You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@kafka.apache.org by "Guozhang Wang (JIRA)" <ji...@apache.org> on 2016/01/14 02:27:39 UTC

[jira] [Created] (KAFKA-3104) Windowed Stream Aggregation Implementation

Guozhang Wang created KAFKA-3104:
------------------------------------

             Summary: Windowed Stream Aggregation Implementation
                 Key: KAFKA-3104
                 URL: https://issues.apache.org/jira/browse/KAFKA-3104
             Project: Kafka
          Issue Type: Sub-task
            Reporter: Guozhang Wang
            Assignee: Guozhang Wang


Proposed approach:
1. Add a K-V store with format <key, window> : <agg-value> with Windowed<K> ser-de and <T> ser-de.
2. Upon receiving a record forwarded from the upstream processor:
2.1. First get the mapping window(s) from the context timestamp.
2.2. For each window, first try fetch from the store, if not exist call initialValue().
2.2. Only call add(..), and never call remove(..)
2.3. Forward the aggregate value with key Windowed<agg-key> and value Change<agg-value>.



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