dont modify connected state on errors

Signed-off-by: Vasiliy Tolstov <v.tolstov@unistack.org>

broker.go

@@ -27,43 +27,45 @@ var (
 
 func (m *hookEvent) OnGroupManageError(err error) {
 	if err != nil {
-		m.connected.Store(0)
+		// m.connected.Store(0)
-		if m.fatalOnError {
+		// if m.fatalOnError {
-			m.log.Fatal(context.TODO(), "kgo.OnGroupManageError", err)
+		m.log.Error(context.TODO(), "kgo.OnGroupManageError", err)
-		}
+		//}
 	}
 }
 
 func (m *hookEvent) OnBrokerConnect(_ kgo.BrokerMetadata, _ time.Duration, _ net.Conn, err error) {
 	if err != nil {
-		m.connected.Store(0)
+		// m.connected.Store(0)
-		if m.fatalOnError {
+		// if m.fatalOnError {
-			m.log.Fatal(context.TODO(), "kgo.OnBrokerConnect", err)
+		m.log.Error(context.TODO(), "kgo.OnBrokerConnect", err)
-		}
+		//}
 	}
 }
 
 func (m *hookEvent) OnBrokerDisconnect(_ kgo.BrokerMetadata, _ net.Conn) {
-	m.connected.Store(0)
+	// m.connected.Store(0)
 }
 
 func (m *hookEvent) OnBrokerWrite(_ kgo.BrokerMetadata, _ int16, _ int, _ time.Duration, _ time.Duration, err error) {
 	if err != nil {
-		m.connected.Store(0)
+		// m.connected.Store(0)
-		if m.fatalOnError {
+		// if m.fatalOnError {
-			m.log.Fatal(context.TODO(), "kgo.OnBrokerWrite", err)
+		m.log.Error(context.TODO(), "kgo.OnBrokerWrite", err)
-		}
+		//}
 	}
 }
 
 func (m *hookEvent) OnBrokerRead(_ kgo.BrokerMetadata, _ int16, _ int, _ time.Duration, _ time.Duration, err error) {
 	if err != nil {
-		m.connected.Store(0)
+		// m.connected.Store(0)
+		m.log.Error(context.TODO(), "kgo.OnBrokerRead", err)
 	}
 }
 
 func (m *hookEvent) OnProduceRecordUnbuffered(_ *kgo.Record, err error) {
 	if err != nil {
-		m.connected.Store(0)
+		// m.connected.Store(0)
+		m.log.Error(context.TODO(), "kgo.OnProduceRecordUnbuffered", err)
 	}
 }

kgo.go

@@ -19,6 +19,7 @@ import (
 	"go.unistack.org/micro/v3/metadata"
 	"go.unistack.org/micro/v3/semconv"
 	"go.unistack.org/micro/v3/tracer"
+	mjitter "go.unistack.org/micro/v3/util/jitter"
 	mrand "go.unistack.org/micro/v3/util/rand"
 )
 
@@ -66,6 +67,8 @@ type Broker struct {
 
 	sync.RWMutex
 	init bool
+
+	done chan struct{}
 }
 
 func (r *Broker) Live() bool {
@@ -143,6 +146,25 @@ func (k *Broker) connect(ctx context.Context, opts ...kgo.Opt) (*kgo.Client, *ho
 			return nil, nil, err
 		}
 		k.connected.Store(1)
+
+		if fatalOnError {
+			go func() {
+				c := 3
+				n := 0
+				tc := mjitter.NewTicker(500*time.Millisecond, 1*time.Second)
+				defer tc.Stop()
+				for range tc.C {
+					if k.connected.Load() == 0 {
+						if n > c {
+							k.opts.Logger.Fatal(context.Background(), "broker fatal error")
+						}
+						n++
+					} else {
+						n = 0
+					}
+				}
+			}()
+		}
 		return c, htracer, nil
 	}
 }
@@ -204,6 +226,7 @@ func (k *Broker) Disconnect(ctx context.Context) error {
 	}
 
 	k.connected.Store(0)
+	close(k.done)
 	return nil
 }
 
@@ -391,12 +414,6 @@ func (k *Broker) Subscribe(ctx context.Context, topic string, handler broker.Han
 		}
 	}
 
-	if options.Context != nil {
-		if v, ok := options.Context.Value(fatalOnErrorKey{}).(bool); ok && v {
-			fatalOnError = v
-		}
-	}
-
 	sub := &Subscriber{
 		topic:        topic,
 		opts:         options,
@@ -492,5 +509,6 @@ func NewBroker(opts ...broker.Option) *Broker {
 		connected: &atomic.Uint32{},
 		opts:      options,
 		kopts:     kopts,
+		done:      make(chan struct{}),
 	}
 }

subscriber.go

@@ -119,7 +119,7 @@ func (s *Subscriber) poll(ctx context.Context) {
 
 				s.Lock()
 				for p, l := range lmap {
-					s.kopts.Meter.Counter(semconv.BrokerGroupLag, "topic", s.topic, "group", s.opts.Group, "partition", strconv.Itoa(int(p)), "lag", strconv.Itoa(int(l.Lag)))
+					s.kopts.Meter.Counter(semconv.BrokerGroupLag, "topic", s.topic, "group", s.opts.Group, "partition", strconv.Itoa(int(p))).Set(uint64(l.Lag))
 				}
 				s.Unlock()
 
@@ -160,7 +160,10 @@ func (s *Subscriber) killConsumers(ctx context.Context, lost map[string][]int32)
 	for topic, partitions := range lost {
 		for _, partition := range partitions {
 			tps := tp{topic, partition}
-			pc := s.consumers[tps]
+			pc, ok := s.consumers[tps]
+			if !ok {
+				continue
+			}
 			delete(s.consumers, tps)
 			close(pc.quit)
 			if s.kopts.Logger.V(logger.DebugLevel) {
@@ -174,7 +177,7 @@ func (s *Subscriber) killConsumers(ctx context.Context, lost map[string][]int32)
 
 func (s *Subscriber) autocommit(_ *kgo.Client, _ *kmsg.OffsetCommitRequest, _ *kmsg.OffsetCommitResponse, err error) {
 	if err != nil {
-		s.connected.Store(0)
+		//		s.connected.Store(0)
 		if s.fatalOnError {
 			s.kopts.Logger.Fatal(context.TODO(), "kgo.AutoCommitCallback error", err)
 		}
@@ -276,7 +279,7 @@ func (pc *consumer) consume() {
 								pc.c.MarkCommitRecords(record)
 							} else {
 								eventPool.Put(p)
-								pc.connected.Store(0)
+								//								pc.connected.Store(0)
 								pc.kopts.Logger.Fatal(pc.kopts.Context, "[kgo] ErrLostMessage wtf?")
 								return
 							}
@@ -293,7 +296,7 @@ func (pc *consumer) consume() {
 						pc.kopts.Meter.Summary(semconv.SubscribeMessageLatencyMicroseconds, "endpoint", record.Topic, "topic", record.Topic).Update(te.Seconds())
 						pc.kopts.Meter.Histogram(semconv.SubscribeMessageDurationSeconds, "endpoint", record.Topic, "topic", record.Topic).Update(te.Seconds())
 						eventPool.Put(p)
-						pc.connected.Store(0)
+						//						pc.connected.Store(0)
 						pc.kopts.Logger.Fatal(pc.kopts.Context, "[kgo] Unmarshal err not handled wtf?")
 						sp.Finish()
 						return
@@ -331,7 +334,7 @@ func (pc *consumer) consume() {
 					pc.c.MarkCommitRecords(record)
 				} else {
 					eventPool.Put(p)
-					pc.connected.Store(0)
+					//					pc.connected.Store(0)
 					pc.kopts.Logger.Fatal(pc.kopts.Context, "[kgo] ErrLostMessage wtf?")
 					sp.SetStatus(tracer.SpanStatusError, "ErrLostMessage")
 					sp.Finish()