[jira] [Updated] (FLINK-20435) Refactor ExecNode and ExecEdge

["Johannes Moser (Jira)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/FLINK-20435?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Johannes Moser updated FLINK-20435:
-----------------------------------
    Fix Version/s: 1.14.0

> Refactor ExecNode and ExecEdge
> ------------------------------
>
>                 Key: FLINK-20435
>                 URL: https://issues.apache.org/jira/browse/FLINK-20435
>             Project: Flink
>          Issue Type: Improvement
>          Components: Table SQL / Planner
>            Reporter: godfrey he
>            Priority: Major
>              Labels: auto-unassigned
>             Fix For: 1.14.0
>
>
> Currently, there are many improvements about ExecNode:
> 1. simplify type parameter of {{ExecNode}}. Currently, {{ExecNode#translateToPlan}} takes {{BatchPlanner}} or {{StreamPlanner}} as a parameter, so {{ExecNode}} has a type parameter {{E <: Planner}}, which indicates the node is a batch node or a streaming node. While in the future, a plan may contain both batch nodes and stream node. The type parameter can be removed, and we will use PlannerBase instead.
> 2. port the implementation of ExecNodes to Java
> 3. separate the implementation of {{FlinkPhysicalRel}} and {{ExecNode}}. Currently, an execution node extends both from {{FlinkPhysicalRel}} and {{ExecNode}}. After a physical node is converted to an exec node, many parameters are unnecessary, such as: RelOptCluster, RelTraitSet, etc. With more optimizations on {{ExecNode}}, We need {{ExecNode}} to be cleaner and simpler. So we will separate the implementation of {{FlinkPhysicalRel}} and {{ExecNode}}.
> Currently, the ExecEdge represents the properties for the input of an operator, the properties include 
> 1. required data distribute for an input (the input corresponds to the [Input|https://github.com/apache/flink/blob/master/flink-streaming-java/src/main/java/org/apache/flink/streaming/api/operators/Input.java])
> 2. DamBehavior which describes the behaviors how an in record may trigger the output of the target operator.
> 3. the priority of this input read by the target operator
> So ExecEdge should be rename to InputProperty, and we will re-introduce ExecEdge which describes how to connect two ExecNodes, and how to shuffle the data between two ExecNodes. Its role should be similar to StreamEdge.
> This is an umbrella issue, we will create more related sub-tasks.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)