[
https://issues.apache.org/jira/browse/KAFKA-8178?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17522184#comment-17522184
]
Jan Doms commented on KAFKA-8178:
---------------------------------
I understand people programming in a reactive style not wanting their threads
to be blocked by kafka, OTOH backpressure is important and often desirable
(e.g. to limit the amount of memory occupied by a process).
As such I actually consider this documented(!) behavior a feature, which users
may depend on.
See e.g.
[https://kafka.apache.org/23/javadoc/index.html?org/apache/kafka/clients/producer/KafkaProducer.html]
and
[https://docs.confluent.io/platform/current/installation/configuration/producer-configs.html#producerconfigs_max.block.ms]
.
> The send is asynchronous and this method will return immediately once the
> record has been stored in the buffer of records waiting to be sent.
I want to explicitly ask to not change this behavior for existing users. If the
changed behavior is to be implemented it should probably be done as a new
overload of the send method allowing users to opt in to it.
> KafkaProducer#send(ProducerRecord,Callback) may block for up to 60 seconds
> --------------------------------------------------------------------------
>
> Key: KAFKA-8178
> URL: https://issues.apache.org/jira/browse/KAFKA-8178
> Project: Kafka
> Issue Type: Bug
> Components: clients, producer
> Reporter: Sergei Egorov
> Priority: Major
>
> Hello. I was running reactor-kafka with [the BlockHound
> agent|https://github.com/reactor/BlockHound] (you can see the progress
> [here|https://github.com/reactor/reactor-kafka/pull/75] and even run it
> yourself) and it detected a very dangerous blocking call in
> KafkaProducer#send(ProducerRecord,Callback) which is supposed to be async:
> {code:java}
> java.lang.Error: Blocking call! java.lang.Object#wait
> at reactor.BlockHound$Builder.lambda$new$0(BlockHound.java:154)
> at reactor.BlockHound$Builder.lambda$install$8(BlockHound.java:254)
> at reactor.BlockHoundRuntime.checkBlocking(BlockHoundRuntime.java:43)
> at java.lang.Object.wait(Object.java)
> at org.apache.kafka.clients.Metadata.awaitUpdate(Metadata.java:181)
> at
> org.apache.kafka.clients.producer.KafkaProducer.waitOnMetadata(KafkaProducer.java:938)
> at
> org.apache.kafka.clients.producer.KafkaProducer.doSend(KafkaProducer.java:823)
> at
> org.apache.kafka.clients.producer.KafkaProducer.send(KafkaProducer.java:803)
> {code}
> it blocks for up to "maxBlockTimeMs" (60 seconds by default)
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
