[SPARK-18137][SQL]Fix RewriteDistinctAggregates UnresolvedException when a UDAF has a foldable TypeCheck by windpiger · Pull Request #15668 · apache/spark

windpiger · 2016-10-28T03:40:11Z

What changes were proposed in this pull request?

In RewriteDistinctAggregates rewrite funtion,after the UDAF's childs are mapped to AttributeRefference, If the UDAF(such as ApproximatePercentile) has a foldable TypeCheck for the input, It will failed because the AttributeRefference is not foldable,then the UDAF is not resolved, and then nullify on the unresolved object will throw a Exception.

In this PR, only map Unfoldable child to AttributeRefference, this can avoid the UDAF's foldable TypeCheck. and then only Expand Unfoldable child, there is no need to Expand a static value(foldable value).

Before sql result

select percentile_approxy(key,0.99999),count(distinct key),sume(distinc key) from src limit 1
org.apache.spark.sql.catalyst.analysis.UnresolvedException: Invalid call to dataType on unresolved object, tree: 'percentile_approx(CAST(src.key AS DOUBLE), CAST(0.99999BD AS DOUBLE), 10000)
at org.apache.spark.sql.catalyst.analysis.UnresolvedAttribute.dataType(unresolved.scala:92)
at org.apache.spark.sql.catalyst.optimizer.RewriteDistinctAggregates$.org$apache$spark$sql$catalyst$optimizer$RewriteDistinctAggregates$$nullify(RewriteDistinctAggregates.scala:261)

After sql result

select percentile_approxy(key,0.99999),count(distinct key),sume(distinc key) from src limit 1
[498.0,309,79136]

How was this patch tested?

Add a test case in HiveUDFSuit.

…hen the UDAF has a foldable TypeCheck

cloud-fan · 2016-10-28T06:02:38Z

OK to test

cloud-fan · 2016-10-28T07:44:47Z

ok to test

SparkQA · 2016-10-28T10:07:33Z

Test build #67698 has finished for PR 15668 at commit 7029e89.

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

hvanhovell · 2016-10-28T16:15:36Z

@windpiger I have taken a quick look at your PR. I think your approach has merit, however I was wondering if it is easier not to extract/move the foldable expressions at all?

cloud-fan · 2016-10-30T14:38:07Z

+1 on @hvanhovell 's idea, literals don't need to be pre-executed.

windpiger · 2016-10-31T02:57:21Z

@hvanhovell @cloud-fan Do you mean that just change distinctAggChildren/regularAggChildren to represent the unfoldable expressions, and then the followings code could be unchanged?

...
val distinctAggChildren = distinctAggGroups.keySet.flatten.toSeq.distinct.filter(!_.foldable)
...

val regularAggChildren = regularAggExprs.flatMap(.aggregateFunction.children).distinct.**filter(!.foldable)**

cloud-fan · 2016-11-01T14:00:32Z

It looks to me that after this PR, we still have literals in expand, can you call df.explain(true) to double check the query plan?

and how about this? https://github.com/apache/spark/compare/master...cloud-fan:showcase?expand=1 It's a little simpler and do avoid puting literals in expand.

windpiger · 2016-11-02T04:03:23Z

@cloud-fan yes,It is simpler,I check the query plan, my test case
sql("SELECTpercentile_approx(key, 0.99999)" + ", count(distinct key),sum(distinct key) FROM src LIMIT 1")
explain is ok as follows:
+- *Expand [List(null, null, 0, cast(key#29 as double)), List(key#29, null, 1, null), List(null, cast(key#29 as bigint), 2, null)], [src.key#38, CAST(src.keyAS BIGINT)#39L, gid#37, CAST(src.keyAS DOUBLE)#40] +- HiveTableScan [key#29], MetastoreRelation default, src

But, I miss one case that the distinct on a constant like:
sql("SELECTpercentile_approx(key, 0.99999)" + ", count(distinct key),sum(distinct key),count(distinct 1) FROM src LIMIT 1")
explain with a literal expand as follows:
+- *Expand [List(null, null, null, 0, cast(key#69 as double)), List(1, null, null, 1, null), List(null, key#69, null, 2, null), List(null, null, cast(key#69 as bigint), 3, null)], [src.key#80, CAST(src.keyAS BIGINT)#81L, gid#79, CAST(src.keyAS DOUBLE)#82] +- HiveTableScan [key#69], MetastoreRelation default, src

…hild;if has unfoldable children,it will only expand the unfoldable children

SparkQA · 2016-11-07T09:44:10Z

Test build #68267 has finished for PR 15668 at commit 8a6dd8d.

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

windpiger · 2016-11-07T10:00:51Z

@cloud-fan I rewrite the expand logic:

If the distinct aggFunction has unfoldable children,it will only expand the unfoldable children;
If the distinct aggFunction only has foldable children, it will only expand the first children, generally the aggFunction should not run foldable TypeChecker for the first child(this is the way to avoid foldable typechecker for the first child).

for example:
select percentile_approx(2,0.99999),sum(distinct 1),count(distinct 1,2,3,1) from src limit 1

explained:
== Physical Plan == CollectLimit 1 +- *HashAggregate(keys=[], functions=[first(if ((gid#177 = 0)) percentile_approx(CAST(2 AS DOUBLE), CAST(0.99999BD AS DOUBLE), 10000)#180 else null, true), sum(if ((gid#177 = 1)) CAST(1 AS BIGINT)#178L else null), count(if ((gid#177 = 2)) 1#179 else null, 2, 3, if ((gid#177 = 2)) 1#179 else null)]) +- Exchange SinglePartition +- *HashAggregate(keys=[], functions=[partial_first(if ((gid#177 = 0)) percentile_approx(CAST(2 AS DOUBLE), CAST(0.99999BD AS DOUBLE), 10000)#180 else null, true), partial_sum(if ((gid#177 = 1)) CAST(1 AS BIGINT)#178L else null), partial_count(if ((gid#177 = 2)) 1#179 else null, 2, 3, if ((gid#177 = 2)) 1#179 else null)]) +- SortAggregate(key=[CAST(1 AS BIGINT)#178L, 1#179, gid#177], functions=[percentile_approx(2.0, 0.99999, 10000, 0, 0)]) +- *Sort [CAST(1 AS BIGINT)#178L ASC NULLS FIRST, 1#179 ASC NULLS FIRST, gid#177 ASC NULLS FIRST], false, 0 +- Exchange hashpartitioning(CAST(1 AS BIGINT)#178L, 1#179, gid#177, 200) +- SortAggregate(key=[CAST(1 AS BIGINT)#178L, 1#179, gid#177], functions=[partial_percentile_approx(2.0, 0.99999, 10000, 0, 0)]) +- *Sort [CAST(1 AS BIGINT)#178L ASC NULLS FIRST, 1#179 ASC NULLS FIRST, gid#177 ASC NULLS FIRST], false, 0 +- *Expand [List(null, null, 0), List(1, null, 1), List(null, 1, 2)], [CAST(1 AS BIGINT)#178L, 1#179, gid#177] +- HiveTableScan MetastoreRelation default, src
and the result is:
+--------------------------------------------------------------------+---------------+--------------------------+ |percentile_approx(CAST(2 AS DOUBLE), CAST(0.99999 AS DOUBLE), 10000)|sum(DISTINCT 1)|count(DISTINCT 1, 2, 3, 1)| +--------------------------------------------------------------------+---------------+--------------------------+ | 2.0| 1| 1| +--------------------------------------------------------------------+---------------+--------------------------+

hvanhovell

This is shaping up nicely. I left a few minor comments.

hvanhovell · 2016-11-07T08:44:50Z

 import org.apache.spark.sql.types.IntegerType

-/**
+/*


can you revert this? this breaks scaladoc.

hvanhovell · 2016-11-07T08:47:25Z

-// NamedExpression. This is done to prevent collisions between distinct and regular aggregate
-// children, in this case attribute reuse causes the input of the regular aggregate to bound to
-// the (nulled out) input of the distinct aggregate.
+ // We are creating a new reference here instead of reusing the attribute in case of a


Revert this change

hvanhovell · 2016-11-07T08:47:39Z

 val transformations: Map[Expression, Expression] =
-(distinctAggOperatorMap.flatMap(_._2) ++
-regularAggOperatorMap.map(e => (e._1, e._3))).toMap
+ (distinctAggOperatorMap.flatMap(_._2) ++


Revert this change

hvanhovell · 2016-11-07T08:48:48Z

 }

 test("Generic UDAF aggregates") {
+ checkAnswer(sql("SELECT percentile_approx(2, 0.99999), " +


Use multiline strings for these tests.

hvanhovell · 2016-11-07T08:52:55Z

- val distinctAggGroups = aggExpressions
- .filter(_.isDistinct)
- .groupBy(_.aggregateFunction.children.toSet)
+ val distinctAggGroups = aggExpressions.filter(_.isDistinct).groupBy{


NIT: Space between groupBy and bracket.

hvanhovell · 2016-11-07T08:53:32Z

- .groupBy(_.aggregateFunction.children.toSet)
+ val distinctAggGroups = aggExpressions.filter(_.isDistinct).groupBy{
+ e =>
+ if (e.aggregateFunction.children.exists(!_.foldable)) {


Just materialize the nonFoldables. Instead of filtering them twice.

hvanhovell · 2016-11-07T08:54:45Z

- af: AggregateFunction)(
- attrs: Expression => Expression): AggregateFunction = {
- af.withNewChildren(af.children.map(attrs)).asInstanceOf[AggregateFunction]
+ def patchAggregateFunctionChildren(af: AggregateFunction)(


NIT: Style, please keep this the way it was.

hvanhovell · 2016-11-07T11:23:56Z

+ // count(distinct 1) will be explained to count(1) after the rewrite function.
+ // Generally, the distinct aggregateFunction should not run
+ // foldable TypeCheck for the first child.
+ e.aggregateFunction.children.take(1).toSet


This is a good catch. It would be great if we could git rid of this by constant folding (not needed in this PR). Another way of getting rid of this, would be by creating a separate processing group for these distincts.

SparkQA · 2016-11-07T12:14:58Z

Test build #68270 has finished for PR 15668 at commit c5b3a3d.

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

SparkQA · 2016-11-07T15:21:09Z

Test build #68274 has finished for PR 15668 at commit 67fc72d.

This patch fails Spark unit tests.
This patch merges cleanly.
This patch adds no public classes.

windpiger · 2016-11-07T16:04:50Z

retest this please

windpiger · 2016-11-08T01:45:49Z

@cloud-fan @hvanhovell could you please help to retest this？

cloud-fan · 2016-11-08T03:43:07Z

retest this please

cloud-fan · 2016-11-08T03:43:20Z

add to whitelist

cloud-fan · 2016-11-08T03:45:47Z

- val distinctAggGroups = aggExpressions
- .filter(_.isDistinct)
- .groupBy(_.aggregateFunction.children.toSet)
+ val distinctAggGroups = aggExpressions.filter(_.isDistinct).groupBy {


nit:
...groupBy { e => ... }

cloud-fan · 2016-11-08T03:50:28Z

LGTM, pending jenkins

SparkQA · 2016-11-08T06:09:09Z

Test build #68315 has finished for PR 15668 at commit 67fc72d.

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

SparkQA · 2016-11-08T08:34:29Z

Test build #68323 has finished for PR 15668 at commit 6e58167.

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

hvanhovell · 2016-11-08T11:08:42Z

LGTM. Merging to master/2.1/2.0. Thanks!

…when a UDAF has a foldable TypeCheck ## What changes were proposed in this pull request? In RewriteDistinctAggregates rewrite funtion,after the UDAF's childs are mapped to AttributeRefference, If the UDAF(such as ApproximatePercentile) has a foldable TypeCheck for the input, It will failed because the AttributeRefference is not foldable,then the UDAF is not resolved, and then nullify on the unresolved object will throw a Exception. In this PR, only map Unfoldable child to AttributeRefference, this can avoid the UDAF's foldable TypeCheck. and then only Expand Unfoldable child, there is no need to Expand a static value(foldable value). **Before sql result** > select percentile_approxy(key,0.99999),count(distinct key),sume(distinc key) from src limit 1 > org.apache.spark.sql.catalyst.analysis.UnresolvedException: Invalid call to dataType on unresolved object, tree: 'percentile_approx(CAST(src.`key` AS DOUBLE), CAST(0.99999BD AS DOUBLE), 10000) > at org.apache.spark.sql.catalyst.analysis.UnresolvedAttribute.dataType(unresolved.scala:92) > at org.apache.spark.sql.catalyst.optimizer.RewriteDistinctAggregates$.org$apache$spark$sql$catalyst$optimizer$RewriteDistinctAggregates$$nullify(RewriteDistinctAggregates.scala:261) **After sql result** > select percentile_approxy(key,0.99999),count(distinct key),sume(distinc key) from src limit 1 > [498.0,309,79136] ## How was this patch tested? Add a test case in HiveUDFSuit. Author: root <root@iZbp1gsnrlfzjxh82cz80vZ.(none)> Closes#15668 from windpiger/RewriteDistinctUDAFUnresolveExcep. (cherry picked from commit c291bd2) Signed-off-by: Herman van Hovell <hvanhovell@databricks.com>

hvanhovell · 2016-11-08T11:12:23Z

@windpiger do you have a JIRA username? So I can credit you on the JIRA.

windpiger · 2016-11-08T11:18:01Z

@hvanhovell username is Song Jun

…when a UDAF has a foldable TypeCheck ## What changes were proposed in this pull request? In RewriteDistinctAggregates rewrite funtion,after the UDAF's childs are mapped to AttributeRefference, If the UDAF(such as ApproximatePercentile) has a foldable TypeCheck for the input, It will failed because the AttributeRefference is not foldable,then the UDAF is not resolved, and then nullify on the unresolved object will throw a Exception. In this PR, only map Unfoldable child to AttributeRefference, this can avoid the UDAF's foldable TypeCheck. and then only Expand Unfoldable child, there is no need to Expand a static value(foldable value). **Before sql result** > select percentile_approxy(key,0.99999),count(distinct key),sume(distinc key) from src limit 1 > org.apache.spark.sql.catalyst.analysis.UnresolvedException: Invalid call to dataType on unresolved object, tree: 'percentile_approx(CAST(src.`key` AS DOUBLE), CAST(0.99999BD AS DOUBLE), 10000) > at org.apache.spark.sql.catalyst.analysis.UnresolvedAttribute.dataType(unresolved.scala:92) > at org.apache.spark.sql.catalyst.optimizer.RewriteDistinctAggregates$.org$apache$spark$sql$catalyst$optimizer$RewriteDistinctAggregates$$nullify(RewriteDistinctAggregates.scala:261) **After sql result** > select percentile_approxy(key,0.99999),count(distinct key),sume(distinc key) from src limit 1 > [498.0,309,79136] ## How was this patch tested? Add a test case in HiveUDFSuit. Author: root <root@iZbp1gsnrlfzjxh82cz80vZ.(none)> Closesapache#15668 from windpiger/RewriteDistinctUDAFUnresolveExcep.

[SPARK-18137][SQL]Fix RewriteDistinctAggregates UnresolvedException w…
7029e89
…hen the UDAF has a foldable TypeCheck

windpiger changed the title ~~[SPARK-18137][SQL]Fix RewriteDistinctAggregates UnresolvedException w…~~[SPARK-18137][SQL]Fix RewriteDistinctAggregates UnresolvedException when a UDAF has a foldable TypeCheckOct 28, 2016

if distinct agg only has foldable children,it will expand the first c…
8a6dd8d
…hild;if has unfoldable children,it will only expand the unfoldable children

fix some code style
c5b3a3d

hvanhovell requested changes Nov 7, 2016
View reviewed changes

fix some code style
67fc72d

cloud-fan reviewed Nov 8, 2016
View reviewed changes

fix a code style
6e58167

asfgit closed this in c291bd2Nov 8, 2016

davidnavas pushed a commit to davidnavas/spark that referenced this pull request Jun 7, 2017
Backport ofapache/pull/15668
e757de5

davidnavas mentioned this pull request Jun 7, 2017
WEB-11368: Skip literals so that HiveUDAF percentile_approx can operate correctly alteryx/spark#186
Merged

Uh oh!

Conversation

windpiger commented Oct 28, 2016

What changes were proposed in this pull request?

How was this patch tested?

Uh oh!

cloud-fan commented Oct 28, 2016

Uh oh!

cloud-fan commented Oct 28, 2016

Uh oh!

SparkQA commented Oct 28, 2016

Uh oh!

hvanhovell commented Oct 28, 2016

Uh oh!

cloud-fan commented Oct 30, 2016

Uh oh!

windpiger commented Oct 31, 2016

Uh oh!

cloud-fan commented Nov 1, 2016

Uh oh!

windpiger commented Nov 2, 2016

Uh oh!

SparkQA commented Nov 7, 2016

Uh oh!

windpiger commented Nov 7, 2016• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

hvanhovell left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

SparkQA commented Nov 7, 2016

Uh oh!

SparkQA commented Nov 7, 2016

Uh oh!

windpiger commented Nov 7, 2016

Uh oh!

windpiger commented Nov 8, 2016

Uh oh!

cloud-fan commented Nov 8, 2016

Uh oh!

cloud-fan commented Nov 8, 2016

Uh oh!

Choose a reason for hiding this comment

Uh oh!

cloud-fan commented Nov 8, 2016

Uh oh!

SparkQA commented Nov 8, 2016

Uh oh!

SparkQA commented Nov 8, 2016

Uh oh!

hvanhovell commented Nov 8, 2016• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

hvanhovell commented Nov 8, 2016

Uh oh!

windpiger commented Nov 8, 2016

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

windpiger commented Nov 7, 2016•
edited
Loading

hvanhovell commented Nov 8, 2016•
edited
Loading