cdc/sink: Sink manager manage checkpoint per table

cdc/sink/manager.go

@@ -23,7 +23,7 @@ import (
 	"github.com/pingcap/errors"
 	"github.com/pingcap/log"
 	"github.com/pingcap/ticdc/cdc/model"
-	redo "github.com/pingcap/ticdc/cdc/redo"
+	"github.com/pingcap/ticdc/cdc/redo"
 	"github.com/prometheus/client_golang/prometheus"
 	"go.uber.org/zap"
 )
@@ -34,10 +34,11 @@ const (
 
 // Manager manages table sinks, maintains the relationship between table sinks and backendSink.
 type Manager struct {
-	backendSink  Sink
-	checkpointTs model.Ts
-	tableSinks   map[model.TableID]*tableSink
-	tableSinksMu sync.Mutex
+	backendSink            Sink
+	tableCheckpointTsMap   sync.Map
+	tableSinks             map[model.TableID]*tableSink
+	tableSinksMu           sync.Mutex
+	changeFeedCheckpointTs uint64
 
 	flushMu  sync.Mutex
 	flushing int64
@@ -57,7 +58,7 @@ func NewManager(
 	drawbackChan := make(chan drawbackMsg, 16)
 	return &Manager{
 		backendSink:               newBufferSink(ctx, backendSink, errCh, checkpointTs, drawbackChan),
-		checkpointTs:              checkpointTs,
+		changeFeedCheckpointTs:    checkpointTs,
 		tableSinks:                make(map[model.TableID]*tableSink),
 		drawbackChan:              drawbackChan,
 		captureAddr:               captureAddr,
@@ -90,15 +91,15 @@ func (m *Manager) Close(ctx context.Context) error {
 	return m.backendSink.Close(ctx)
 }
 
-func (m *Manager) getMinEmittedTs() model.Ts {
+func (m *Manager) getMinEmittedTs(tableID model.TableID) model.Ts {
 	m.tableSinksMu.Lock()
 	defer m.tableSinksMu.Unlock()
 	if len(m.tableSinks) == 0 {
-		return m.getCheckpointTs()
+		return m.getCheckpointTs(tableID)
 	}
 	minTs := model.Ts(math.MaxUint64)
-	for _, tableSink := range m.tableSinks {
-		resolvedTs := tableSink.getResolvedTs()
+	for _, tblSink := range m.tableSinks {
+		resolvedTs := tblSink.getResolvedTs()
 		if minTs > resolvedTs {
 			minTs = resolvedTs
 		}
@@ -111,19 +112,19 @@ func (m *Manager) flushBackendSink(ctx context.Context, tableID model.TableID) (
 	// which will cause a lot of lock contention and blocking in high concurrency cases.
 	// So here we use flushing as a lightweight lock to improve the lock competition problem.
 	if !atomic.CompareAndSwapInt64(&m.flushing, 0, 1) {
-		return m.getCheckpointTs(), nil
+		return m.getCheckpointTs(tableID), nil
 	}
 	m.flushMu.Lock()
 	defer func() {
 		m.flushMu.Unlock()
 		atomic.StoreInt64(&m.flushing, 0)
 	}()
-	minEmittedTs := m.getMinEmittedTs()
+	minEmittedTs := m.getMinEmittedTs(tableID)
 	checkpointTs, err := m.backendSink.FlushRowChangedEvents(ctx, tableID, minEmittedTs)
 	if err != nil {
-		return m.getCheckpointTs(), errors.Trace(err)
+		return m.getCheckpointTs(tableID), errors.Trace(err)
 	}
-	atomic.StoreUint64(&m.checkpointTs, checkpointTs)
+	m.tableCheckpointTsMap.Store(tableID, checkpointTs)
 	return checkpointTs, nil
 }
 
@@ -145,8 +146,13 @@ func (m *Manager) destroyTableSink(ctx context.Context, tableID model.TableID) e
 	return m.backendSink.Barrier(ctx, tableID)
 }
 
-func (m *Manager) getCheckpointTs() uint64 {
-	return atomic.LoadUint64(&m.checkpointTs)
+func (m *Manager) getCheckpointTs(tableID model.TableID) uint64 {
+	checkPoints, ok := m.tableCheckpointTsMap.Load(tableID)
+	if ok {
+		return checkPoints.(uint64)
+	}
+	// this function is call when something unexpected, return change level checkpoint is safe
+	return m.changeFeedCheckpointTs
 }
 
 type drawbackMsg struct {

cdc/sink/manager_test.go

@@ -35,9 +35,17 @@ var _ = check.Suite(&managerSuite{})
 
 type checkSink struct {
 	*check.C
-	rows           []*model.RowChangedEvent
+	rows           map[model.TableID][]*model.RowChangedEvent
 	rowsMu         sync.Mutex
-	lastResolvedTs uint64
+	lastResolvedTs map[model.TableID]uint64
+}
+
+func newCheckSink(c *check.C) *checkSink {
+	return &checkSink{
+		C:              c,
+		rows:           make(map[model.TableID][]*model.RowChangedEvent),
+		lastResolvedTs: make(map[model.TableID]uint64),
+	}
 }
 
 func (c *checkSink) Initialize(ctx context.Context, tableInfo []*model.SimpleTableInfo) error {
@@ -47,7 +55,9 @@ func (c *checkSink) Initialize(ctx context.Context, tableInfo []*model.SimpleTab
 func (c *checkSink) EmitRowChangedEvents(ctx context.Context, rows ...*model.RowChangedEvent) error {
 	c.rowsMu.Lock()
 	defer c.rowsMu.Unlock()
-	c.rows = append(c.rows, rows...)
+	for _, row := range rows {
+		c.rows[row.Table.TableID] = append(c.rows[row.Table.TableID], row)
+	}
 	return nil
 }
 
@@ -59,20 +69,21 @@ func (c *checkSink) FlushRowChangedEvents(ctx context.Context, tableID model.Tab
 	c.rowsMu.Lock()
 	defer c.rowsMu.Unlock()
 	var newRows []*model.RowChangedEvent
-	for _, row := range c.rows {
-		if row.CommitTs <= c.lastResolvedTs {
-			return c.lastResolvedTs, errors.Errorf("commit-ts(%d) is not greater than lastResolvedTs(%d)", row.CommitTs, c.lastResolvedTs)
+	rows := c.rows[tableID]
+	for _, row := range rows {
+		if row.CommitTs <= c.lastResolvedTs[tableID] {
+			return c.lastResolvedTs[tableID], errors.Errorf("commit-ts(%d) is not greater than lastResolvedTs(%d)", row.CommitTs, c.lastResolvedTs)
 		}
 		if row.CommitTs > resolvedTs {
 			newRows = append(newRows, row)
 		}
 	}
 
-	c.Assert(c.lastResolvedTs, check.LessEqual, resolvedTs)
-	c.lastResolvedTs = resolvedTs
-	c.rows = newRows
+	c.Assert(c.lastResolvedTs[tableID], check.LessEqual, resolvedTs)
+	c.lastResolvedTs[tableID] = resolvedTs
+	c.rows[tableID] = newRows
 
-	return c.lastResolvedTs, nil
+	return c.lastResolvedTs[tableID], nil
 }
 
 func (c *checkSink) EmitCheckpointTs(ctx context.Context, ts uint64) error {
@@ -92,7 +103,7 @@ func (s *managerSuite) TestManagerRandom(c *check.C) {
 	ctx, cancel := context.WithCancel(context.Background())
 	defer cancel()
 	errCh := make(chan error, 16)
-	manager := NewManager(ctx, &checkSink{C: c}, errCh, 0, "", "")
+	manager := NewManager(ctx, newCheckSink(c), errCh, 0, "", "")
 	defer manager.Close(ctx)
 	goroutineNum := 10
 	rowNum := 100
@@ -147,7 +158,7 @@ func (s *managerSuite) TestManagerAddRemoveTable(c *check.C) {
 	ctx, cancel := context.WithCancel(context.Background())
 	defer cancel()
 	errCh := make(chan error, 16)
-	manager := NewManager(ctx, &checkSink{C: c}, errCh, 0, "", "")
+	manager := NewManager(ctx, newCheckSink(c), errCh, 0, "", "")
 	defer manager.Close(ctx)
 	goroutineNum := 200
 	var wg sync.WaitGroup
@@ -234,7 +245,7 @@ func (s *managerSuite) TestManagerDestroyTableSink(c *check.C) {
 	defer cancel()
 
 	errCh := make(chan error, 16)
-	manager := NewManager(ctx, &checkSink{C: c}, errCh, 0, "", "")
+	manager := NewManager(ctx, newCheckSink(c), errCh, 0, "", "")
 	defer manager.Close(ctx)
 
 	tableID := int64(49)
@@ -255,7 +266,7 @@ func (s *managerSuite) TestManagerDestroyTableSink(c *check.C) {
 func BenchmarkManagerFlushing(b *testing.B) {
 	ctx, cancel := context.WithCancel(context.Background())
 	errCh := make(chan error, 16)
-	manager := NewManager(ctx, &checkSink{}, errCh, 0, "", "")
+	manager := NewManager(ctx, newCheckSink(nil), errCh, 0, "", "")
 
 	// Init table sinks.
 	goroutineNum := 2000

cdc/sink/table_sink.go

@@ -71,7 +71,7 @@ func (t *tableSink) FlushRowChangedEvents(ctx context.Context, tableID model.Tab
 
 	err := t.manager.backendSink.EmitRowChangedEvents(ctx, resolvedRows...)
 	if err != nil {
-		return t.manager.getCheckpointTs(), errors.Trace(err)
+		return t.manager.getCheckpointTs(tableID), errors.Trace(err)
 	}
 	atomic.StoreUint64(&t.emittedTs, resolvedTs)
 	ckpt, err := t.flushRedoLogs(ctx, resolvedTs)
@@ -85,7 +85,7 @@ func (t *tableSink) flushRedoLogs(ctx context.Context, resolvedTs uint64) (uint6
 	if t.redoManager.Enabled() {
 		err := t.redoManager.FlushLog(ctx, t.tableID, resolvedTs)
 		if err != nil {
-			return t.manager.getCheckpointTs(), err
+			return t.manager.getCheckpointTs(t.tableID), err
 		}
 	}
 	return 0, nil