[jira] [Comment Edited] (BEAM-1777) If PipelineEnforcement throws an exception after Pipeline.run() fails, it overwrites the original failure

["Stas Levin (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/BEAM-1777?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15936232#comment-15936232 ] 

Stas Levin edited comment on BEAM-1777 at 3/22/17 12:47 PM:
------------------------------------------------------------

I wonder if this can be resolved by removing the {{finally}} around {{enforcement.get().afterPipelineExecution()}}.
At first glance, if {{Pipeline#run()}} fails there's little point in moving on to checking enforcements and the likes.

I'll look into it.


was (Author: staslev):
I wonder if this can be resolved by removing the {{finally}} around {{enforcement.get().afterPipelineExecution()}}.
At first glance, if {{Pipeline#run()}} fails there's little point in moving on to checking enforcements and the likes.

> If PipelineEnforcement throws an exception after Pipeline.run() fails, it overwrites the original failure
> ---------------------------------------------------------------------------------------------------------
>
>                 Key: BEAM-1777
>                 URL: https://issues.apache.org/jira/browse/BEAM-1777
>             Project: Beam
>          Issue Type: Bug
>          Components: sdk-java-core
>            Reporter: Thomas Groh
>            Assignee: Stas Levin
>
> The exception is thrown out of the finally block after the original exception was already thrown, and is not properly suppressed.
> If the Pipeline throws an exception in construction (including validation failures), this can cause {{AbandonedNodeEnforcement}} to attempt to traverse a {{Pipeline}} which is invalid, which throws another exception. That exception is improperly propagated instead of the original failure, which complicates debugging.
> An example is using an unkeyed input PCollection to a Stateful ParDo. The validation will fail, but the error message will suggest that a node has incompletely specified outputs (which is a consequence rather than a cause of having an invalid transform).



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