I wanted to follow up on this as I think it could help others.
Oleg, I created a patch for PutKafka/PublishKafka that removes the limit on the
size of the message which can be sent. I’d don’t know that it’s what you want
to solve the JIRA, but I’m thinking not. I was just an expediency for me.
However, if you are interested, let me know.
I discovered that GetKafka/ConsumeKafka can read messages > 1MiB if you add a
dynamic property to the processor. fetch.message.max.bytes needs to be set for
GetKafka and max.partition.fetch.bytes needs to be set for consume Kafka.
Cheers,
Chris McDermott
Remote Business Analytics
STaTS/StoreFront Remote
HPE Storage
Hewlett Packard Enterprise
Mobile: +1 978-697-5315
On 8/21/16, 6:45 PM, "Oleg Zhurakousky" <ozhurakou...@hortonworks.com> wrote:
Chris
The “. . . querying the max.message.bytes. . .” is exactly what I had in
mind. As you mentioned earlier, that is where the default value came from in
the first place. So, yes that is what we’re going to link together.
Cheers
Oleg
> On Aug 21, 2016, at 6:15 PM, McDermott, Chris Kevin (MSDU -
STaTS/StorefrontRemote) <chris.mcderm...@hpe.com> wrote:
>
> Thanks for getting back, Oleg. I’d be happy to send demarcated messages;
the problem is reassembling them at the other end of the pipe. I’ve done a lot
of searching for techniques to do this. They all seem to have major draw backs
in terms of reliable message delivery or in terms of garbage collection (not in
the sense of Java GC, but cleanup of files bounced off of a shared file
system.) The nice thing about Kafka is its atomic, it has replicated delivery,
and guaranteed GC semantics. My use case has fairly low throughput
requirements (thousands, not millions of TPM) where most messages are fairly
small but a few are larger.
>
> It would be nice if the Kafka client could learn the max message size
from Kafka itself by querying the max.message.bytes on the topic, rather than
have the flow designer be required to set it on the producer Processors. For
now, though I’d be happy going back to the old behavior where its set on the
producer Processors. On the flip side I am also concerned that the clients
(GetKafka and ConsumeKafka) do not expose a max message parameter. That will
be equally problematic.
>
> Anyway, enough of my blathering.
>
> Yours and the communities help is greatly appreciated.
>
> Thanks,
>
> Chris McDermott
>
> Remote Business Analytics
> STaTS/StoreFront Remote
> HPE Storage
> Hewlett Packard Enterprise
> Mobile: +1 978-697-5315
>
>
>
> On 8/21/16, 3:43 PM, "Oleg Zhurakousky" <ozhurakou...@hortonworks.com>
wrote:
>
> Chris
>
> Sorry for getting late to this
>
> This indeed was intentional, primarily to follow Kafka’s best
practices where Kafka was not designed to be a general data transfer system of
arbitrary size but rather “manageable” size. Also, as you know we have ability
to demarcate messages essentially allowing you to send a very large FlowFile
that will be parsed into chunks where each chunk will end up to be a separate
Kafka message.
> That said, we did consider that at some point we may expose it as
configuration property and it seems like the time has come.
>
> Cheers
> Oleg
>
>> On Aug 20, 2016, at 7:03 PM, McDermott, Chris Kevin (MSDU -
STaTS/StorefrontRemote) <chris.mcderm...@hpe.com> wrote:
>>
>> Jira is https://issues.apache.org/jira/browse/NIFI-2614.
>>
>> Thanks,
>>
>> Chris McDermott
>>
>> Remote Business Analytics
>> STaTS/StoreFront Remote
>> HPE Storage
>> Hewlett Packard Enterprise
>> Mobile: +1 978-697-5315
>>
>>
>>
>> On 8/20/16, 6:57 PM, "McDermott, Chris Kevin (MSDU -
STaTS/StorefrontRemote)" <chris.mcderm...@hpe.com> wrote:
>>
>> I’ll raise a JIRA, Joe.
>>
>> Thanks,
>>
>> Chris McDermott
>>
>> Remote Business Analytics
>> STaTS/StoreFront Remote
>> HPE Storage
>> Hewlett Packard Enterprise
>> Mobile: +1 978-697-5315
>>
>>
>>
>> On 8/20/16, 6:52 PM, "Joe Witt" <joe.w...@gmail.com> wrote:
>>
>> If no jira is raised sooner I'll raise one and get it sorted.
>>
>> On Aug 20, 2016 6:40 PM, "Andrew Psaltis"
<psaltis.and...@gmail.com> wrote:
>>
>>> Hi Chris,
>>> Sorry for not catching that code path. I am not sure if it is actually a
>>> regression as I took a look at the 1.0.0-BETA code and it matches the
>>> 0.7.0, specifically this comment block:
>>>
>>> /*
>>> * We're using the default value from Kafka. We are using it to control
the
>>> * message size before it goes to to Kafka thus limiting possibility of a
>>> * late failures in Kafka client.
>>> */
>>>
>>> found at[1] leads me to believe it was intentional and not a regression.
>>> Looking at the 0.6.1 release code it appears that PutKafka used a
default
>>> of 5 MB [2].
>>>
>>> I can speculate on the reasoning behind it, however, I will refrain from
>>> opining on it as I was not involved in any of the conversations related
to
>>> the change and enforcement of the 1 MB max.
>>>
>>> [1]
>>> https://github.com/apache/nifi/blob/rel/nifi-1.0.0-BETA-
>>> official/nifi-nar-bundles/nifi-kafka-bundle/nifi-kafka-
>>> processors/src/main/java/org/apache/nifi/processors/kafka/
>>> PublishingContext.java#L36-L41
>>> [2]
>>> https://github.com/apache/nifi/blob/rel/nifi-0.6.1/nifi-
>>> nar-bundles/nifi-kafka-bundle/nifi-kafka-processors/src/
>>> main/java/org/apache/nifi/processors/kafka/PutKafka.java#L169-L176
>>>
>>> Thanks,
>>> Andrew
>>>
>>> On Sat, Aug 20, 2016 at 6:09 PM, McDermott, Chris Kevin (MSDU -
>>> STaTS/StorefrontRemote) <chris.mcderm...@hpe.com> wrote:
>>>
>>>> Thanks, Andrew.
>>>>
>>>> I’ve set all of the right broker configs to allow larger messages.
>>>> Believe me I spent a lot of time banging my head against the wall
>>> thinking
>>>> that the broker and topic configs were wrong.
>>>>
>>>> PublisingKafka uses PublishingContext. That class has bean property
>>>> called maxRequestSize, which defaults to 1048576. As far as I can tell
>>> the
>>>> setMaxRequestSize() method is never called (except by some test code.)
>>>> KafkaPublisher.publish() calls Max Record Size.getMaxRequestSize() and
>>>> passes the result to the constructor for StreamDemarcator. The
publish
>>>> method then calls the StreamDemarcator. getNextToken(), which in turns
>>>> calls StreamDemarcator.fill() which compares the stream position
against
>>>> the maxRequestSize and throws the exception with this line.
>>>>
>>>> throw new IllegalStateException("Maximum allowed data size of " +
>>>> this.maxDataSize + " exceeded.");
>>>>
>>>> Which matches what I see in the nifi-app.log file…
>>>>
>>>> 2016-08-20 22:03:05,470 ERROR [Timer-Driven Process Thread-8]
>>>> o.apache.nifi.processors.kafka.PutKafka
>>>> java.lang.IllegalStateException: Maximum allowed data size of 1048576
>>>> exceeded.
>>>> at org.apache.nifi.stream.io.util.StreamDemarcator.fill(
>>> StreamDemarcator.java:153)
>>>> ~[nifi-utils-0.7.0.jar:0.7.0]
>>>> at org.apache.nifi.stream.io.util.StreamDemarcator.
>>>> nextToken(StreamDemarcator.java:105) ~[nifi-utils-0.7.0.jar:0.7.0]
>>>> at org.apache.nifi.processors.kafka.KafkaPublisher.publish(
>>> KafkaPublisher.java:129)
>>>> ~[nifi-kafka-processors-0.7.0.jar:0.7.0]
>>>> at org.apache.nifi.processors.kafka.PutKafka$1.process(
>>> PutKafka.java:315)
>>>> ~[nifi-kafka-processors-0.7.0.jar:0.7.0]
>>>> at org.apache.nifi.controller.repository.
>>>> StandardProcessSession.read(StandardProcessSession.java:1851)
>>>> ~[nifi-framework-core-0.7.0.jar:0.7.0]
>>>> at org.apache.nifi.controller.repository.
>>>> StandardProcessSession.read(StandardProcessSession.java:1822)
>>>> ~[nifi-framework-core-0.7.0.jar:0.7.0]
>>>> at org.apache.nifi.processors.kafka.PutKafka.
>>>> doRendezvousWithKafka(PutKafka.java:311) ~[nifi-kafka-processors-0.7.0.
>>>> jar:0.7.0]
>>>> at org.apache.nifi.processors.kafka.PutKafka.
>>>> rendezvousWithKafka(PutKafka.java:287) ~[nifi-kafka-processors-0.7.0.
>>>> jar:0.7.0]
>>>> at org.apache.nifi.processors.kafka.AbstractKafkaProcessor.
>>>> onTrigger(AbstractKafkaProcessor.java:76)
~[nifi-kafka-processors-0.7.0.
>>>> jar:0.7.0]
>>>> at org.apache.nifi.controller.StandardProcessorNode.onTrigger(
>>>> StandardProcessorNode.java:1054) [nifi-framework-core-0.7.0.jar:0.7.0]
>>>> at org.apache.nifi.controller.tasks.ContinuallyRunProcessorTask.
>>>> call(ContinuallyRunProcessorTask.java:136) [nifi-framework-core-0.7.0.
>>>> jar:0.7.0]
>>>> at org.apache.nifi.controller.tasks.ContinuallyRunProcessorTask.
>>>> call(ContinuallyRunProcessorTask.java:47) [nifi-framework-core-0.7.0.
>>>> jar:0.7.0]
>>>> at org.apache.nifi.controller.scheduling.
>>>> TimerDrivenSchedulingAgent$1.run(TimerDrivenSchedulingAgent.java:127)
>>>> [nifi-framework-core-0.7.0.jar:0.7.0]
>>>> at java.util.concurrent.Executors$RunnableAdapter.
>>> call(Executors.java:511)
>>>> [na:1.8.0_45]
>>>> at java.util.concurrent.FutureTask.runAndReset(
>>> FutureTask.java:308)
>>>> [na:1.8.0_45]
>>>> at java.util.concurrent.ScheduledThreadPoolExecutor$
>>>> ScheduledFutureTask.access$301(ScheduledThreadPoolExecutor.java:180)
>>>> [na:1.8.0_45]
>>>> at java.util.concurrent.ScheduledThreadPoolExecutor$
>>>> ScheduledFutureTask.run(ScheduledThreadPoolExecutor.java:294)
>>>> [na:1.8.0_45]
>>>> at java.util.concurrent.ThreadPoolExecutor.runWorker(
>>> ThreadPoolExecutor.java:1142)
>>>> [na:1.8.0_45]
>>>> at java.util.concurrent.ThreadPoolExecutor$Worker.run(
>>> ThreadPoolExecutor.java:617)
>>>> [na:1.8.0_45]
>>>> at java.lang.Thread.run(Thread.java:745) [na:1.8.0_45]
>>>>
>>>> This occurs using PublishKafka, and PutKafka. Setting the Max Record
>>> Size
>>>> property in the PutKafka processor has no affect on this. Note the
stack
>>>> trace above is from the PutKafka processor with Max Record Size set to
>>> 10MB.
>>>>
>>>> I believe that this a regression from 0.6.0.
>>>>
>>>> Chris McDermott
>>>>
>>>> Remote Business Analytics
>>>> STaTS/StoreFront Remote
>>>> HPE Storage
>>>> Hewlett Packard Enterprise
>>>> Mobile: +1 978-697-5315
>>>>
>>>>
>>>>
>>>> On 8/20/16, 3:48 PM, "Andrew Psaltis" <psaltis.and...@gmail.com> wrote:
>>>>
>>>> Hi Chris,
>>>> Regarding the PutKafka processor looking at this block[1] of the
>>>> PutKafka
>>>> code, it has a default size of 1 MB, but it does not restrict the
>>>> size. The
>>>> DATA_SIZE_VALIDATOR does a sanity check and also enforces that
>>>> the supported value entered is the correct format <value> [B|
>>>> KB|MB|GB|TB].
>>>> Later on in the code at this block[2], the value is set on the Kafka
>>>> config, again this does not enforce a value maximum.
>>>>
>>>> In regards to the PublishKafka processor I do not see where it
>>> accepts
>>>> a
>>>> size nor restrict the size at all.
>>>>
>>>> Have you adjusted the 'message.max.bytes' config value for your
>>>> broker(s)?
>>>> The default value for that is 1 MB [3] (The url references the 0.8
>>>> Kafka,
>>>> however I believe this default has been stable since the early days
>>> of
>>>> the
>>>> project.)
>>>>
>>>> If you really do want to send messages that are larger than 1 MB in
>>>> size, I
>>>> would highly recommending reading this post[4] from Gwen Shapira. It
>>>> does
>>>> a great job of outlining the things you need to take into
>>>> consideration.
>>>> This will also point you to the relevant configs in Kafka that will
>>>> need to
>>>> be adjusted if you decide to go this route.
>>>>
>>>>
>>>> Thanks,
>>>> Andrew
>>>>
>>>> [1]
>>>> https://github.com/apache/nifi/blob/rel/nifi-0.7.0/nifi-
>>>> nar-bundles/nifi-kafka-bundle/nifi-kafka-processors/src/
>>>> main/java/org/apache/nifi/processors/kafka/PutKafka.java#L174-L180
>>>> [2]
>>>> https://github.com/apache/nifi/blob/rel/nifi-0.7.0/nifi-
>>>> nar-bundles/nifi-kafka-bundle/nifi-kafka-processors/src/
>>>> main/java/org/apache/nifi/processors/kafka/PutKafka.java#L495
>>>> [3] https://kafka.apache.org/08/configuration.html
>>>> [4] http://ingest.tips/2015/01/21/handling-large-messages-kafka/
>>>>
>>>> On Sat, Aug 20, 2016 at 3:25 PM, McDermott, Chris Kevin (MSDU -
>>>> STaTS/StorefrontRemote) <chris.mcderm...@hpe.com> wrote:
>>>>
>>>>> Hi folks,
>>>>>
>>>>>
>>>>>
>>>>> From experimentation and looking at the code it seems that the max
>>>> message
>>>>> size that can be sent via the PublishKafka and PutKafka processors
>>>> in 0.7.0
>>>>> is 1MB. Can someone please confirm my read on this?
>>>>>
>>>>>
>>>>>
>>>>> Thanks,
>>>>>
>>>>>
>>>>>
>>>>> Chris McDermott
>>>>>
>>>>>
>>>>>
>>>>> Remote Business Analytics
>>>>>
>>>>> STaTS/StoreFront Remote
>>>>>
>>>>> HPE Storage
>>>>>
>>>>> Hewlett Packard Enterprise
>>>>>
>>>>> Mobile: +1 978-697-5315
>>>>>
>>>>>
>>>>>
>>>>>
>>>>
>>>>
>>>> --
>>>> Thanks,
>>>> Andrew
>>>>
>>>> Subscribe to my book: Streaming Data <http://manning.com/psaltis>
>>>> <https://www.linkedin.com/pub/andrew-psaltis/1/17b/306>
>>>> twiiter: @itmdata <http://twitter.com/intent/
>>> user?screen_name=itmdata>
>>>>
>>>>
>>>>
>>>
>>>
>>> --
>>> Thanks,
>>> Andrew
>>>
>>> Subscribe to my book: Streaming Data <http://manning.com/psaltis>
>>> <https://www.linkedin.com/pub/andrew-psaltis/1/17b/306>
>>> twiiter: @itmdata <http://twitter.com/intent/user?screen_name=itmdata>
>>>
>>
>>
>>
>>
>
>
>
