You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@samza.apache.org by "Jagadish (JIRA)" <ji...@apache.org> on 2016/04/15 00:02:25 UTC

[jira] [Updated] (SAMZA-934) Race condition in Kafka System Producer

     [ https://issues.apache.org/jira/browse/SAMZA-934?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Jagadish updated SAMZA-934:
---------------------------
    Description: 
There is a race condition in the KafkaSystemProducer class in the access to the sendFailed flag. (in the main and the callback threads). It is possible for a 

A. a failure to be masked  (or)
B. a failure to repeatedly re-throw exceptions in a loop,

by the write to this flag in another thread. This may lead to holes in processing of messages.

Ideally, failure should immediately shut down the producer in the *same* callback thread. This will cancel future sends / buffered sends in the producer. 

  was:
There is a race condition in the KafkaSystemProducer class in the access to the sendFailed flag. (in the main and the callback threads). It is possible for a A. a failure to be masked  
B. a failure to repeatedly re-throw exceptions in a loop, by the write to this flag by another thread.

Ideally, failure should immediately shut down the producer in the *same* callback thread. This will cancel future sends / buffered sends in the producer. 



> Race condition in Kafka System Producer
> ---------------------------------------
>
>                 Key: SAMZA-934
>                 URL: https://issues.apache.org/jira/browse/SAMZA-934
>             Project: Samza
>          Issue Type: Bug
>            Reporter: Jagadish
>            Assignee: Navina Ramesh
>
> There is a race condition in the KafkaSystemProducer class in the access to the sendFailed flag. (in the main and the callback threads). It is possible for a 
> A. a failure to be masked  (or)
> B. a failure to repeatedly re-throw exceptions in a loop,
> by the write to this flag in another thread. This may lead to holes in processing of messages.
> Ideally, failure should immediately shut down the producer in the *same* callback thread. This will cancel future sends / buffered sends in the producer. 



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)