tools/ut: accelerate the build process and scatter test

.github/workflows/pd-tests.yaml

@@ -27,19 +27,27 @@ jobs:
       matrix:
         include:
           - worker_id: 1
-            name: 'Unit Test(1)'
+            name: 'Unit Test(core)'
           - worker_id: 2
-            name: 'Unit Test(2)'
+            name: 'Unit Test(schedule)'
           - worker_id: 3
-            name: 'Tools Test'
+            name: 'Unit Test(others)'
           - worker_id: 4
-            name: 'Client Integration Test'
+            name: 'Tests(core)'
           - worker_id: 5
-            name: 'TSO Integration Test'
+            name: 'Tests(others)'
           - worker_id: 6
-            name: 'MicroService Integration Test'
+            name: 'Tools Test'
+          - worker_id: 7
+            name: 'Client Integration Test'
+          - worker_id: 8
+            name: 'TSO Integration Test'
+          - worker_id: 9
+            name: 'MicroService Integration(Core)'
+          - worker_id: 10
+            name: 'MicroService Integration(TSO)'
     outputs:
-      job-total: 6
+      job-total: 10
     steps:
       - name: Checkout code
         uses: actions/checkout@v4
@@ -52,6 +60,9 @@ jobs:
         run: |
           make ci-test-job JOB_INDEX=$WORKER_ID
           mv covprofile covprofile_$WORKER_ID
+          if [ -f junitfile ]; then
+            cat junitfile
+          fi
       - name: Upload coverage result ${{ matrix.worker_id }}
         uses: actions/upload-artifact@v4
         with:

.gitignore

@@ -19,6 +19,7 @@ docs/swagger/*
 *.before
 covprofile
 coverage.tmp
+junitfile
 package.list
 report.xml
 coverage.xml

Makefile

@@ -228,7 +228,7 @@ failpoint-disable: install-tools
 ut: pd-ut
 	@$(FAILPOINT_ENABLE)
 	# only run unit tests
-	./bin/pd-ut run --ignore tests --race
+	./bin/pd-ut run --ignore tests --race --junitfile ./junitfile
 	@$(CLEAN_UT_BINARY)
 	@$(FAILPOINT_DISABLE)
 

pkg/core/region_test.go

@@ -731,7 +731,7 @@ func BenchmarkRandomSetRegion(b *testing.B) {
 
 func TestGetRegionSizeByRange(t *testing.T) {
 	regions := NewRegionsInfo()
-	nums := 1000010
+	nums := 100001
 	for i := 0; i < nums; i++ {
 		peer := &metapb.Peer{StoreId: 1, Id: uint64(i + 1)}
 		endKey := []byte(fmt.Sprintf("%20d", i+1))

scripts/ci-subtask.sh

@@ -3,34 +3,51 @@
 # ./ci-subtask.sh <TOTAL_TASK_N> <TASK_INDEX>
 
 ROOT_PATH_COV=$(pwd)/covprofile
+ROOT_PATH_JUNITFILE=$(pwd)/junitfile
 # Currently, we only have 3 integration tests, so we can hardcode the task index.
 integrations_dir=$(pwd)/tests/integrations
 
 case $1 in
     1)
-        # unit tests ignore `tests`
-        ./bin/pd-ut run --race --ignore tests --coverprofile $ROOT_PATH_COV || exit 1
+        # unit tests ignore `tests`, `server`, `schedule`,`utils` and `encryption`
+        ./bin/pd-ut run --ignore tests,server,pkg/schedule,pkg/utils,pkg/encryption --race --coverprofile $ROOT_PATH_COV --junitfile $ROOT_PATH_JUNITFILE || exit 1
         ;;
     2)
-        # unit tests only in `tests`
-        ./bin/pd-ut run tests --race --coverprofile $ROOT_PATH_COV || exit 1
+        # unit tests only in `schedule` and `encryption` without `tests`
+        ./bin/pd-ut run schedule,encryption --ignore tests --race --coverprofile $ROOT_PATH_COV --junitfile $ROOT_PATH_JUNITFILE || exit 1
         ;;
     3)
+        # unit tests only in `server` and `utils` without `tests`
+        ./bin/pd-ut run server,utils --ignore tests --race --coverprofile $ROOT_PATH_COV --junitfile $ROOT_PATH_JUNITFILE || exit 1
+        ;;
+    4)
+        # tests only in `tests` without `server/api, server/cluster and `server/config`
+        ./bin/pd-ut run tests --ignore tests/server/api,tests/server/cluster,tests/server/config --race --coverprofile $ROOT_PATH_COV --junitfile $ROOT_PATH_JUNITFILE || exit 1
+        ;;
+    5)
+        # tests only in `server/api, server/cluster and `server/config` in `tests`
+        ./bin/pd-ut run tests/server/api,tests/server/cluster,tests/server/config --race --coverprofile $ROOT_PATH_COV --junitfile $ROOT_PATH_JUNITFILE || exit 1
+        ;;
+    6)
         # tools tests
         cd ./tools && make ci-test-job && cat covprofile >> $ROOT_PATH_COV || exit 1
         ;;
-    4)
+    7)
         # integration test client
-        ./bin/pd-ut it run client --race --coverprofile $ROOT_PATH_COV || exit 1
+        ./bin/pd-ut it run client --race --coverprofile $ROOT_PATH_COV --junitfile $ROOT_PATH_JUNITFILE || exit 1
         # client tests
         cd ./client && make ci-test-job && cat covprofile >> $ROOT_PATH_COV || exit 1
         ;;
-    5)
+    8)
         # integration test tso
-        ./bin/pd-ut it run tso --race --coverprofile $ROOT_PATH_COV || exit 1
+        ./bin/pd-ut it run tso --race --ignore mcs/tso --coverprofile $ROOT_PATH_COV --junitfile $ROOT_PATH_JUNITFILE || exit 1
         ;;
-    6)
-        # integration test mcs
-        ./bin/pd-ut it run mcs --race --coverprofile $ROOT_PATH_COV || exit 1
+    9)
+        # integration test mcs without mcs/tso
+        ./bin/pd-ut it run mcs --race --ignore mcs/tso --coverprofile $ROOT_PATH_COV --junitfile $ROOT_PATH_JUNITFILE || exit 1
+        ;;
+    10)
+        # integration test mcs/tso
+        ./bin/pd-ut it run mcs/tso --race --coverprofile $ROOT_PATH_COV --junitfile $ROOT_PATH_JUNITFILE || exit 1
         ;;
 esac

server/api/region_test.go

@@ -333,7 +333,7 @@ func TestRegionsWithKillRequest(t *testing.T) {
 	url := fmt.Sprintf("%s%s/api/v1/regions", addr, apiPrefix)
 	mustBootstrapCluster(re, svr)
 
-	regionCount := 100000
+	regionCount := 10000
 	tu.GenerateTestDataConcurrently(regionCount, func(i int) {
 		r := core.NewTestRegionInfo(uint64(i+2), 1,
 			[]byte(fmt.Sprintf("%09d", i)),

tests/integrations/client/client_test.go

@@ -217,7 +217,7 @@ func TestLeaderTransferAndMoveCluster(t *testing.T) {
 	}()
 
 	// Transfer leader.
-	for i := 0; i < 5; i++ {
+	for i := 0; i < 3; i++ {
 		oldLeaderName := cluster.WaitLeader()
 		err := cluster.GetServer(oldLeaderName).ResignLeader()
 		re.NoError(err)

tests/integrations/mcs/tso/keyspace_group_manager_test.go

@@ -67,7 +67,7 @@ func (suite *tsoKeyspaceGroupManagerTestSuite) allocID() uint32 {
 	return uint32(id)
 }
 
-func TestTSOKeyspaceGroupManager(t *testing.T) {
+func TestTSOKeyspaceGroupManagerSuite(t *testing.T) {
 	suite.Run(t, &tsoKeyspaceGroupManagerTestSuite{})
 }
 

tests/integrations/mcs/tso/server_test.go

@@ -260,8 +260,8 @@ func TestWaitAPIServiceReady(t *testing.T) {
 	}
 }
 
-type APIServerForwardTestSuite struct {
-	suite.Suite
+type APIServerForward struct {
+	re               *require.Assertions
 	ctx              context.Context
 	cancel           context.CancelFunc
 	cluster          *tests.TestCluster
@@ -270,13 +270,11 @@ type APIServerForwardTestSuite struct {
 	pdClient         pd.Client
 }
 
-func TestAPIServerForwardTestSuite(t *testing.T) {
-	suite.Run(t, new(APIServerForwardTestSuite))
-}
-
-func (suite *APIServerForwardTestSuite) SetupTest() {
+func NewAPIServerForward(re *require.Assertions) APIServerForward {
+	suite := APIServerForward{
+		re: re,
+	}
 	var err error
-	re := suite.Require()
 	suite.ctx, suite.cancel = context.WithCancel(context.Background())
 	suite.cluster, err = tests.NewTestAPICluster(suite.ctx, 3)
 	re.NoError(err)
@@ -294,11 +292,12 @@ func (suite *APIServerForwardTestSuite) SetupTest() {
 	suite.pdClient, err = pd.NewClientWithContext(context.Background(),
 		[]string{suite.backendEndpoints}, pd.SecurityOption{}, pd.WithMaxErrorRetry(1))
 	re.NoError(err)
+	return suite
 }
 
-func (suite *APIServerForwardTestSuite) TearDownTest() {
-	re := suite.Require()
+func (suite *APIServerForward) ShutDown() {
 	suite.pdClient.Close()
+	re := suite.re
 
 	etcdClient := suite.pdLeader.GetEtcdClient()
 	clusterID := strconv.FormatUint(suite.pdLeader.GetClusterID(), 10)
@@ -314,8 +313,10 @@ func (suite *APIServerForwardTestSuite) TearDownTest() {
 	re.NoError(failpoint.Disable("github.com/tikv/pd/client/usePDServiceMode"))
 }
 
-func (suite *APIServerForwardTestSuite) TestForwardTSORelated() {
-	re := suite.Require()
+func TestForwardTSORelated(t *testing.T) {
+	re := require.New(t)
+	suite := NewAPIServerForward(re)
+	defer suite.ShutDown()
 	// Unable to use the tso-related interface without tso server
 	suite.checkUnavailableTSO(re)
 	tc, err := tests.NewTestTSOCluster(suite.ctx, 1, suite.backendEndpoints)
@@ -325,8 +326,10 @@ func (suite *APIServerForwardTestSuite) TestForwardTSORelated() {
 	suite.checkAvailableTSO(re)
 }
 
-func (suite *APIServerForwardTestSuite) TestForwardTSOWhenPrimaryChanged() {
-	re := suite.Require()
+func TestForwardTSOWhenPrimaryChanged(t *testing.T) {
+	re := require.New(t)
+	suite := NewAPIServerForward(re)
+	defer suite.ShutDown()
 
 	tc, err := tests.NewTestTSOCluster(suite.ctx, 2, suite.backendEndpoints)
 	re.NoError(err)
@@ -363,10 +366,11 @@ func (suite *APIServerForwardTestSuite) TestForwardTSOWhenPrimaryChanged() {
 	suite.checkAvailableTSO(re)
 }
 
-func (suite *APIServerForwardTestSuite) TestResignTSOPrimaryForward() {
+func TestResignTSOPrimaryForward(t *testing.T) {
+	re := require.New(t)
+	suite := NewAPIServerForward(re)
+	defer suite.ShutDown()
 	// TODO: test random kill primary with 3 nodes
-	re := suite.Require()
-
 	tc, err := tests.NewTestTSOCluster(suite.ctx, 2, suite.backendEndpoints)
 	re.NoError(err)
 	defer tc.Destroy()
@@ -388,8 +392,10 @@ func (suite *APIServerForwardTestSuite) TestResignTSOPrimaryForward() {
 	}
 }
 
-func (suite *APIServerForwardTestSuite) TestResignAPIPrimaryForward() {
-	re := suite.Require()
+func TestResignAPIPrimaryForward(t *testing.T) {
+	re := require.New(t)
+	suite := NewAPIServerForward(re)
+	defer suite.ShutDown()
 
 	tc, err := tests.NewTestTSOCluster(suite.ctx, 2, suite.backendEndpoints)
 	re.NoError(err)
@@ -410,8 +416,10 @@ func (suite *APIServerForwardTestSuite) TestResignAPIPrimaryForward() {
 	}
 }
 
-func (suite *APIServerForwardTestSuite) TestForwardTSOUnexpectedToFollower1() {
-	re := suite.Require()
+func TestForwardTSOUnexpectedToFollower1(t *testing.T) {
+	re := require.New(t)
+	suite := NewAPIServerForward(re)
+	defer suite.ShutDown()
 	suite.checkForwardTSOUnexpectedToFollower(func() {
 		// unary call will retry internally
 		// try to update gc safe point
@@ -421,8 +429,10 @@ func (suite *APIServerForwardTestSuite) TestForwardTSOUnexpectedToFollower1() {
 	})
 }
 
-func (suite *APIServerForwardTestSuite) TestForwardTSOUnexpectedToFollower2() {
-	re := suite.Require()
+func TestForwardTSOUnexpectedToFollower2(t *testing.T) {
+	re := require.New(t)
+	suite := NewAPIServerForward(re)
+	defer suite.ShutDown()
 	suite.checkForwardTSOUnexpectedToFollower(func() {
 		// unary call will retry internally
 		// try to set external ts
@@ -433,16 +443,18 @@ func (suite *APIServerForwardTestSuite) TestForwardTSOUnexpectedToFollower2() {
 	})
 }
 
-func (suite *APIServerForwardTestSuite) TestForwardTSOUnexpectedToFollower3() {
-	re := suite.Require()
+func TestForwardTSOUnexpectedToFollower3(t *testing.T) {
+	re := require.New(t)
+	suite := NewAPIServerForward(re)
+	defer suite.ShutDown()
 	suite.checkForwardTSOUnexpectedToFollower(func() {
 		_, _, err := suite.pdClient.GetTS(suite.ctx)
 		re.Error(err)
 	})
 }
 
-func (suite *APIServerForwardTestSuite) checkForwardTSOUnexpectedToFollower(checkTSO func()) {
-	re := suite.Require()
+func (suite *APIServerForward) checkForwardTSOUnexpectedToFollower(checkTSO func()) {
+	re := suite.re
 	tc, err := tests.NewTestTSOCluster(suite.ctx, 2, suite.backendEndpoints)
 	re.NoError(err)
 	tc.WaitForDefaultPrimaryServing(re)
@@ -477,7 +489,7 @@ func (suite *APIServerForwardTestSuite) checkForwardTSOUnexpectedToFollower(chec
 	tc.Destroy()
 }
 
-func (suite *APIServerForwardTestSuite) addRegions() {
+func (suite *APIServerForward) addRegions() {
 	leader := suite.cluster.GetServer(suite.cluster.WaitLeader())
 	rc := leader.GetServer().GetRaftCluster()
 	for i := 0; i < 3; i++ {
@@ -491,7 +503,7 @@ func (suite *APIServerForwardTestSuite) addRegions() {
 	}
 }
 
-func (suite *APIServerForwardTestSuite) checkUnavailableTSO(re *require.Assertions) {
+func (suite *APIServerForward) checkUnavailableTSO(re *require.Assertions) {
 	_, _, err := suite.pdClient.GetTS(suite.ctx)
 	re.Error(err)
 	// try to update gc safe point
@@ -502,7 +514,7 @@ func (suite *APIServerForwardTestSuite) checkUnavailableTSO(re *require.Assertio
 	re.Error(err)
 }
 
-func (suite *APIServerForwardTestSuite) checkAvailableTSO(re *require.Assertions) {
+func (suite *APIServerForward) checkAvailableTSO(re *require.Assertions) {
 	mcs.WaitForTSOServiceAvailable(suite.ctx, re, suite.pdClient)
 	// try to get ts
 	_, _, err := suite.pdClient.GetTS(suite.ctx)

tests/integrations/tso/client_test.go

@@ -71,13 +71,13 @@ func (suite *tsoClientTestSuite) getBackendEndpoints() []string {
 	return strings.Split(suite.backendEndpoints, ",")
 }
 
-func TestLegacyTSOClient(t *testing.T) {
+func TestLegacyTSOClientSuite(t *testing.T) {
 	suite.Run(t, &tsoClientTestSuite{
 		legacy: true,
 	})
 }
 
-func TestMicroserviceTSOClient(t *testing.T) {
+func TestMicroserviceTSOClientSuite(t *testing.T) {
 	suite.Run(t, &tsoClientTestSuite{
 		legacy: false,
 	})

tests/integrations/tso/consistency_test.go

@@ -53,13 +53,13 @@ type tsoConsistencyTestSuite struct {
 	tsoClient tsopb.TSOClient
 }
 
-func TestLegacyTSOConsistency(t *testing.T) {
+func TestLegacyTSOConsistencySuite(t *testing.T) {
 	suite.Run(t, &tsoConsistencyTestSuite{
 		legacy: true,
 	})
 }
 
-func TestMicroserviceTSOConsistency(t *testing.T) {
+func TestMicroserviceTSOConsistencySuite(t *testing.T) {
 	suite.Run(t, &tsoConsistencyTestSuite{
 		legacy: false,
 	})

tests/server/api/scheduler_test.go

@@ -40,18 +40,28 @@ const apiPrefix = "/pd"
 
 type scheduleTestSuite struct {
 	suite.Suite
-	env *tests.SchedulingTestEnvironment
+	env     *tests.SchedulingTestEnvironment
+	runMode tests.SchedulerMode
 }
 
-func TestScheduleTestSuite(t *testing.T) {
-	suite.Run(t, new(scheduleTestSuite))
+func TestPDSchedulingTestSuite(t *testing.T) {
+	suite.Run(t, &scheduleTestSuite{
+		runMode: tests.PDMode,
+	})
+}
+
+func TestAPISchedulingTestSuite(t *testing.T) {
+	suite.Run(t, &scheduleTestSuite{
+		runMode: tests.APIMode,
+	})
 }
 
 func (suite *scheduleTestSuite) SetupSuite() {
 	re := suite.Require()
 	re.NoError(failpoint.Enable("github.com/tikv/pd/pkg/schedule/changeCoordinatorTicker", `return(true)`))
 	re.NoError(failpoint.Enable("github.com/tikv/pd/server/cluster/skipStoreConfigSync", `return(true)`))
 	suite.env = tests.NewSchedulingTestEnvironment(suite.T())
+	suite.env.RunMode = suite.runMode
 }
 
 func (suite *scheduleTestSuite) TearDownSuite() {