[ 
https://issues.apache.org/jira/browse/FLINK-2525?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14732295#comment-14732295
 ] 

ASF GitHub Bot commented on FLINK-2525:
---------------------------------------

Github user ffbin commented on the pull request:

    https://github.com/apache/flink/pull/1046#issuecomment-138051794
  
    @mjsax @StephanEwen Thanks. I have finish the changes at the core classes. 
    when user use env.getConfig().setGlobalJobParameters(conf); to set storm 
configuration,  I need convert Configuration into map<Object,Object>, toMap() 
is not enough, So i add getConfDataClone() to get a clone of inner config 
hashmap.
    I have a look at how to use TaskConfiguration instead of JobConfiguration.
    TaskConfiguration transfer path is:
    TaskConfiguration <-- TaskDeploymentDescriptor <-- JobVertex <-- 
StreamConfig 
    In StreamingJobGraphGenerator.java setVertexConfig() function set 
TaskConfiguration, like this:
                ...
                config.setNumberOfOutputs(nonChainableOutputs.size());
                config.setNonChainedOutputs(nonChainableOutputs);
                config.setChainedOutputs(chainableOutputs);
                
config.setCheckpointingEnabled(streamGraph.isCheckpointingEnabled());
                ...
    If we want to use TaskConfiguration instead of JobConfiguration, we need 
add special interface to set stormConfig in StreamGraph and StreamConfig.
    I think it maybe break the separation between the storm-compatibility and 
core code.
    what is your opinion?


> Add configuration support in Storm-compatibility
> ------------------------------------------------
>
>                 Key: FLINK-2525
>                 URL: https://issues.apache.org/jira/browse/FLINK-2525
>             Project: Flink
>          Issue Type: New Feature
>          Components: Storm Compatibility
>            Reporter: fangfengbin
>            Assignee: fangfengbin
>
> Spouts and Bolt are initialized by a call to `Spout.open(...)` and 
> `Bolt.prepare()`, respectively. Both methods have a config `Map` as first 
> parameter. This map is currently not populated. Thus, Spouts and Bolts cannot 
> be configure with user defined parameters. In order to support this feature, 
> spout and bolt wrapper classes need to be extended to create a proper `Map` 
> object. Furthermore, the clients need to be extended to take a `Map`, 
> translate it into a Flink `Configuration` that is forwarded to the wrappers 
> for proper initialization of the map.



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

Reply via email to