[pick-7.1] region_request: ignore resource group errors that not rela…

…tive storag layer (#1354) (#1464)

ref #1322

Signed-off-by: nolouch <[email protected]>

internal/locate/region_request.go

@@ -64,6 +64,7 @@ import (
 	"github.com/tikv/client-go/v2/metrics"
 	"github.com/tikv/client-go/v2/tikvrpc"
 	"github.com/tikv/client-go/v2/util"
+	"github.com/tikv/pd/client/errs"
 	pderr "github.com/tikv/pd/client/errs"
 )
 
@@ -1679,7 +1680,9 @@ func (s *RegionRequestSender) sendReqToRegion(
 	}
 
 	if err != nil {
-		s.rpcError = err
+		if isRPCError(err) {
+			s.rpcError = err
+		}
 
 		// Because in rpc logic, context.Cancel() will be transferred to rpcContext.Cancel error. For rpcContext cancel,
 		// we need to retry the request. But for context cancel active, for example, limitExec gets the required rows,
@@ -1701,6 +1704,11 @@ func (s *RegionRequestSender) sendReqToRegion(
 	return
 }
 
+func isRPCError(err error) bool {
+	// exclude ErrClientResourceGroupThrottled
+	return err != nil && errs.ErrClientResourceGroupThrottled.NotEqual(err)
+}
+
 func (s *RegionRequestSender) getStoreToken(st *Store, limit int64) error {
 	// Checking limit is not thread safe, preferring this for avoiding load in loop.
 	count := st.tokenCount.Load()
@@ -1747,16 +1755,6 @@ func (s *RegionRequestSender) onSendFail(bo *retry.Backoffer, ctx *RPCContext, e
 		}
 	}
 
-	if ctx.Store != nil && ctx.Store.storeType == tikvrpc.TiFlashCompute {
-		s.regionCache.InvalidateTiFlashComputeStoresIfGRPCError(err)
-	} else if ctx.Meta != nil {
-		if s.replicaSelector != nil {
-			s.replicaSelector.onSendFailure(bo, err)
-		} else {
-			s.regionCache.OnSendFail(bo, ctx, s.NeedReloadRegion(ctx), err)
-		}
-	}
-
 	// don't need to retry for ResourceGroup error
 	if errors.Is(err, pderr.ErrClientResourceGroupThrottled) {
 		return err
@@ -1769,6 +1767,16 @@ func (s *RegionRequestSender) onSendFail(bo *retry.Backoffer, ctx *RPCContext, e
 		return err
 	}
 
+	if ctx.Store != nil && ctx.Store.storeType == tikvrpc.TiFlashCompute {
+		s.regionCache.InvalidateTiFlashComputeStoresIfGRPCError(err)
+	} else if ctx.Meta != nil {
+		if s.replicaSelector != nil {
+			s.replicaSelector.onSendFailure(bo, err)
+		} else {
+			s.regionCache.OnSendFail(bo, ctx, s.NeedReloadRegion(ctx), err)
+		}
+	}
+
 	// Retry on send request failure when it's not canceled.
 	// When a store is not available, the leader of related region should be elected quickly.
 	// TODO: the number of retry time should be limited:since region may be unavailable

internal/locate/region_request_test.go

@@ -61,6 +61,7 @@ import (
 	"github.com/tikv/client-go/v2/internal/mockstore/mocktikv"
 	"github.com/tikv/client-go/v2/internal/retry"
 	"github.com/tikv/client-go/v2/tikvrpc"
+	pderr "github.com/tikv/pd/client/errs"
 	"google.golang.org/grpc"
 )
 
@@ -144,6 +145,41 @@ func (s *testRegionRequestToSingleStoreSuite) TestOnRegionError() {
 	}()
 }
 
+func (s *testRegionRequestToSingleStoreSuite) TestOnSendFailByResourceGroupThrottled() {
+	req := tikvrpc.NewRequest(tikvrpc.CmdRawPut, &kvrpcpb.RawPutRequest{
+		Key:   []byte("key"),
+		Value: []byte("value"),
+	})
+	region, err := s.cache.LocateRegionByID(s.bo, s.region)
+	s.Nil(err)
+	s.NotNil(region)
+
+	// test ErrClientResourceGroupThrottled handled by regionRequestSender
+	func() {
+		oc := s.regionRequestSender.client
+		defer func() {
+			s.regionRequestSender.client = oc
+		}()
+		s.regionRequestSender.regionCache.storeMu.Lock()
+		storeOld := s.regionRequestSender.regionCache.storeMu.stores[1]
+		s.regionRequestSender.regionCache.storeMu.Unlock()
+		epoch := storeOld.epoch
+		s.regionRequestSender.client = &fnClient{fn: func(ctx context.Context, addr string, req *tikvrpc.Request, timeout time.Duration) (response *tikvrpc.Response, err error) {
+			return nil, pderr.ErrClientResourceGroupThrottled
+		}}
+		bo := retry.NewBackofferWithVars(context.Background(), 5, nil)
+		_, _, err := s.regionRequestSender.SendReq(bo, req, region.Region, time.Second)
+		s.NotNil(err)
+		s.regionRequestSender.regionCache.storeMu.Lock()
+		storeNew := s.regionRequestSender.regionCache.storeMu.stores[1]
+		s.regionRequestSender.regionCache.storeMu.Unlock()
+		//  not mark the store need be refill, then the epoch should not be changed.
+		s.Equal(epoch, storeNew.epoch)
+		// no rpc error if the error is ErrClientResourceGroupThrottled
+		s.Nil(s.regionRequestSender.rpcError)
+	}()
+}
+
 func (s *testRegionRequestToSingleStoreSuite) TestOnSendFailedWithStoreRestart() {
 	req := tikvrpc.NewRequest(tikvrpc.CmdRawPut, &kvrpcpb.RawPutRequest{
 		Key:   []byte("key"),