etcd_worker: batch etcd patch (pingcap#3277) (pingcap#3393)

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

@@ -281,7 +281,8 @@ 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/tiflow/cdc/puller/sorter"
 	"github.com/pingcap/tiflow/cdc/sink"
 	"github.com/pingcap/tiflow/pkg/config"
+	"github.com/pingcap/tiflow/pkg/orchestrator"
 	"github.com/prometheus/client_golang/prometheus"
 )
 
@@ -37,6 +38,7 @@ func init() {
 	sink.InitMetrics(registry)
 	entry.InitMetrics(registry)
 	sorter.InitMetrics(registry)
+	orchestrator.InitMetrics(registry)
 	if config.NewReplicaImpl {
 		processor.InitMetrics(registry)
 		tablepipeline.InitMetrics(registry)

cdc/task_test.go

@@ -156,7 +156,9 @@ func (s *taskSuite) TestWatch(c *check.C) {
 	}
 
 	// Watch with a normal context
-	ch := s.w.Watch(context.Background())
+	ctx, cancel = context.WithCancel(context.Background())
+	defer cancel()
+	ch := s.w.Watch(ctx)
 
 	// Trigger the ErrCompacted error
 	c.Assert(failpoint.Enable("github.com/pingcap/tiflow/cdc.restart_task_watch", "50%off"), check.IsNil)

errors.toml

@@ -206,6 +206,16 @@ 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

go.mod

@@ -8,6 +8,7 @@ require (
 	github.com/Shopify/sarama v1.27.2
 	github.com/alecthomas/template v0.0.0-20190718012654-fb15b899a751
 	github.com/apache/pulsar-client-go v0.1.1
+	github.com/benbjohnson/clock v1.1.0
 	github.com/bradleyjkemp/grpc-tools v0.2.5
 	github.com/cenkalti/backoff v2.2.1+incompatible
 	github.com/chzyer/readline v0.0.0-20180603132655-2972be24d48e