dm-relay: create binlog reader from relay, refactor relay

dm/dm/worker/config_test.go

@@ -17,17 +17,20 @@ import (
 	"flag"
 	"os"
 	"strings"
+	"testing"
 
 	"github.com/kami-zh/go-capturer"
 	"github.com/pingcap/check"
 
 	"github.com/pingcap/ticdc/dm/pkg/terror"
 )
 
-var (
-	defaultConfigFile = "./dm-worker.toml"
-	_                 = check.Suite(&testConfigSuite{})
-)
+var defaultConfigFile = "./dm-worker.toml"
+
+func TestConfigSuite(t *testing.T) {
+	_ = check.Suite(&testConfigSuite{})
+	check.TestingT(t)
+}
 
 type testConfigSuite struct{}
 

dm/dm/worker/relay.go

@@ -50,10 +50,8 @@ type RelayHolder interface {
 	Result() *pb.ProcessResult
 	// Update updates relay config online
 	Update(ctx context.Context, cfg *config.SourceConfig) error
-	// RegisterListener registers a relay listener
-	RegisterListener(el relay.Listener)
-	// UnRegisterListener unregisters a relay listener
-	UnRegisterListener(el relay.Listener)
+	// Relay returns relay object
+	Relay() relay.Process
 }
 
 // NewRelayHolder is relay holder initializer
@@ -310,12 +308,8 @@ func (h *realRelayHolder) EarliestActiveRelayLog() *streamer.RelayLogInfo {
 	return h.relay.ActiveRelayLog()
 }
 
-func (h *realRelayHolder) RegisterListener(el relay.Listener) {
-	h.relay.RegisterListener(el)
-}
-
-func (h *realRelayHolder) UnRegisterListener(el relay.Listener) {
-	h.relay.UnRegisterListener(el)
+func (h *realRelayHolder) Relay() relay.Process {
+	return h.relay
 }
 
 /******************** dummy relay holder ********************/
@@ -326,14 +320,16 @@ type dummyRelayHolder struct {
 	stage       pb.Stage
 	relayBinlog string
 
-	cfg *config.SourceConfig
+	cfg    *config.SourceConfig
+	relay2 relay.Process
 }
 
 // NewDummyRelayHolder creates a new RelayHolder.
 func NewDummyRelayHolder(cfg *config.SourceConfig) RelayHolder {
 	return &dummyRelayHolder{
-		cfg:   cfg,
-		stage: pb.Stage_New,
+		cfg:    cfg,
+		stage:  pb.Stage_New,
+		relay2: &relay.Relay{},
 	}
 }
 
@@ -436,8 +432,6 @@ func (d *dummyRelayHolder) Stage() pb.Stage {
 	return d.stage
 }
 
-func (d *dummyRelayHolder) RegisterListener(el relay.Listener) {
-}
-
-func (d *dummyRelayHolder) UnRegisterListener(el relay.Listener) {
+func (d *dummyRelayHolder) Relay() relay.Process {
+	return d.relay2
 }

dm/dm/worker/relay_test.go

@@ -15,6 +15,7 @@ package worker
 
 import (
 	"context"
+	"testing"
 	"time"
 
 	"github.com/go-mysql-org/go-mysql/mysql"
@@ -26,14 +27,18 @@ import (
 	"github.com/pingcap/ticdc/dm/dm/unit"
 	"github.com/pingcap/ticdc/dm/pkg/binlog"
 	"github.com/pingcap/ticdc/dm/pkg/gtid"
+	"github.com/pingcap/ticdc/dm/pkg/log"
 	pkgstreamer "github.com/pingcap/ticdc/dm/pkg/streamer"
 	"github.com/pingcap/ticdc/dm/pkg/utils"
 	"github.com/pingcap/ticdc/dm/relay"
 )
 
 type testRelay struct{}
 
-var _ = Suite(&testRelay{})
+func TestRelay(t *testing.T) {
+	_ = Suite(&testRelay{})
+	TestingT(t)
+}
 
 /*********** dummy relay log process unit, used only for testing *************/
 
@@ -46,6 +51,10 @@ type DummyRelay struct {
 	reloadErr     error
 }
 
+func (d *DummyRelay) NewReader(logger log.Logger, cfg *relay.BinlogReaderConfig) *relay.BinlogReader {
+	return nil
+}
+
 func (d *DummyRelay) RegisterListener(el relay.Listener) {
 }
 

dm/dm/worker/server_test.go

@@ -53,13 +53,12 @@ const (
 var etcdErrCompacted = v3rpc.ErrCompacted
 
 func TestServer(t *testing.T) {
+	_ = Suite(&testServer{})
 	TestingT(t)
 }
 
 type testServer struct{}
 
-var _ = Suite(&testServer{})
-
 func (t *testServer) SetUpSuite(c *C) {
 	err := log.InitLogger(&log.Config{})
 	c.Assert(err, IsNil)
@@ -121,7 +120,7 @@ func (t *testServer) TestServer(c *C) {
 		cfg.UseRelay = false
 		return NewRealSubTask(cfg, etcdClient, worker)
 	}
-	createUnits = func(cfg *config.SubTaskConfig, etcdClient *clientv3.Client, worker string, notifier relay.EventNotifier) []unit.Unit {
+	createUnits = func(cfg *config.SubTaskConfig, etcdClient *clientv3.Client, worker string, relay relay.Process) []unit.Unit {
 		mockDumper := NewMockUnit(pb.UnitType_Dump)
 		mockLoader := NewMockUnit(pb.UnitType_Load)
 		mockSync := NewMockUnit(pb.UnitType_Sync)
@@ -551,7 +550,9 @@ func (t *testServer) testSubTaskRecover(c *C, s *Server, dir string) {
 
 func (t *testServer) testStopWorkerWhenLostConnect(c *C, s *Server, etcd *embed.Etcd) {
 	etcd.Close()
-	time.Sleep(retryConnectSleepTime + time.Duration(defaultKeepAliveTTL+3)*time.Second)
+	c.Assert(utils.WaitSomething(60, time.Second, func() bool {
+		return s.getWorker(true) == nil
+	}), IsTrue)
 	c.Assert(s.getWorker(true), IsNil)
 }
 

dm/dm/worker/source_worker.go

@@ -353,11 +353,9 @@ func (w *SourceWorker) EnableRelay() (err error) {
 		w.observeRelayStage(w.relayCtx, w.etcdClient, revRelay)
 	}()
 
-	w.relayHolder.RegisterListener(w.subTaskHolder)
-
 	w.relayEnabled.Store(true)
 	w.l.Info("relay enabled")
-	w.subTaskHolder.resetAllSubTasks(true)
+	w.subTaskHolder.resetAllSubTasks(w.getRelayWithoutLock())
 	return nil
 }
 
@@ -382,12 +380,11 @@ func (w *SourceWorker) DisableRelay() {
 		w.l.Info("finish refreshing task checker")
 	}
 
-	w.subTaskHolder.resetAllSubTasks(false)
+	w.subTaskHolder.resetAllSubTasks(nil)
 
 	if w.relayHolder != nil {
 		r := w.relayHolder
 		w.relayHolder = nil
-		r.UnRegisterListener(w.subTaskHolder)
 		r.Close()
 	}
 	if w.relayPurger != nil {
@@ -519,7 +516,15 @@ func (w *SourceWorker) StartSubTask(cfg *config.SubTaskConfig, expectStage pb.St
 	}
 
 	w.l.Info("subtask created", zap.Stringer("config", cfg2))
-	st.Run(expectStage)
+	st.Run(expectStage, w.getRelayWithoutLock())
+	return nil
+}
+
+// caller should make sure w.Lock is locked before calling this method.
+func (w *SourceWorker) getRelayWithoutLock() relay.Process {
+	if w.relayHolder != nil {
+		return w.relayHolder.Relay()
+	}
 	return nil
 }
 
@@ -566,10 +571,10 @@ func (w *SourceWorker) OperateSubTask(name string, op pb.TaskOp) error {
 		err = st.Pause()
 	case pb.TaskOp_Resume:
 		w.l.Info("resume sub task", zap.String("task", name))
-		err = st.Resume()
+		err = st.Resume(w.getRelayWithoutLock())
 	case pb.TaskOp_AutoResume:
 		w.l.Info("auto_resume sub task", zap.String("task", name))
-		err = st.Resume()
+		err = st.Resume(w.getRelayWithoutLock())
 	default:
 		err = terror.ErrWorkerUpdateTaskStage.Generatef("invalid operate %s on subtask %v", op, name)
 	}
@@ -1039,5 +1044,5 @@ func (w *SourceWorker) HandleError(ctx context.Context, req *pb.HandleWorkerErro
 		return terror.ErrWorkerSubTaskNotFound.Generate(req.Task)
 	}
 
-	return st.HandleError(ctx, req)
+	return st.HandleError(ctx, req, w.getRelayWithoutLock())
 }

dm/dm/worker/source_worker_test.go

@@ -18,6 +18,7 @@ import (
 	"fmt"
 	"sync"
 	"sync/atomic"
+	"testing"
 	"time"
 
 	"github.com/DATA-DOG/go-sqlmock"
@@ -115,7 +116,10 @@ func (t *testServer) testWorker(c *C) {
 
 type testServer2 struct{}
 
-var _ = Suite(&testServer2{})
+func TestServer2(t *testing.T) {
+	_ = Suite(&testServer2{})
+	TestingT(t)
+}
 
 func (t *testServer2) SetUpSuite(c *C) {
 	err := log.InitLogger(&log.Config{})
@@ -242,7 +246,7 @@ var _ = Suite(&testWorkerFunctionalities{})
 func (t *testWorkerFunctionalities) SetUpSuite(c *C) {
 	NewRelayHolder = NewDummyRelayHolder
 	NewSubTask = NewRealSubTask
-	createUnits = func(cfg *config.SubTaskConfig, etcdClient *clientv3.Client, worker string, notifier relay.EventNotifier) []unit.Unit {
+	createUnits = func(cfg *config.SubTaskConfig, etcdClient *clientv3.Client, worker string, relay relay.Process) []unit.Unit {
 		atomic.AddInt32(&t.createUnitCount, 1)
 		mockDumper := NewMockUnit(pb.UnitType_Dump)
 		mockLoader := NewMockUnit(pb.UnitType_Load)
@@ -417,7 +421,7 @@ func (t *testWorkerEtcdCompact) SetUpSuite(c *C) {
 		cfg.UseRelay = false
 		return NewRealSubTask(cfg, etcdClient, worker)
 	}
-	createUnits = func(cfg *config.SubTaskConfig, etcdClient *clientv3.Client, worker string, notifier relay.EventNotifier) []unit.Unit {
+	createUnits = func(cfg *config.SubTaskConfig, etcdClient *clientv3.Client, worker string, relay relay.Process) []unit.Unit {
 		mockDumper := NewMockUnit(pb.UnitType_Dump)
 		mockLoader := NewMockUnit(pb.UnitType_Load)
 		mockSync := NewMockUnit(pb.UnitType_Sync)

dm/dm/worker/subtask.go

@@ -45,22 +45,12 @@ const (
 	waitRelayCatchupTimeout = 30 * time.Second
 )
 
-type relayNotifier struct {
-	// ch with size = 1, we only need to be notified whether binlog file of relay changed, not how many times
-	ch chan interface{}
-}
-
-// Notified implements streamer.EventNotifier.
-func (r relayNotifier) Notified() chan interface{} {
-	return r.ch
-}
-
 // createRealUnits is subtask units initializer
 // it can be used for testing.
 var createUnits = createRealUnits
 
 // createRealUnits creates process units base on task mode.
-func createRealUnits(cfg *config.SubTaskConfig, etcdClient *clientv3.Client, workerName string, notifier relay.EventNotifier) []unit.Unit {
+func createRealUnits(cfg *config.SubTaskConfig, etcdClient *clientv3.Client, workerName string, relay relay.Process) []unit.Unit {
 	failpoint.Inject("mockCreateUnitsDumpOnly", func(_ failpoint.Value) {
 		log.L().Info("create mock worker units with dump unit only", zap.String("failpoint", "mockCreateUnitsDumpOnly"))
 		failpoint.Return([]unit.Unit{dumpling.NewDumpling(cfg)})
@@ -75,7 +65,7 @@ func createRealUnits(cfg *config.SubTaskConfig, etcdClient *clientv3.Client, wor
 		} else {
 			us = append(us, loader.NewLightning(cfg, etcdClient, workerName))
 		}
-		us = append(us, syncer.NewSyncer(cfg, etcdClient, notifier))
+		us = append(us, syncer.NewSyncer(cfg, etcdClient, relay))
 	case config.ModeFull:
 		// NOTE: maybe need another checker in the future?
 		us = append(us, dumpling.NewDumpling(cfg))
@@ -85,7 +75,7 @@ func createRealUnits(cfg *config.SubTaskConfig, etcdClient *clientv3.Client, wor
 			us = append(us, loader.NewLightning(cfg, etcdClient, workerName))
 		}
 	case config.ModeIncrement:
-		us = append(us, syncer.NewSyncer(cfg, etcdClient, notifier))
+		us = append(us, syncer.NewSyncer(cfg, etcdClient, relay))
 	default:
 		log.L().Error("unsupported task mode", zap.String("subtask", cfg.Name), zap.String("task mode", cfg.Mode))
 	}
@@ -119,8 +109,6 @@ type SubTask struct {
 	etcdClient *clientv3.Client
 
 	workerName string
-
-	notifier relay.EventNotifier
 }
 
 // NewSubTask is subtask initializer
@@ -143,15 +131,14 @@ func NewSubTaskWithStage(cfg *config.SubTaskConfig, stage pb.Stage, etcdClient *
 		cancel:     cancel,
 		etcdClient: etcdClient,
 		workerName: workerName,
-		notifier:   &relayNotifier{ch: make(chan interface{}, 1)},
 	}
 	updateTaskMetric(st.cfg.Name, st.cfg.SourceID, st.stage, st.workerName)
 	return &st
 }
 
 // initUnits initializes the sub task processing units.
-func (st *SubTask) initUnits() error {
-	st.units = createUnits(st.cfg, st.etcdClient, st.workerName, st.notifier)
+func (st *SubTask) initUnits(relay relay.Process) error {
+	st.units = createUnits(st.cfg, st.etcdClient, st.workerName, relay)
 	if len(st.units) < 1 {
 		return terror.ErrWorkerNoAvailUnits.Generate(st.cfg.Name, st.cfg.Mode)
 	}
@@ -211,15 +198,15 @@ func (st *SubTask) initUnits() error {
 
 // Run runs the sub task.
 // TODO: check concurrent problems.
-func (st *SubTask) Run(expectStage pb.Stage) {
+func (st *SubTask) Run(expectStage pb.Stage, relay relay.Process) {
 	if st.Stage() == pb.Stage_Finished || st.Stage() == pb.Stage_Running {
 		st.l.Warn("prepare to run a subtask with invalid stage",
 			zap.Stringer("current stage", st.Stage()),
 			zap.Stringer("expected stage", expectStage))
 		return
 	}
 
-	if err := st.initUnits(); err != nil {
+	if err := st.initUnits(relay); err != nil {
 		st.l.Error("fail to initial subtask", log.ShortError(err))
 		st.fail(err)
 		return
@@ -511,9 +498,9 @@ func (st *SubTask) Pause() error {
 
 // Resume resumes the paused sub task
 // TODO: similar to Run, refactor later.
-func (st *SubTask) Resume() error {
+func (st *SubTask) Resume(relay relay.Process) error {
 	if !st.initialized.Load() {
-		st.Run(pb.Stage_Running)
+		st.Run(pb.Stage_Running, relay)
 		return nil
 	}
 
@@ -714,7 +701,7 @@ func (st *SubTask) fail(err error) {
 }
 
 // HandleError handle error for syncer unit.
-func (st *SubTask) HandleError(ctx context.Context, req *pb.HandleWorkerErrorRequest) error {
+func (st *SubTask) HandleError(ctx context.Context, req *pb.HandleWorkerErrorRequest, relay relay.Process) error {
 	syncUnit, ok := st.currUnit.(*syncer.Syncer)
 	if !ok {
 		return terror.ErrWorkerOperSyncUnitOnly.Generate(st.currUnit.Type())
@@ -726,7 +713,7 @@ func (st *SubTask) HandleError(ctx context.Context, req *pb.HandleWorkerErrorReq
 	}
 
 	if st.Stage() == pb.Stage_Paused {
-		err = st.Resume()
+		err = st.Resume(relay)
 	}
 	return err
 }
@@ -738,11 +725,3 @@ func updateTaskMetric(task, sourceID string, stage pb.Stage, workerName string)
 		taskState.WithLabelValues(task, sourceID, workerName).Set(float64(stage))
 	}
 }
-
-func (st *SubTask) relayNotify() {
-	// skip if there's pending notify
-	select {
-	case st.notifier.Notified() <- struct{}{}:
-	default:
-	}
-}