etcd_worker: batch etcd patch (#3277) (#3390)

cdc/capture.go

@@ -162,7 +162,7 @@ func (c *Capture) Run(ctx context.Context) (err error) {
 			return errors.Trace(err)
 		}
 		log.Info("start to listen processor task...")
-		if err := etcdWorker.Run(ctx, c.session, 200*time.Millisecond); err != nil {
+		if err := etcdWorker.Run(ctx, c.session, 200*time.Millisecond, c.info.AdvertiseAddr); err != nil {
 			// We check ttl of lease instead of check `session.Done`, because
 			// `session.Done` is only notified when etcd client establish a
 			// new keepalive request, there could be a time window as long as

cdc/capture/capture.go

@@ -280,7 +280,8 @@ func (c *Capture) runEtcdWorker(ctx cdcContext.Context, reactor orchestrator.Rea
 	if err != nil {
 		return errors.Trace(err)
 	}
-	if err := etcdWorker.Run(ctx, c.session, timerInterval); err != nil {
+	captureAddr := c.info.AdvertiseAddr
+	if err := etcdWorker.Run(ctx, c.session, timerInterval, captureAddr); err != nil {
 		// We check ttl of lease instead of check `session.Done`, because
 		// `session.Done` is only notified when etcd client establish a
 		// new keepalive request, there could be a time window as long as

cdc/metrics.go

@@ -23,6 +23,7 @@ import (
 	"github.com/pingcap/ticdc/cdc/puller/sorter"
 	"github.com/pingcap/ticdc/cdc/sink"
 	"github.com/pingcap/ticdc/pkg/config"
+	"github.com/pingcap/ticdc/pkg/orchestrator"
 	"github.com/prometheus/client_golang/prometheus"
 )
 
@@ -36,6 +37,7 @@ func init() {
 	puller.InitMetrics(registry)
 	sink.InitMetrics(registry)
 	entry.InitMetrics(registry)
+	orchestrator.InitMetrics(registry)
 	sorter.InitMetrics(registry)
 	if config.NewReplicaImpl {
 		processor.InitMetrics(registry)

cdc/model/reactor_state.go

@@ -100,6 +100,7 @@ func (s *GlobalReactorState) Update(key util.EtcdKey, value []byte, _ bool) erro
 }
 
 // GetPatches implements the ReactorState interface
+// Every []DataPatch slice in [][]DataPatch slice is the patches of a ChangefeedReactorState
 func (s *GlobalReactorState) GetPatches() [][]orchestrator.DataPatch {
 	pendingPatches := s.pendingPatches
 	for _, changefeedState := range s.Changefeeds {

errors.toml

@@ -196,6 +196,16 @@ error = '''
 the etcd txn should be aborted and retried immediately
 '''
 
+["CDC:ErrEtcdTxnOpsExceed"]
+error = '''
+patch ops:%d of a single changefeed exceed etcd txn max ops:%d
+'''
+
+["CDC:ErrEtcdTxnSizeExceed"]
+error = '''
+patch size:%d of a single changefeed exceed etcd txn max size:%d
+'''
+
 ["CDC:ErrEventFeedAborted"]
 error = '''
 single event feed aborted