[jira] [Comment Edited] (KAFKA-6582) Partitions get underreplicated, with a single ISR, and doesn't recover. Other brokers do not take over and we need to manually restart the broker.

["Sebastian Schmitz (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/KAFKA-6582?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16776419#comment-16776419 ] 

Sebastian Schmitz edited comment on KAFKA-6582 at 2/24/19 9:59 PM:
-------------------------------------------------------------------

We are using 2.1.0

Will update to 2.1.1 today to see if it's related to the mentioned deadlock


was (Author: paxi):
We are using 2.1.0

> Partitions get underreplicated, with a single ISR, and doesn't recover. Other brokers do not take over and we need to manually restart the broker.
> --------------------------------------------------------------------------------------------------------------------------------------------------
>
>                 Key: KAFKA-6582
>                 URL: https://issues.apache.org/jira/browse/KAFKA-6582
>             Project: Kafka
>          Issue Type: Bug
>          Components: network
>    Affects Versions: 1.0.0
>         Environment: Ubuntu 16.04
> Linux kafka04 4.4.0-109-generic #132-Ubuntu SMP Tue Jan 9 19:52:39 UTC 2018 x86_64 x86_64 x86_64 GNU/Linux
> java version "9.0.1"
> Java(TM) SE Runtime Environment (build 9.0.1+11)
> Java HotSpot(TM) 64-Bit Server VM (build 9.0.1+11, mixed mode) 
> but also tried with the latest JVM 8 before with the same result.
>            Reporter: Jurriaan Pruis
>            Priority: Major
>         Attachments: Screenshot 2019-01-18 at 13.08.17.png, Screenshot 2019-01-18 at 13.16.59.png
>
>
> Partitions get underreplicated, with a single ISR, and doesn't recover. Other brokers do not take over and we need to manually restart the 'single ISR' broker (if you describe the partitions of replicated topic it is clear that some partitions are only in sync on this broker).
> This bug resembles KAFKA-4477 a lot, but since that issue is marked as resolved this is probably something else but similar.
> We have the same issue (or at least it looks pretty similar) on Kafka 1.0. 
> Since upgrading to Kafka 1.0 in November 2017 we've had these issues (we've upgraded from Kafka 0.10.2.1).
> This happens almost every 24-48 hours on a random broker. This is why we currently have a cronjob which restarts every broker every 24 hours. 
> During this issue the ISR shows the following server log: 
> {code:java}
> [2018-02-20 12:02:08,342] WARN Attempting to send response via channel for which there is no open connection, connection id 10.132.0.32:9092-10.14.148.20:56352-96708 (kafka.network.Processor)
> [2018-02-20 12:02:08,364] WARN Attempting to send response via channel for which there is no open connection, connection id 10.132.0.32:9092-10.14.150.25:54412-96715 (kafka.network.Processor)
> [2018-02-20 12:02:08,349] WARN Attempting to send response via channel for which there is no open connection, connection id 10.132.0.32:9092-10.14.149.18:35182-96705 (kafka.network.Processor)
> [2018-02-20 12:02:08,379] WARN Attempting to send response via channel for which there is no open connection, connection id 10.132.0.32:9092-10.14.150.25:54456-96717 (kafka.network.Processor)
> [2018-02-20 12:02:08,448] WARN Attempting to send response via channel for which there is no open connection, connection id 10.132.0.32:9092-10.14.159.20:36388-96720 (kafka.network.Processor)
> [2018-02-20 12:02:08,683] WARN Attempting to send response via channel for which there is no open connection, connection id 10.132.0.32:9092-10.14.157.110:41922-96740 (kafka.network.Processor)
> {code}
> Also on the ISR broker, the controller log shows this:
> {code:java}
> [2018-02-20 12:02:14,927] INFO [Controller-3-to-broker-3-send-thread]: Controller 3 connected to 10.132.0.32:9092 (id: 3 rack: null) for sending state change requests (kafka.controller.RequestSendThread)
> [2018-02-20 12:02:14,927] INFO [Controller-3-to-broker-0-send-thread]: Controller 3 connected to 10.132.0.10:9092 (id: 0 rack: null) for sending state change requests (kafka.controller.RequestSendThread)
> [2018-02-20 12:02:14,928] INFO [Controller-3-to-broker-1-send-thread]: Controller 3 connected to 10.132.0.12:9092 (id: 1 rack: null) for sending state change requests (kafka.controller.RequestSendThread){code}
> And the non-ISR brokers show these kind of errors:
>  
> {code:java}
> 2018-02-20 12:02:29,204] WARN [ReplicaFetcher replicaId=1, leaderId=3, fetcherId=0] Error in fetch to broker 3, request (type=FetchRequest, replicaId=1, maxWait=500, minBytes=1, maxBytes=10485760, fetchData={......................}, isolationLevel=READ_UNCOMMITTED) (kafka.server.ReplicaFetcherThread)
> java.io.IOException: Connection to 3 was disconnected before the response was read
>  at org.apache.kafka.clients.NetworkClientUtils.sendAndReceive(NetworkClientUtils.java:95)
>  at kafka.server.ReplicaFetcherBlockingSend.sendRequest(ReplicaFetcherBlockingSend.scala:96)
>  at kafka.server.ReplicaFetcherThread.fetch(ReplicaFetcherThread.scala:205)
>  at kafka.server.ReplicaFetcherThread.fetch(ReplicaFetcherThread.scala:41)
>  at kafka.server.AbstractFetcherThread.processFetchRequest(AbstractFetcherThread.scala:149)
>  at kafka.server.AbstractFetcherThread.doWork(AbstractFetcherThread.scala:113)
>  at kafka.utils.ShutdownableThread.run(ShutdownableThread.scala:64)
> {code}
>  



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)