region: fix the potential panic . (#7143)

server/core/region.go

@@ -22,11 +22,13 @@
 	"sort"
 	"strings"
 	"sync/atomic"
+	"time"
 	"unsafe"
 
 	"github.com/docker/go-units"
 	"github.com/gogo/protobuf/proto"
 	"github.com/pingcap/errors"
+	"github.com/pingcap/failpoint"
 	"github.com/pingcap/kvproto/pkg/metapb"
 	"github.com/pingcap/kvproto/pkg/pdpb"
 	"github.com/pingcap/kvproto/pkg/replication_modepb"
@@ -885,7 +887,16 @@
 }
 
 // UpdateSubTree updates the subtree.
+<<<<<<< HEAD:server/core/region.go
 func (r *RegionsInfo) UpdateSubTree(region, origin *RegionInfo, toRemove []*RegionInfo, rangeChanged bool) {
+=======
+func (r *RegionsInfo) UpdateSubTree(region, origin *RegionInfo, overlaps []*RegionInfo, rangeChanged bool) {
+	failpoint.Inject("UpdateSubTree", func() {
+		if origin == nil {
+			time.Sleep(time.Second)
+		}
+	})
+>>>>>>> 54219d649 (region: fix the potential panic .  (#7143)):pkg/core/region.go
 	r.st.Lock()
 	defer r.st.Unlock()
 	if origin != nil {
@@ -894,8 +905,17 @@
 			// TODO: Improve performance by deleting only the different peers.
 			r.removeRegionFromSubTreeLocked(origin)
 		} else {
-			r.updateSubTreeStat(origin, region)
-			r.subRegions[region.GetID()].RegionInfo = region
+			// The region tree and the subtree update is not atomic and the region tree is updated first.
+			// If there are two thread needs to update region tree,
+			// t1: thread-A  update region tree
+			// 										t2: thread-B: update region tree again
+			//										t3: thread-B: update subtree
+			// t4: thread-A: update region subtree
+			// to keep region tree consistent with subtree, we need to drop this update.
+			if tree, ok := r.subRegions[region.GetID()]; ok {
+				r.updateSubTreeStat(origin, region)
+				tree.RegionInfo = region
+			}
 			return
 		}
 	}
@@ -909,7 +929,7 @@
 	r.subRegions[region.GetID()] = item
 	// It has been removed and all information needs to be updated again.
 	// Set peers then.
 	setPeer := func(peersMap map[uint64]*regionTree, storeID uint64, item *regionItem) {
 		store, ok := peersMap[storeID]
 		if !ok {
 			store = newRegionTree()
@@ -930,7 +950,7 @@
 		}
 	}
 
 	setPeers := func(peersMap map[uint64]*regionTree, peers []*metapb.Peer) {
 		for _, peer := range peers {
 			storeID := peer.GetStoreId()
 			setPeer(peersMap, storeID, item)

server/core/region_test.go

@@ -22,6 +22,7 @@ import (
 	"testing"
 	"time"
 
+	"github.com/pingcap/failpoint"
 	"github.com/pingcap/kvproto/pkg/metapb"
 	"github.com/pingcap/kvproto/pkg/pdpb"
 	"github.com/stretchr/testify/require"
@@ -451,6 +452,18 @@ func TestRegionKey(t *testing.T) {
 	}
 }
 
+func TestSetRegionConcurrence(t *testing.T) {
+	re := require.New(t)
+	re.NoError(failpoint.Enable("github.com/tikv/pd/pkg/core/UpdateSubTree", `return()`))
+	regions := NewRegionsInfo()
+	region := NewTestRegionInfo(1, 1, []byte("a"), []byte("b"))
+	go func() {
+		regions.AtomicCheckAndPutRegion(region)
+	}()
+	regions.AtomicCheckAndPutRegion(region)
+	re.NoError(failpoint.Disable("github.com/tikv/pd/pkg/core/UpdateSubTree"))
+}
+
 func TestSetRegion(t *testing.T) {
 	re := require.New(t)
 	regions := NewRegionsInfo()