[ 
https://issues.apache.org/jira/browse/STORM-927?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14640158#comment-14640158
 ] 

ASF GitHub Bot commented on STORM-927:
--------------------------------------

Github user DevInsanity commented on the pull request:

    https://github.com/apache/storm/pull/623#issuecomment-124442778
  
    @caofangkun 
    No, we're using a flux based topology and deploying to a remote cluster 
(via `storm.cmd jar project.jar org.apache.storm.flux.Flux --remote 
project.yaml --filter project.properties`).
    
    The problem is that because storm.cmd launches the actual process via 
`start /b` then it doesn't wait for the topology to deploy before returning to 
the caller, so you can't get any error codes back from the deployer (for 
example, if the yaml is invalid).
    
    We're actually invoking this from a maven antrun plugin task as part of 
build process. In that case, it's even worse, as in the case of an error, not 
only does maven still report a build successful (as the antrun plugin's exec 
task doesn't see an error code), but none of the output is captured either. 
With the change, the output from the deployment is captured by maven properly.
    
    While we could specify the storm_debug option explicitly (which would still 
need the quotes added around the %JAVA% to work properly when java is installed 
under Program Files), the jar command doesn't appear to be intended to launch a 
background server, so should wait for it's action to finish before returning.


> Storm.cmd on Windows shouldn't exit until deployment complete
> -------------------------------------------------------------
>
>                 Key: STORM-927
>                 URL: https://issues.apache.org/jira/browse/STORM-927
>             Project: Apache Storm
>          Issue Type: Improvement
>    Affects Versions: 0.10.0
>         Environment: Windows Server 2008 R2
>            Reporter: Paul Milliken
>            Priority: Minor
>              Labels: windows
>
> The storm.cmd script returns before the topology has been deployed. This 
> causes problems when attempting to automate deployments, as detecting 
> failures is made difficult.
> It looks like this is because all storm command are ultimately invoked via 
> "start /b" unless STORM_DEBUG is set. Attempting to workaround by setting 
> STORM_DEBUG doesn't work, as the fix in STORM-322 hasn't been applied to the 
> debug command.
> It looks like a relatively simple fix. Apply the STORM-322 fix to the debug 
> route, and make the jar command use it by default.



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

Reply via email to