[
https://issues.apache.org/jira/browse/STORM-17?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Rick Kellogg updated STORM-17:
------------------------------
Component/s: storm-core
> Spouts should be their own ackers
> ---------------------------------
>
> Key: STORM-17
> URL: https://issues.apache.org/jira/browse/STORM-17
> Project: Apache Storm
> Issue Type: Improvement
> Components: storm-core
> Reporter: James Xu
>
> This would create the following improvements:
> https://github.com/nathanmarz/storm/issues/282
> 1) Operational: no more need to specify the number of ackers
> 2) More accurate tuple tree tracking: When this is done, spouts can generate
> the tuple ids and guarantee that they're unique by checking if it already
> exists already. With the current acker approach, the birthday problem takes
> effect, and occasionally two different spout tuples will have the same id,
> causing them both to timeout and fail. See this thread for more details:
> http://groups.google.com/group/storm-user/browse_frm/thread/f3267a4a605757e9
> 3) Better performance: No need for the ack init message from spout -> acker,
> and no need for ack/fail message from acker -> spout
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
