[BUG] pulsar.(*producer).Send() panic when topic is deleted

[Violet-eve-garden]

Expected behavior

When topic is deleted, should return Topic Not Found error instead of panic.

Actual behavior

Steps to reproduce

1. create topic;
2. create producer and send messages in loop;
3. force delete topic while producer is connected;
4. wait for a period of time, you can see the panic;

System configuration

Pulsar version: 0.11.1

[Violet-eve-garden]

pulsar-client-go/pulsar/producer_impl.go

Lines 204 to 217 in 60f2908

	if oldProducers != nil {
	if oldNumPartitions == newNumPartitions {
	p.log.Debug("Number of partitions in topic has not changed")
	return nil
	}
	p.log.WithField("old_partitions", oldNumPartitions).
	WithField("new_partitions", newNumPartitions).
	Info("Changed number of partitions in topic")
	}
	p.producers = make([]Producer, newNumPartitions)

• The bug may be caused by the runBackgroundPartitionDiscovery behavior.
• When topic is deleted or the number of partitions is changed, it unsafely change p.producers which is the member of struct producer.

[gunli]

Hi @Violet-eve-garden TopicIsNotFound is handled in the latest master branch, would you please try with the latest code?

You can update it by:
go get -u github.com/apache/pulsar-client-go@bd11581867c88f93e4b6f247d82914f9eb4ee476