[
https://issues.apache.org/jira/browse/FLINK-7018?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Xu Pingyong updated FLINK-7018:
-------------------------------
Description:
Motivation:
1. StreamGraph is a graph consisted of some streamNodes. So virtual
nodes (such as select, sideOutput, partition) should be moved away from it.
Main iterfaces of StreamGraph should be as following：
addSource(StreamNode sourceNode)
addSink(StreamNode sinkNode)
addOperator(StreamNode streamNode)
addEdge(Integer upStreamVertexID, Integer downStreamVertexID,
StreamEdge.InputOrder inputOrder, StreamPartitioner<?> partitioner,
List<String> outputNames, OutputTag outputTag)
getJobGraph()
2. StreamExecutionEnvironment should not be in StreamGraph, I create
StreamGraphProperties which extracts all env information the streamGraph needs
from StreamExecutionEnvironment. It contains:
1) executionConfig
2) checkpointConfig
3) timeCharacteristic
4) stateBackend
5) chainingEnabled
6) cachedFiles
7) jobName
Related Changes:
I moved the part of dealing with virtual nodes to StreamGraphGenerator.
And get properties of StreamGraph from StreamGraphProperties instead of
StreamExecutionEnvironment.
It is only a code abstraction internally.
was:
Motivation:
1. StreamGraph is a graph consisted of some streamNodes. So virtual
nodes (such as select, sideOutput, partition) should be moved away from it.
Main iterfaces of StreamGraph should be as following：
addSource(StreamNode sourceNode)
addSink(StreamNode sinkNode)
addOperator(StreamNode streamNode)
addEdge(Integer upStreamVertexID, Integer downStreamVertexID,
StreamEdge.InputOrder inputOrder, StreamPartitioner<?> partitioner,
List<String> outputNames, OutputTag outputTag)
getJobGraph()
2. StreamExecutionEnvironment should not be in StreamGraph, I create
StreamGraphProperties which extracts all information the streamGraph needs from
StreamExecutionEnvironment. It contains:
1) executionConfig
2) checkpointConfig
3) timeCharacteristic
4) stateBackend
5) chainingEnabled
6) cachedFiles
7) jobName
Related Changes:
I moved the part of dealing with virtual nodes to StreamGraphGenerator.
And get properties of StreamGraph from StreamGraphProperties instead of
StreamExecutionEnvironment.
It is only a code abstraction internally.
> Refactor streamGraph to make interfaces clear
> ---------------------------------------------
>
> Key: FLINK-7018
> URL: https://issues.apache.org/jira/browse/FLINK-7018
> Project: Flink
> Issue Type: Improvement
> Components: DataStream API
> Reporter: Xu Pingyong
> Assignee: Xu Pingyong
>
> Motivation:
> 1. StreamGraph is a graph consisted of some streamNodes. So virtual
> nodes (such as select, sideOutput, partition) should be moved away from it.
> Main iterfaces of StreamGraph should be as following：
> addSource(StreamNode sourceNode)
> addSink(StreamNode sinkNode)
> addOperator(StreamNode streamNode)
> addEdge(Integer upStreamVertexID, Integer downStreamVertexID,
> StreamEdge.InputOrder inputOrder, StreamPartitioner<?> partitioner,
> List<String> outputNames, OutputTag outputTag)
> getJobGraph()
> 2. StreamExecutionEnvironment should not be in StreamGraph, I create
> StreamGraphProperties which extracts all env information the streamGraph
> needs from StreamExecutionEnvironment. It contains:
> 1) executionConfig
> 2) checkpointConfig
> 3) timeCharacteristic
> 4) stateBackend
> 5) chainingEnabled
> 6) cachedFiles
> 7) jobName
>
> Related Changes:
> I moved the part of dealing with virtual nodes to
> StreamGraphGenerator. And get properties of StreamGraph from
> StreamGraphProperties instead of StreamExecutionEnvironment.
>
> It is only a code abstraction internally.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
