884,15 @@ func (cc *clientConn) handleQuery(ctx context.Context, sql string) (err error) {
 		metrics.ExecuteErrorCounter.WithLabelValues(metrics.ExecuteErrorToLabel(err)).Inc()
 		return errors.Trace(err)
 	}
+	cc.mu.Lock()
+	cc.mu.resultSets = rs
+	status := atomic.LoadInt32(&cc.status)
+	if status == connStatusShutdown || status == connStatusWaitShutdown {
+		cc.mu.Unlock()
+		killConn(cc)
+		return errors.New("killed by another connection")
+	}
+	cc.mu.Unlock()
 	if rs != nil {
 		if len(rs) == 1 {
 			err = cc.writeResultset(ctx, rs[0], false, 0, 0)

server/driver_tidb.go

@@ -16,7 +16,6 @@ package server
 import (
 	"crypto/tls"
 	"fmt"
-
 	"github.com/pingcap/errors"
 	"github.com/pingcap/parser/ast"
 	"github.com/pingcap/parser/auth"
@@ -31,6 +30,7 @@ import (
 	"github.com/pingcap/tidb/util/chunk"
 	"github.com/pingcap/tidb/util/sqlexec"
 	"golang.org/x/net/context"
+	"sync/atomic"
 )
 
 // TiDBDriver implements IDriver.
@@ -336,7 +336,7 @@ type tidbResultSet struct {
 	recordSet sqlexec.RecordSet
 	columns   []*ColumnInfo
 	rows      []chunk.Row
-	closed    bool
+	closed    int32
 }
 
 func (trs *tidbResultSet) NewChunk() *chunk.Chunk {
@@ -359,10 +359,9 @@ func (trs *tidbResultSet) GetFetchedRows() []chunk.Row {
 }
 
 func (trs *tidbResultSet) Close() error {
-	if trs.closed {
+	if !atomic.CompareAndSwapInt32(&trs.closed, 0, 1) {
 		return nil
 	}
-	trs.closed = true
 	return trs.recordSet.Close()
 }
 

server/server.go

@@ -351,19 +351,25 @@ func (s *Server) Kill(connectionID uint64, query bool) {
 		return
 	}
 
-	killConn(conn, query)
-}
-
-func killConn(conn *clientConn, query bool) {
 	if !query {
 		// Mark the client connection status as WaitShutdown, when the goroutine detect
 		// this, it will end the dispatch loop and exit.
 		atomic.StoreInt32(&conn.status, connStatusWaitShutdown)
 	}
+	killConn(conn)
+}
 
+func killConn(conn *clientConn) {
 	conn.mu.RLock()
+	resultSets := conn.mu.resultSets
 	cancelFunc := conn.mu.cancelFunc
 	conn.mu.RUnlock()
+	for _, resultSet := range resultSets {
+		// resultSet.Close() is reentrant so it's safe to kill a same connID multiple times
+		if err := resultSet.Close(); err != nil {
+			logutil.Logger(context.Background()).Error("close result set error", zap.Uint32("connID", conn.connectionID), zap.Error(err))
+		}
+	}
 	if cancelFunc != nil {
 		cancelFunc()
 	}
@@ -378,12 +384,7 @@ func (s *Server) KillAllConnections() {
 	for _, conn := range s.clients {
 		atomic.StoreInt32(&conn.status, connStatusShutdown)
 		terror.Log(errors.Trace(conn.closeWithoutLock()))
-		conn.mu.RLock()
-		cancelFunc := conn.mu.cancelFunc
-		conn.mu.RUnlock()
-		if cancelFunc != nil {
-			cancelFunc()
-		}
+		killConn(conn)
 	}
 }