You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@kafka.apache.org by "ASF GitHub Bot (JIRA)" <ji...@apache.org> on 2015/11/12 02:15:11 UTC
[jira] [Commented] (KAFKA-2813) selector doesn't close socket
connection on non-IOExceptions
[ https://issues.apache.org/jira/browse/KAFKA-2813?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15001441#comment-15001441 ]
ASF GitHub Bot commented on KAFKA-2813:
---------------------------------------
GitHub user junrao opened a pull request:
https://github.com/apache/kafka/pull/501
KAFKA-2813: selector doesn't close socket connection on non-IOExceptions
Patched Selector.poll() to close the connection on any exception.
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/junrao/kafka KAFKA-2813
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/kafka/pull/501.patch
To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:
This closes #501
----
commit 2a4dfd4d63f3b3383d9ce01fce7c2be151ef9f78
Author: Jun Rao <ju...@gmail.com>
Date: 2015-11-12T01:11:49Z
KAFKA-2813: selector doesn't close socket connection on non-IOExceptions
----
> selector doesn't close socket connection on non-IOExceptions
> ------------------------------------------------------------
>
> Key: KAFKA-2813
> URL: https://issues.apache.org/jira/browse/KAFKA-2813
> Project: Kafka
> Issue Type: Bug
> Components: core
> Reporter: Jun Rao
> Assignee: Jun Rao
> Priority: Blocker
> Fix For: 0.9.0.0
>
>
> When running a system test, we saw lots of entries like the following. The issue is that when the current leader switches to the follower, we will truncate the log in the follower. It's possible there is a concurrent fetch request being served at this moment. If this happens, we throw a KafkaException when trying to send the fetch response (in FileMessageSet). The exception will propagate through Selector.poll(). Selector catches IOException and closes the corresponding socket. However, KafkaException is not an IOException. Since the socket is not closed, Selector.poll() will keep accessing the socket and keep getting the same error.
> [2015-11-11 07:25:01,150] ERROR Processor got uncaught exception. (kafka.network.Processor)
> kafka.common.KafkaException: Size of FileMessageSet /mnt/kafka-data-logs/test_topic-0/00000000000000000000.log has been truncated during write: old size 16368, new size 0
> at kafka.log.FileMessageSet.writeTo(FileMessageSet.scala:158)
> at kafka.api.PartitionDataSend.writeTo(FetchResponse.scala:77)
> at org.apache.kafka.common.network.MultiSend.writeTo(MultiSend.java:81)
> at kafka.api.TopicDataSend.writeTo(FetchResponse.scala:148)
> at org.apache.kafka.common.network.MultiSend.writeTo(MultiSend.java:81)
> at kafka.api.FetchResponseSend.writeTo(FetchResponse.scala:291)
> at org.apache.kafka.common.network.KafkaChannel.send(KafkaChannel.java:165)
> at org.apache.kafka.common.network.KafkaChannel.write(KafkaChannel.java:152)
> at org.apache.kafka.common.network.Selector.poll(Selector.java:301)
> at kafka.network.Processor.run(SocketServer.scala:413)
> at java.lang.Thread.run(Thread.java:745)
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)