kafka-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Dana Powers <dana.pow...@gmail.com>
Subject RE: KafkaProducer block on send
Date Fri, 08 Apr 2016 14:31:11 GMT
The prior discussion explained:

(1) The code you point to blocks for a maximum of max.block.ms, which is
user configurable. It does not block indefinitely with no user control as
you suggest. You are free to configure this to 0 if you like at it will not
block at all. Have you tried this like I suggested before?

(2) Even if you convinced people to remove waitOnMetadata, the send method
*still* blocks on memory back pressure (also configured by max.block.ms).
This is for good reason:

while True:

Can quickly devour all of you local memory and crash your process if the
outflow rate decreases, say if brokers go down or network partition occurs.

I totally agree with Oleg.

As documentation says the producers send data in an asynchronous way and it
is enforced by the send method signature with a Future returned.
It can't block indefinitely without returning to the caller.
I'm mean, you can decide that the code inside the send method blocks
indefinitely but in an "asynchronous way", it should first return a Future
to the caller that can handle it.


Paolo PatiernoSenior Software Engineer (IoT) @ Red Hat
Microsoft MVP on Windows Embedded & IoTMicrosoft Azure Advisor
Twitter : @ppatierno
Linkedin : paolopatierno
Blog : DevExperience

> Subject: KafkaProducer block on send
> From: ozhurakousky@hortonworks.com
> To: users@kafka.apache.org
> Date: Thu, 7 Apr 2016 13:04:49 +0000
> I know it’s been discussed before, but that conversation never really
concluded with any reasonable explanation, so I am bringing it up again as
I believe this is a bug that would need to be fixed in some future release.
> Can someone please explain the rational for the following code in
> @Override
> public Future<RecordMetadata> send(ProducerRecord<K, V> record, Callback
callback) {
>         try {
>             // first make sure the metadata for the topic is available
>             long waitedOnMetadataMs = waitOnMetadata(record.topic(),
> . . .
> }
> By definition the method that returns Future implies that caller decides
how long to wait for the completion via Future.get(TIMETOWAIT). In this
case there is an explicit blocking call (waitOnMetadata), that can hang
infinitely (regardless of the reasons) which essentially results in user’s
code deadlock since the Future may never be returned in the first place.
> Thoughts?
> Oleg

  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message