[SPARK-21555][SQL] RuntimeReplaceable should be compared semantically by its canonicalized child by viirya · Pull Request #18761 · apache/spark

viirya · 2017-07-28T14:38:21Z

What changes were proposed in this pull request?

When there are aliases (these aliases were added for nested fields) as parameters in RuntimeReplaceable, as they are not in the children expression, those aliases can't be cleaned up in analyzer rule CleanupAliases.

An expression nvl(foo.foo1, "value") can be resolved to two semantically different expressions in a group by query because they contain different aliases.

Because those aliases are not children of RuntimeReplaceable which is an UnaryExpression. So we can't trim the aliases out by simple transforming the expressions in CleanupAliases.

If we want to replace the non-children aliases in RuntimeReplaceable, we need to add more codes to RuntimeReplaceable and modify all expressions of RuntimeReplaceable. It makes the interface ugly IMO.

Consider those aliases will be replaced later at optimization and so they're no harm, this patch chooses to simply override canonicalized of RuntimeReplaceable.

One concern is about CleanupAliases. Because it actually cannot clean up ALL aliases inside a plan. To make caller of this rule notice that, this patch adds a comment to CleanupAliases.

How was this patch tested?

Added test.

SparkQA · 2017-07-28T17:08:36Z

Test build #80025 has finished for PR 18761 at commit d0d5804.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

viirya · 2017-07-29T00:39:37Z

cc @cloud-fan @gatorsmile Please take a look when you have some time. Thanks.

gatorsmile · 2017-07-29T06:19:51Z

 override def nullable: Boolean = child.nullable
 override def foldable: Boolean = child.foldable
 override def dataType: DataType = child.dataType
+ override lazy val canonicalized: Expression = child.canonicalized


Although this fixes the issue, the root cause is the unneeded aliases (these aliases were added for nested fields) are not properly cleaned up in these RuntimeReplaceable expressions by the rule CleanupAliases.

I know it. However, the aliases actually are replaced by Optimizer because RuntimeReplaceable. So they are no harm.

Btw, those aliases are not children of RuntimeReplaceable which is an UnaryExpression. So we can't trim the aliases out by simple transforming the expressions in CleanupAliases.
If we want to replace the non-children aliases in RuntimeReplaceable, we need to add more codes to RuntimeReplaceable and modify all expressions of RuntimeReplaceable. It makes the interface ugly IMO.
Consider those aliases will be replaced soon and no harm of them, that's why I choose to simply override canonicalized.

First, my only concern is CleanupAliases might be called in the other Analyzer rules. These unneeded aliases could break the assumptions made by the caller of CleanupAliases.
Second, you should post what I said above in the PR description. So far, the description is not clear to the reviewers.
Third, I am ok about the fix, but we should write the comments to explain why we did it here. We need to add two comments. One is in CleanupAliases ; another is above this line to explain why we add canonicalized.

Updated PR description and comments accordingly.

If you meant there is an assumption that CleanupAliases would clean up ALL aliases in a plan, that is not correct. It can only clean up the aliases reachable by expression transformation. I added a comment to clarify this.

gatorsmile · 2017-07-29T07:45:05Z

+CREATE TABLE test(a INT, foo STRUCT<foo1:STRING,foo2:STRING>) USING parquet;
+INSERT INTO test VALUES(1, ("value1", "value2"));
+SELECT nvl(foo.foo1, "value"), count(*) FROM test GROUP BY nvl(foo.foo1, "value");
+DROP TABLE test;


You can simplify the test case to something like
CREATE TEMPORARY VIEW tempView1 ASVALUES (1, NAMED_STRUCT('col1', 'gamma', 'col2', 'delta')) AS T(id, st)")SELECT nvl(st.col1, "value"), count(*) FROM from tempView1 GROUP BY nvl(st.col1, "value")

gatorsmile · 2017-07-29T07:45:39Z

BTW, please also update the PR title.

viirya · 2017-07-29T08:27:33Z

I hope this PR description can be good for you.

SparkQA · 2017-07-29T10:49:21Z

Test build #80039 has finished for PR 18761 at commit 82d54ed.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

gatorsmile · 2017-07-29T17:01:46Z

It looks much better now. Since CleanupAliases has such an issue, could you check whether the other expressions are facing the same issue?

gatorsmile · 2017-07-29T17:01:51Z

LGTM

… by its canonicalized child ## What changes were proposed in this pull request? When there are aliases (these aliases were added for nested fields) as parameters in `RuntimeReplaceable`, as they are not in the children expression, those aliases can't be cleaned up in analyzer rule `CleanupAliases`. An expression `nvl(foo.foo1, "value")` can be resolved to two semantically different expressions in a group by query because they contain different aliases. Because those aliases are not children of `RuntimeReplaceable` which is an `UnaryExpression`. So we can't trim the aliases out by simple transforming the expressions in `CleanupAliases`. If we want to replace the non-children aliases in `RuntimeReplaceable`, we need to add more codes to `RuntimeReplaceable` and modify all expressions of `RuntimeReplaceable`. It makes the interface ugly IMO. Consider those aliases will be replaced later at optimization and so they're no harm, this patch chooses to simply override `canonicalized` of `RuntimeReplaceable`. One concern is about `CleanupAliases`. Because it actually cannot clean up ALL aliases inside a plan. To make caller of this rule notice that, this patch adds a comment to `CleanupAliases`. ## How was this patch tested? Added test. Author: Liang-Chi Hsieh <viirya@gmail.com> Closes#18761 from viirya/SPARK-21555. (cherry picked from commit 9c8109e) Signed-off-by: gatorsmile <gatorsmile@gmail.com>

gatorsmile · 2017-07-29T17:04:39Z

Thanks! Merging to master/2.2/2.1

… by its canonicalized child ## What changes were proposed in this pull request? When there are aliases (these aliases were added for nested fields) as parameters in `RuntimeReplaceable`, as they are not in the children expression, those aliases can't be cleaned up in analyzer rule `CleanupAliases`. An expression `nvl(foo.foo1, "value")` can be resolved to two semantically different expressions in a group by query because they contain different aliases. Because those aliases are not children of `RuntimeReplaceable` which is an `UnaryExpression`. So we can't trim the aliases out by simple transforming the expressions in `CleanupAliases`. If we want to replace the non-children aliases in `RuntimeReplaceable`, we need to add more codes to `RuntimeReplaceable` and modify all expressions of `RuntimeReplaceable`. It makes the interface ugly IMO. Consider those aliases will be replaced later at optimization and so they're no harm, this patch chooses to simply override `canonicalized` of `RuntimeReplaceable`. One concern is about `CleanupAliases`. Because it actually cannot clean up ALL aliases inside a plan. To make caller of this rule notice that, this patch adds a comment to `CleanupAliases`. ## How was this patch tested? Added test. Author: Liang-Chi Hsieh <viirya@gmail.com> Closes#18761 from viirya/SPARK-21555. (cherry picked from commit 9c8109e) Signed-off-by: gatorsmile <gatorsmile@gmail.com>

viirya · 2017-07-30T00:03:55Z

@gatorsmile OK. Usually the constructor expression parameters should also be in children. RuntimeReplaceable is a special case. I'd check if there are other expressions like it.

… by its canonicalized child ## What changes were proposed in this pull request? When there are aliases (these aliases were added for nested fields) as parameters in `RuntimeReplaceable`, as they are not in the children expression, those aliases can't be cleaned up in analyzer rule `CleanupAliases`. An expression `nvl(foo.foo1, "value")` can be resolved to two semantically different expressions in a group by query because they contain different aliases. Because those aliases are not children of `RuntimeReplaceable` which is an `UnaryExpression`. So we can't trim the aliases out by simple transforming the expressions in `CleanupAliases`. If we want to replace the non-children aliases in `RuntimeReplaceable`, we need to add more codes to `RuntimeReplaceable` and modify all expressions of `RuntimeReplaceable`. It makes the interface ugly IMO. Consider those aliases will be replaced later at optimization and so they're no harm, this patch chooses to simply override `canonicalized` of `RuntimeReplaceable`. One concern is about `CleanupAliases`. Because it actually cannot clean up ALL aliases inside a plan. To make caller of this rule notice that, this patch adds a comment to `CleanupAliases`. ## How was this patch tested? Added test. Author: Liang-Chi Hsieh <viirya@gmail.com> Closesapache#18761 from viirya/SPARK-21555. (cherry picked from commit 9c8109e) Signed-off-by: gatorsmile <gatorsmile@gmail.com>

… by its canonicalized child When there are aliases (these aliases were added for nested fields) as parameters in `RuntimeReplaceable`, as they are not in the children expression, those aliases can't be cleaned up in analyzer rule `CleanupAliases`. An expression `nvl(foo.foo1, "value")` can be resolved to two semantically different expressions in a group by query because they contain different aliases. Because those aliases are not children of `RuntimeReplaceable` which is an `UnaryExpression`. So we can't trim the aliases out by simple transforming the expressions in `CleanupAliases`. If we want to replace the non-children aliases in `RuntimeReplaceable`, we need to add more codes to `RuntimeReplaceable` and modify all expressions of `RuntimeReplaceable`. It makes the interface ugly IMO. Consider those aliases will be replaced later at optimization and so they're no harm, this patch chooses to simply override `canonicalized` of `RuntimeReplaceable`. One concern is about `CleanupAliases`. Because it actually cannot clean up ALL aliases inside a plan. To make caller of this rule notice that, this patch adds a comment to `CleanupAliases`. Added test. Author: Liang-Chi Hsieh <viirya@gmail.com> Closesapache#18761 from viirya/SPARK-21555. (cherry picked from commit 9c8109e) Signed-off-by: gatorsmile <gatorsmile@gmail.com>

GROUP BY should work with expressions with Nvl.
d0d5804

gatorsmile reviewed Jul 29, 2017
View reviewed changes

Address comments.
82d54ed

viirya changed the title ~~[SPARK-21555][SQL] GROUP BY should work with expressions with Nvl~~[SPARK-21555][SQL] RuntimeReplaceable should be compared semantically by its canonicalized childJul 29, 2017

asfgit closed this in 9c8109eJul 29, 2017

viirya deleted the SPARK-21555 branch December 27, 2023 18:34

Uh oh!

Conversation

viirya commented Jul 28, 2017• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What changes were proposed in this pull request?

How was this patch tested?

Uh oh!

SparkQA commented Jul 28, 2017

Uh oh!

viirya commented Jul 29, 2017

Uh oh!

gatorsmileJul 29, 2017• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

viiryaJul 29, 2017

Choose a reason for hiding this comment

Uh oh!

viiryaJul 29, 2017

Choose a reason for hiding this comment

Uh oh!

gatorsmileJul 29, 2017

Choose a reason for hiding this comment

Uh oh!

viiryaJul 29, 2017

Choose a reason for hiding this comment

Uh oh!

viiryaJul 29, 2017

Choose a reason for hiding this comment

Uh oh!

gatorsmileJul 29, 2017

Choose a reason for hiding this comment

Uh oh!

gatorsmile commented Jul 29, 2017

Uh oh!

viirya commented Jul 29, 2017

Uh oh!

SparkQA commented Jul 29, 2017

Uh oh!

gatorsmile commented Jul 29, 2017

Uh oh!

gatorsmile commented Jul 29, 2017

Uh oh!

gatorsmile commented Jul 29, 2017

Uh oh!

viirya commented Jul 30, 2017

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

viirya commented Jul 28, 2017•
edited
Loading

gatorsmileJul 29, 2017•
edited
Loading