[jira] [Resolved] (STORM-171) Add "progress" method to OutputCollector

["Stig Rohde Døssing (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/STORM-171?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Stig Rohde Døssing resolved STORM-171.
--------------------------------------
       Resolution: Fixed
    Fix Version/s: 2.0.0
                   1.0.0

This sounds similar to https://issues.apache.org/jira/browse/STORM-1549. I'm marking this one as resolved. Feel free to reopen if https://issues.apache.org/jira/browse/STORM-1549 doesn't cover this issue.

> Add "progress" method to OutputCollector
> ----------------------------------------
>
>                 Key: STORM-171
>                 URL: https://issues.apache.org/jira/browse/STORM-171
>             Project: Apache Storm
>          Issue Type: New Feature
>          Components: storm-core
>            Reporter: James Xu
>            Priority: Minor
>             Fix For: 1.0.0, 2.0.0
>
>
> https://github.com/nathanmarz/storm/issues/168
> void progress(Tuple input)
> This would send a message back to the spout (through the acker) to increase the timeout for the roots of the tuple. Would be useful if the processing times of a tuple is highly variable. The timeout should reset to TOPOLOGY_MESSAGE_TIMEOUT plus the current time.
> -----------
> xumingming: what's the content of the tuple passed to the method progress?
> -----------
> nathanmarz: It would be the tuple that was passed to "execute". For example, in pseudocode:
> execute(Tuple input):
> while(...):
> // do some processing
> _collector.progress(input)
> _collector.ack(input)
> So the progress method would extend the timeout for the spout tuples at the root of "input"
> Just a note – a "progress" method could be a key to implementing MapReduce on top of Storm (along with more powerful grouping/scheduling capabilities and the ability to make use of local disk).
> -----------
> rohitprasad15: I wanted to take up this issue. I learnt Clojure and Storm a week back, but want to get deep into both.
> I can briefly describe my approach - 
> 1. Modify task.clj. Implement ^void progress [this ^Tuple tuple] as part of output-collector. Send a message to the acker, in a way similar to how ack method does it.
> 2. Modify acker.clj. Change execute() of IBolt implementation. Need to add ACKER-PROGRESS-STREAM-ID, to distinguish that its a progress message, and then finally reset the timeout to maxTopologyMessageTimeout.
> Is this approach in the right direction?



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