[ 
https://issues.apache.org/jira/browse/BEAM-3646?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Thomas Groh resolved BEAM-3646.
-------------------------------
       Resolution: Fixed
    Fix Version/s: 2.4.0

> Add comments about appropriate use of DoFn.Teardown
> ---------------------------------------------------
>
>                 Key: BEAM-3646
>                 URL: https://issues.apache.org/jira/browse/BEAM-3646
>             Project: Beam
>          Issue Type: Bug
>          Components: sdk-java-core
>            Reporter: Thomas Groh
>            Assignee: Thomas Groh
>            Priority: Critical
>             Fix For: 2.4.0
>
>          Time Spent: 40m
>  Remaining Estimate: 0h
>
> Because the {{Teardown}} method has no relation to the atomicity of 
> processing and commiting of output, it is EXTREMELY DANGEROUS to use to flush 
> outputs, and buffered data there is extremely likely to never be flushed. If 
> a DoFn instance with buffered data is lost (for example, via worker/machine 
> failure), and the runner has committed the result of processing that input, 
> the data is lost.
>  
> Not commenting on this being the case can cause users to believe that 
> (especially if running a batch pipeline) that their data will be flushed on 
> pipeline completion. This is very dangerous behavior that we do not warn of 
> sufficiently.



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

Reply via email to