flink-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Gyula Fora (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (FLINK-1503) GSoC project: Batch and Streaming integration through new operators and unified API
Date Wed, 18 Mar 2015 16:03:38 GMT

    [ https://issues.apache.org/jira/browse/FLINK-1503?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14367368#comment-14367368

Gyula Fora commented on FLINK-1503:


As you can probably tell from the description this a very complex issue with lot of possibilities
and implications.

You are always welcome to ask anything regarding this on the dev-list. (Or you can also reach
out directly to me)

As a first step I would probably try to understand whats going on under the hood in the Flink
runtime , and try to come up with some use cases that would need batch streaming interaction.

Having a use-case in mind can be a very good guide. 

I also did some prototype implementations a while back here: https://github.com/mbalassi/flink/tree/batch-integration

The branch is completely outdated but if you checkout you can run the "lambda example"

> GSoC project: Batch and Streaming integration through new operators and unified API
> -----------------------------------------------------------------------------------
>                 Key: FLINK-1503
>                 URL: https://issues.apache.org/jira/browse/FLINK-1503
>             Project: Flink
>          Issue Type: New Feature
>          Components: Java API, Scala API, Streaming
>            Reporter: Gyula Fora
>            Priority: Minor
>              Labels: gsoc2015, java, scala
> Currently the Flink batch and streaming API-s (java and scala) and runtimes work independently
from each other without operators to allow interactions between DataStreams and DataSets in
a fault-tolerant manner.
> The goal is to modify the execution environments and the runtime layer to allow these
> Possible runtime changes to add:
> -Interaction through intermediate files
> -Interaction by connection the execution graphs 
> Possible new operators implement:
> -Converting a dataset to a datastream (either by directly streaming in the results or
periodically executing the dataset transformations)
> -Hash joining a datastream with a dataset by key
> -Other binary operators with streams and sets
> The implementations should work with the fault tolerance mechanism provided by then (exaclty-once
or at-least-once guarantees).

This message was sent by Atlassian JIRA

View raw message