Merge branch 'master' into add_campaign_times

pkg/mcs/resourcemanager/server/config.go

@@ -127,8 +127,9 @@ func (rmc *ControllerConfig) Adjust(meta *configutil.ConfigMetaData) {
 	if !meta.IsDefined("ltb-token-rpc-max-delay") {
 		configutil.AdjustDuration(&rmc.LTBTokenRPCMaxDelay, defaultLTBTokenRPCMaxDelay)
 	}
-	failpoint.Inject("enableDegradedMode", func() {
+	failpoint.Inject("enableDegradedModeAndTraceLog", func() {
 		configutil.AdjustDuration(&rmc.DegradedModeWaitDuration, time.Second)
+		configutil.AdjustBool(&rmc.EnableControllerTraceLog, true)
 	})
 }
 

server/util.go

@@ -17,7 +17,6 @@ package server
 import (
 	"context"
 	"net/http"
-	"net/http/pprof"
 	"path/filepath"
 	"strings"
 
@@ -133,10 +132,6 @@ func combineBuilderServerHTTPService(ctx context.Context, svr *Server, serviceBu
 
 	apiService.UseHandler(router)
 	userHandlers[pdAPIPrefix] = apiService
-
-	// fix issue https://github.com/tikv/pd/issues/7253
-	// FIXME: remove me after upgrade
-	userHandlers["/debug/pprof/trace"] = http.HandlerFunc(pprof.Trace)
 	return userHandlers, nil
 }
 

tests/integrations/mcs/resourcemanager/resource_manager_test.go

@@ -28,6 +28,7 @@ import (
 
 	"github.com/pingcap/failpoint"
 	rmpb "github.com/pingcap/kvproto/pkg/resource_manager"
+	"github.com/pingcap/log"
 	"github.com/stretchr/testify/require"
 	"github.com/stretchr/testify/suite"
 	pd "github.com/tikv/pd/client"
@@ -37,6 +38,7 @@ import (
 	"github.com/tikv/pd/pkg/utils/typeutil"
 	"github.com/tikv/pd/tests"
 	"go.uber.org/goleak"
+	"go.uber.org/zap"
 
 	// Register Service
 	_ "github.com/tikv/pd/pkg/mcs/registry"
@@ -64,7 +66,7 @@ func (suite *resourceManagerClientTestSuite) SetupSuite() {
 	var err error
 	re := suite.Require()
 
-	re.NoError(failpoint.Enable("github.com/tikv/pd/pkg/mcs/resourcemanager/server/enableDegradedMode", `return(true)`))
+	re.NoError(failpoint.Enable("github.com/tikv/pd/pkg/mcs/resourcemanager/server/enableDegradedModeAndTraceLog", `return(true)`))
 	re.NoError(failpoint.Enable("github.com/tikv/pd/pkg/member/skipCampaignLeaderCheck", "return(true)"))
 
 	suite.ctx, suite.clean = context.WithCancel(context.Background())
@@ -151,7 +153,7 @@ func (suite *resourceManagerClientTestSuite) TearDownSuite() {
 	suite.client.Close()
 	suite.cluster.Destroy()
 	suite.clean()
-	re.NoError(failpoint.Disable("github.com/tikv/pd/pkg/mcs/resourcemanager/server/enableDegradedMode"))
+	re.NoError(failpoint.Disable("github.com/tikv/pd/pkg/mcs/resourcemanager/server/enableDegradedModeAndTraceLog"))
 	re.NoError(failpoint.Disable("github.com/tikv/pd/pkg/member/skipCampaignLeaderCheck"))
 }
 
@@ -410,11 +412,9 @@ func (suite *resourceManagerClientTestSuite) TestResourceGroupController() {
 	testCases := []struct {
 		resourceGroupName string
 		tcs               []tokenConsumptionPerSecond
-		len               int
 	}{
 		{
 			resourceGroupName: rg.Name,
-			len:               8,
 			tcs: []tokenConsumptionPerSecond{
 				{rruTokensAtATime: 50, wruTokensAtATime: 20, times: 100, waitDuration: 0},
 				{rruTokensAtATime: 50, wruTokensAtATime: 100, times: 100, waitDuration: 0},
@@ -434,7 +434,7 @@ func (suite *resourceManagerClientTestSuite) TestResourceGroupController() {
 		v := false
 		<-tricker.C
 		for _, cas := range testCases {
-			if i >= cas.len {
+			if i >= len(cas.tcs) {
 				continue
 			}
 			v = true
@@ -452,8 +452,9 @@ func (suite *resourceManagerClientTestSuite) TestResourceGroupController() {
 				sum += time.Since(startTime)
 				rgsController.OnResponse(cas.resourceGroupName, rreq, rres)
 				rgsController.OnResponse(cas.resourceGroupName, wreq, wres)
-				time.Sleep(1000 * time.Microsecond)
+				time.Sleep(time.Millisecond)
 			}
+			log.Info("finished test case", zap.Int("index", i), zap.Duration("sum", sum), zap.Duration("waitDuration", cas.tcs[i].waitDuration))
 			re.LessOrEqual(sum, buffDuration+cas.tcs[i].waitDuration)
 		}
 		i++
@@ -1479,12 +1480,12 @@ func (suite *resourceManagerClientTestSuite) TestResourceGroupControllerConfigCh
 	readBaseCost := 1.5
 	defaultCfg := controller.DefaultConfig()
 	expectCfg := server.ControllerConfig{
-		// failpoint enableDegradedMode will setup and set it be 1s.
+		// failpoint enableDegradedModeAndTraceLog will set it be 1s and enable trace log.
 		DegradedModeWaitDuration: typeutil.NewDuration(time.Second),
+		EnableControllerTraceLog: true,
 		LTBMaxWaitDuration:       typeutil.Duration(defaultCfg.LTBMaxWaitDuration),
 		LTBTokenRPCMaxDelay:      typeutil.Duration(defaultCfg.LTBTokenRPCMaxDelay),
 		RequestUnit:              server.RequestUnitConfig(defaultCfg.RequestUnit),
-		EnableControllerTraceLog: defaultCfg.EnableControllerTraceLog,
 	}
 	expectRUCfg := controller.GenerateRUConfig(defaultCfg)
 	expectRUCfg.DegradedModeWaitDuration = time.Second

tests/integrations/mcs/scheduling/server_test.go

@@ -106,6 +106,7 @@ func (suite *serverTestSuite) TestAllocIDAfterLeaderChange() {
 	pd2, err := suite.cluster.Join(suite.ctx)
 	re.NoError(err)
 	err = pd2.Run()
+	re.NotEmpty(suite.cluster.WaitLeader())
 	re.NoError(err)
 	tc, err := tests.NewTestSchedulingCluster(suite.ctx, 1, suite.backendEndpoints)
 	re.NoError(err)

tests/server/join/join_test.go

@@ -58,6 +58,7 @@ func TestSimpleJoin(t *testing.T) {
 	re.NoError(err)
 	_, err = os.Stat(filepath.Join(pd2.GetConfig().DataDir, "join"))
 	re.False(os.IsNotExist(err))
+	re.NotEmpty(cluster.WaitLeader())
 	members, err = etcdutil.ListEtcdMembers(ctx, client)
 	re.NoError(err)
 	re.Len(members.Members, 2)
@@ -73,6 +74,7 @@ func TestSimpleJoin(t *testing.T) {
 	re.NoError(err)
 	_, err = os.Stat(filepath.Join(pd3.GetConfig().DataDir, "join"))
 	re.False(os.IsNotExist(err))
+	re.NotEmpty(cluster.WaitLeader())
 	members, err = etcdutil.ListEtcdMembers(ctx, client)
 	re.NoError(err)
 	re.Len(members.Members, 3)

tests/server/watch/leader_watch_test.go

@@ -57,6 +57,7 @@ func TestWatcher(t *testing.T) {
 	re.NoError(failpoint.Enable("github.com/tikv/pd/server/delayWatcher", `pause`))
 	err = pd3.Run()
 	re.NoError(err)
+	re.NotEmpty(cluster.WaitLeader())
 	time.Sleep(200 * time.Millisecond)
 	re.Equal(pd1.GetConfig().Name, pd3.GetLeader().GetName())
 	err = pd1.Stop()