kafka-jira mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Guozhang Wang (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (KAFKA-5825) Streams not processing when exactly once is set
Date Wed, 13 Sep 2017 00:40:00 GMT

    [ https://issues.apache.org/jira/browse/KAFKA-5825?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16163954#comment-16163954

Guozhang Wang commented on KAFKA-5825:

I looked carefully through the logs and here is what I can find:

1) the producer did successfully send the message with the transaction, and the coordinator's
last state transition is to transit from prepareCommit to completeCommit:

17:44:52.908 [kafka-request-handler-2] DEBUG kafka.coordinator.transaction.TransactionMetadata
- TransactionalId aa39562f-8b63-42be-8aea-61b10d7d56eb prepare transition from PrepareCommit
to TxnTransitMetadata(producerId=2000, producerEpoch=0, txnTimeoutMs=60000, txnState=CompleteCommit,
topicPartitions=Set(), txnStartTimestamp=1505061892804, txnLastUpdateTimestamp=1505061892908)

Note that it was long after the starting of the test, because producer was keeping trying
to find the coordinator in a loop, that is mainly because the default num.partitions for transaction
logs is 50, which takes a lot of time to set up with replication factor of 3.

2) during the mean time, Streams consumer keep fetching data from but cannot get any:

17:44:53.910 [exactly-once-762724d8-71d8-4c1d-a924-970fc5a81e9f-StreamThread-1] DEBUG org.apache.kafka.clients.consumer.internals.Fetcher
- Fetch READ_UNCOMMITTED at offset 1 for partition my-topic-0 returned fetch data (error=NONE,
highWaterMark=1, lastStableOffset = -1, logStartOffset = 0, abortedTransactions = null, recordsSizeInBytes=0)
17:44:53.910 [exactly-once-762724d8-71d8-4c1d-a924-970fc5a81e9f-StreamThread-1] DEBUG org.apache.kafka.clients.consumer.internals.Fetcher
- Added READ_UNCOMMITTED fetch request for partition my-topic-0 at offset 1 to node
(id: 0 rack: null)
17:44:53.910 [exactly-once-762724d8-71d8-4c1d-a924-970fc5a81e9f-StreamThread-1] DEBUG org.apache.kafka.clients.consumer.internals.Fetcher
- Sending READ_UNCOMMITTED fetch for partitions [my-topic-0] to broker (id:
0 rack: null)

So I cannot see any obvious correctness issues, and I'd suspect it is because the default
num.partitions is too large to set up for your test (note that without EOS turned on this
topic would not be created).

I'd suggest try changing {transaction.state.log.num.partitions} config to smaller (3?) number,
and also consider reducing replication factor for your testing, and see if it works.

> Streams not processing when exactly once is set
> -----------------------------------------------
>                 Key: KAFKA-5825
>                 URL: https://issues.apache.org/jira/browse/KAFKA-5825
>             Project: Kafka
>          Issue Type: Bug
>          Components: streams
>    Affects Versions:
>         Environment: EmbeddedKafka running on Windows.  Relevant files attached.
>            Reporter: Ryan Worsley
>         Attachments: build.sbt, log4j.properties, log-output.txt, Tests.scala
> +Set-up+
> I'm using [EmbeddedKafka|https://github.com/manub/scalatest-embedded-kafka/] for ScalaTest.
> This spins up a single broker internally on a random port.
> I've written two tests - the first without transactions, the second with.  They're nearly
identical apart from the config and the transactional semantics.  I've written the transactional
version based on Neha's [blog|https://www.confluent.io/blog/exactly-once-semantics-are-possible-heres-how-apache-kafka-does-it/]
which is the closest thing I could find to instructions.
> The tests wait until a single message is processed by the streams topology, they use
this message to complete a promise that the test is waiting on.  Once the promise completes
the test verifies the value of the promise as being the expected value of the message.
> +Observed behaviour+
> The first test passes fine, the second test times out, the stream processor never seems
to read the transactional message.
> +Notes+
> I've attached my build.sbt, log4j.properties and my Tests.scala file in order to make
it as easy as possible for someone to re-create.  I'm running on Windows and using Scala as
this reflects my workplace.  I completely expect there to be some configuration issue that's
causing this, but am unable to proceed at this time.
> Related information: https://github.com/manub/scalatest-embedded-kafka/issues/82

This message was sent by Atlassian JIRA

View raw message