Mailing-List: contact dev-help@flink.apache.org; run by ezmlm
Precedence: bulk
Reply-To: dev@flink.apache.org
Date: Tue, 5 Jan 2016 11:18:40 +0000 (UTC)
From: "Aljoscha Krettek (JIRA)" <jira@apache.org>
To: dev@flink.apache.org
Message-ID: <JIRA.12927250.1451992712000.20327.1451992720227@Atlassian.JIRA>
In-Reply-To: <JIRA.12927250.1451992712000@Atlassian.JIRA>
References: <JIRA.12927250.1451992712000@Atlassian.JIRA>
 <JIRA.12927250.1451992712203@arcas>
Subject: [jira] [Created] (FLINK-3200) Use Partitioned State Abstraction in
 WindowOperator
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit

Aljoscha Krettek created FLINK-3200:
---------------------------------------

             Summary: Use Partitioned State Abstraction in WindowOperator
                 Key: FLINK-3200
                 URL: https://issues.apache.org/jira/browse/FLINK-3200
             Project: Flink
          Issue Type: Improvement
          Components: Streaming
    Affects Versions: 1.0.0
            Reporter: Aljoscha Krettek
            Assignee: Aljoscha Krettek


Right now, the WindowOperator uses snapshotState and restoreState to do custom serialization/deserialization of the windowing state. Moving this to partitioned state would allow us to re-scale in the future. Also, once the partitioned state has support to run on ManagedMemory/Out-of-core state the WindowOperator would also automatically benefit from this, allowing very large windows and large window state.

The necessary steps are these:
- Enhance state interface with new primitive states: ReducableState, ListState, etc.. Also make state scoped to a namespace. Namespaces are necessary because we need state that is scoped by key and by window (window would be the namespace in this case)
- Enhance timer interface to also make timers scoped to a key/namespace
- Change WindowOperator to use these new interfaces


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