[
https://issues.apache.org/jira/browse/FLINK-20478?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17243679#comment-17243679
]
Jark Wu commented on FLINK-20478:
---------------------------------
I think the current "Physical Execution Plan" is not very useful to users. It
has almost the same information with "Optimized Logical Plan" or the new
"Optimized Physical Plan". Even more, it doesn't display well when node is
reused [1]. A json execution plan FLINK-19687 is a better format for it, but it
is very verbose.
Thus, my thought is:
1) Display "Optimized Exeuction Plan" by default (your example in description)
2) show additional "Json Exeuction Plan" only if
ExplainDetail.JSON_EXECUTION_PLAN is enabled.
[1]:
https://github.com/apache/flink/blob/master/flink-table/flink-table-planner-blink/src/test/resources/org/apache/flink/table/api/stream/ExplainTest.xml#L376
> Adjust the explain result
> -------------------------
>
> Key: FLINK-20478
> URL: https://issues.apache.org/jira/browse/FLINK-20478
> Project: Flink
> Issue Type: Sub-task
> Components: Table SQL / Planner
> Reporter: godfrey he
> Priority: Major
>
> Currently, the explain result includes "Abstract Syntax Tree", "Optimized
> Logical Plan" and "Physical Execution Plan". While the "Optimized Logical
> Plan" is an {{ExecNode}} graph, and the
> "[ExplainDetail|https://github.com/apache/flink/blob/master/flink-table/flink-table-api-java/src/main/java/org/apache/flink/table/api/ExplainDetail.java]";
> represents the expected explain details, including {{ESTIMATED_COST}} and
> {{CHANGELOG_MODE}} now. Those types can only used for Calicte {{RelNode}}s
> instead of {{ExecNode}}. So I suggest to make the following adjustments:
> 1. Keep "Abstract Syntax Tree" as it, which represents the original
> (un-optimized) {{RelNode}} graph converted from {{SqlNode}}.
> 2. Rename "Optimized Logical Plan" to "Optimized Physical Plan", which
> represents the optimized physical {{RelNode}} graph composed of
> {{FlinkPhysicalRel}}. {{ESTIMATED_COST}} and {{CHANGELOG_MODE}} describe the
> expected explain details for "Optimized Physical Plan".
> 3.Replace "Physical Execution Plan" with "Optimized Execution Plan", which
> represents the optimized {{ExecNode}} graph. Currently, many optimizations
> are based on {{ExecNode}} graph, such as sub-plan reuse, multiple input
> rewrite. We may introduce more optimizations in the future. So there are more
> and more difference between "Optimized Physical Plan" and "Optimized
> Execution Plan". We do not want to show tow execution plans, and "Physical
> Execution Plan" for {{StreamGraph}} is less important than "Optimized
> Execution Plan". If we want to introduce "Physical Execution Plan" in the
> future, we can add a type named "PHYSICAL_EXECUTION_PLAN" in
> {{ExplainDetail}} to support it. There is already an issue to do the similar
> things, [FLINK-19687|https://issues.apache.org/jira/browse/FLINK-19687]
> The following example show the explain result after adjustment:
> {code}
> == Abstract Syntax Tree ==
> LogicalLegacySink(name=[`default_catalog`.`default_database`.`upsertSink1`],
> fields=[a, cnt])
> +- LogicalProject(a=[$0], cnt=[$1])
> +- LogicalFilter(condition=[>($1, 10)])
> +- LogicalAggregate(group=[{0}], cnt=[COUNT()])
> +- LogicalProject(a=[$0])
> +- LogicalTableScan(table=[[default_catalog, default_database,
> MyTable1]])
> LogicalLegacySink(name=[`default_catalog`.`default_database`.`upsertSink2`],
> fields=[a, cnt])
> +- LogicalProject(a=[$0], cnt=[$1])
> +- LogicalFilter(condition=[<($1, 10)])
> +- LogicalAggregate(group=[{0}], cnt=[COUNT()])
> +- LogicalProject(a=[$0])
> +- LogicalTableScan(table=[[default_catalog, default_database,
> MyTable1]])
> == Optimized Physical Plan ==
> LegacySink(name=[`default_catalog`.`default_database`.`upsertSink1`],
> fields=[a, cnt])
> +- Calc(select=[a, cnt], where=[>(cnt, 10)])
> +- GroupAggregate(groupBy=[a], select=[a, COUNT(*) AS cnt])
> +- Exchange(distribution=[hash[a]])
> +- Calc(select=[a])
> +- DataStreamScan(table=[[default_catalog, default_database,
> MyTable1]], fields=[a, b, c])
> LegacySink(name=[`default_catalog`.`default_database`.`upsertSink2`],
> fields=[a, cnt])
> +- Calc(select=[a, cnt], where=[<(cnt, 10)])
> +- GroupAggregate(groupBy=[a], select=[a, COUNT(*) AS cnt])
> +- Exchange(distribution=[hash[a]])
> +- Calc(select=[a])
> +- DataStreamScan(table=[[default_catalog, default_database,
> MyTable1]], fields=[a, b, c])
> == Optimized Execution Plan ==
> GroupAggregate(groupBy=[a], select=[a, COUNT(*) AS cnt], reuse_id=[1])
> +- Exchange(distribution=[hash[a]])
> +- Calc(select=[a])
> +- DataStreamScan(table=[[default_catalog, default_database,
> MyTable1]], fields=[a, b, c])
> LegacySink(name=[`default_catalog`.`default_database`.`upsertSink1`],
> fields=[a, cnt])
> +- Calc(select=[a, cnt], where=[>(cnt, 10)])
> +- Reused(reference_id=[1])
> LegacySink(name=[`default_catalog`.`default_database`.`upsertSink2`],
> fields=[a, cnt])
> +- Calc(select=[a, cnt], where=[<(cnt, 10)])
> +- Reused(reference_id=[1])
> {code}
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
