Mailing-List: contact dev-help@kafka.apache.org; run by ezmlm
Precedence: bulk
Reply-To: dev@kafka.apache.org
Date: Tue, 7 Mar 2017 11:36:38 +0000 (UTC)
From: "Dan (JIRA)" <jira@apache.org>
To: dev@kafka.apache.org
Message-ID: <JIRA.13048447.1488783831000.769.1488886598562@Atlassian.JIRA>
In-Reply-To: <JIRA.13048447.1488783831000@Atlassian.JIRA>
References: <JIRA.13048447.1488783831000@Atlassian.JIRA> <JIRA.13048447.1488783831454@jira-lw-us.apache.org>
Subject: [jira] [Commented] (KAFKA-4845) KafkaConsumer.seekToEnd cannot take
 effect when integrating with spark streaming
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit
archived-at: Tue, 07 Mar 2017 11:36:45 -0000


    [ https://issues.apache.org/jira/browse/KAFKA-4845?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15899299#comment-15899299 ] 

Dan commented on KAFKA-4845:
----------------------------

I checked 0.10.2.0 again and it was indeed fixed. Thanks a lot!

> KafkaConsumer.seekToEnd cannot take effect when integrating with spark streaming
> --------------------------------------------------------------------------------
>
>                 Key: KAFKA-4845
>                 URL: https://issues.apache.org/jira/browse/KAFKA-4845
>             Project: Kafka
>          Issue Type: Bug
>          Components: clients
>    Affects Versions: 0.10.1.0, 0.10.1.1, 0.10.2.0
>            Reporter: Dan
>            Assignee: Vahid Hashemian
>
> When integrating with spark streaming, kafka consumer cannot get the latest offsets except for one partition. The  code snippet is as follows: 
> {code}
> protected def latestOffsets(): Map[TopicPartition, Long] = {
>     val c = consumer
>     c.poll(0)
>     val parts = c.assignment().asScala
>     val newPartitions = parts.diff(currentOffsets.keySet)
>     currentOffsets = currentOffsets ++ newPartitions.map(tp => tp -> c.position(tp)).toMap
>     c.pause(newPartitions.asJava)
>     c.seekToEnd(currentOffsets.keySet.asJava)
>     parts.map(tp => tp -> c.position(tp)).toMap
>   }
> {code}
> When calling consumer.position(topicPartition), it will call updateFetchPositions(Collections.singleton(partition)):
> The bug lies in updateFetchPositions(Set<TopicPartition> partitions):
> {code}
>         fetcher.resetOffsetsIfNeeded(partitions);    // reset to latest offset for current partition
>         if (!subscriptions.hasAllFetchPositions()) {  // called seekToEnd for all partitions before, so this sentence will be true 
>             coordinator.refreshCommittedOffsetsIfNeeded();
>             fetcher.updateFetchPositions(partitions);  // reset to committed offsets for current partition
>         }
> {code}
> So eventually there is only one partition(the last partition in assignment) can get latest offset while all the others get the committed offset.


--
This message was sent by Atlassian JIRA
(v6.3.15#6346)