### What changes were proposed in this pull request?
This PR changes key data types check in `HashJoin` to use `sameType`.
### Why are the changes needed?
Looks at the resolving condition of `SetOperation`, it requires only each left data types should be `sameType` as the right ones. Logically the `EqualTo` expression in equi-join, also requires only left data type `sameType` as right data type. Then `HashJoin` requires left keys data type exactly the same as right keys data type, looks not reasonable.
It makes inconsistent results when doing `except` between two dataframes.
If two dataframes don't have nested fields, even their field nullable property different, `HashJoin` passes the key type check because it checks field individually so field nullable property is ignored.
If two dataframes have nested fields like struct, `HashJoin` fails the key type check because now it compare two struct types and nullable property now affects.
### Does this PR introduce _any_ user-facing change?
Yes. Making consistent `except` operation between dataframes.
### How was this patch tested?
Unit test.
Closes#29555 from viirya/SPARK-32693.
Authored-by: Liang-Chi Hsieh <viirya@gmail.com>
Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
### What changes were proposed in this pull request?
This PR let JDBC clients identify spark interval columns properly.
### Why are the changes needed?
JDBC users can query interval values through thrift server, create views with interval columns, e.g.
```sql
CREATE global temp view view1 as select interval 1 day as i;
```
but when they want to get the details of the columns of view1, the will fail with `Unrecognized type name: INTERVAL`
```
Caused by: java.lang.IllegalArgumentException: Unrecognized type name: INTERVAL
at org.apache.hadoop.hive.serde2.thrift.Type.getType(Type.java:170)
at org.apache.spark.sql.hive.thriftserver.ThriftserverShimUtils$.toJavaSQLType(ThriftserverShimUtils.scala:53)
at org.apache.spark.sql.hive.thriftserver.SparkGetColumnsOperation.$anonfun$addToRowSet$1(SparkGetColumnsOperation.scala:157)
at scala.collection.Iterator.foreach(Iterator.scala:941)
at scala.collection.Iterator.foreach$(Iterator.scala:941)
at scala.collection.AbstractIterator.foreach(Iterator.scala:1429)
at scala.collection.IterableLike.foreach(IterableLike.scala:74)
at scala.collection.IterableLike.foreach$(IterableLike.scala:73)
at org.apache.spark.sql.types.StructType.foreach(StructType.scala:102)
at org.apache.spark.sql.hive.thriftserver.SparkGetColumnsOperation.addToRowSet(SparkGetColumnsOperation.scala:149)
at org.apache.spark.sql.hive.thriftserver.SparkGetColumnsOperation.$anonfun$runInternal$6(SparkGetColumnsOperation.scala:113)
at org.apache.spark.sql.hive.thriftserver.SparkGetColumnsOperation.$anonfun$runInternal$6$adapted(SparkGetColumnsOperation.scala:112)
at scala.Option.foreach(Option.scala:407)
at org.apache.spark.sql.hive.thriftserver.SparkGetColumnsOperation.$anonfun$runInternal$5(SparkGetColumnsOperation.scala:112)
at org.apache.spark.sql.hive.thriftserver.SparkGetColumnsOperation.$anonfun$runInternal$5$adapted(SparkGetColumnsOperation.scala:111)
at scala.collection.mutable.ResizableArray.foreach(ResizableArray.scala:62)
at scala.collection.mutable.ResizableArray.foreach$(ResizableArray.scala:55)
at scala.collection.mutable.ArrayBuffer.foreach(ArrayBuffer.scala:49)
at org.apache.spark.sql.hive.thriftserver.SparkGetColumnsOperation.runInternal(SparkGetColumnsOperation.scala:111)
... 34 more
```
### Does this PR introduce _any_ user-facing change?
YES,
#### before
#### after
### How was this patch tested?
new tests
Closes#29539 from yaooqinn/SPARK-32696.
Authored-by: Kent Yao <yaooqinn@hotmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
Currently, EmptyHashedRelation and HashedRelationWithAllNullKeys is an object, and it will cause JavaDeserialization Exception as following
```
20/08/26 11:13:30 WARN [task-result-getter-2] TaskSetManager: Lost task 34.0 in stage 57.0 (TID 18076, emr-worker-5.cluster-183257, executor 18): java.io.InvalidClassException: org.apache.spark.sql.execution.joins.EmptyHashedRelation$; no valid constructor
at java.io.ObjectStreamClass$ExceptionInfo.newInvalidClassException(ObjectStreamClass.java:169)
at java.io.ObjectStreamClass.checkDeserialize(ObjectStreamClass.java:874)
at java.io.ObjectInputStream.readOrdinaryObject(ObjectInputStream.java:2042)
at java.io.ObjectInputStream.readObject0(ObjectInputStream.java:1572)
at java.io.ObjectInputStream.readObject(ObjectInputStream.java:430)
at org.apache.spark.serializer.JavaDeserializationStream.readObject(JavaSerializer.scala:76)
at org.apache.spark.broadcast.TorrentBroadcast$.$anonfun$unBlockifyObject$4(TorrentBroadcast.scala:328)
```
This PR includes
* Using case object instead to fix serialization issue.
* Also change EmptyHashedRelation not to extend NullAwareHashedRelation since it's already being used in other non-NAAJ joins.
### Why are the changes needed?
It will cause BHJ failed when buildSide is Empty and BHJ(NAAJ) failed when buildSide with null partition keys.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
* Existing UT.
* Run entire TPCDS for E2E coverage.
Closes#29547 from leanken/leanken-SPARK-32705.
Authored-by: xuewei.linxuewei <xuewei.linxuewei@alibaba-inc.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
This is a follow up PR to #29328 to apply the same constraint where `path` option cannot coexist with path parameter to `DataFrameWriter.save()`, `DataStreamReader.load()` and `DataStreamWriter.start()`.
### Why are the changes needed?
The current behavior silently overwrites the `path` option if path parameter is passed to `DataFrameWriter.save()`, `DataStreamReader.load()` and `DataStreamWriter.start()`.
For example,
```
Seq(1).toDF.write.option("path", "/tmp/path1").parquet("/tmp/path2")
```
will write the result to `/tmp/path2`.
### Does this PR introduce _any_ user-facing change?
Yes, if `path` option coexists with path parameter to any of the above methods, it will throw `AnalysisException`:
```
scala> Seq(1).toDF.write.option("path", "/tmp/path1").parquet("/tmp/path2")
org.apache.spark.sql.AnalysisException: There is a 'path' option set and save() is called with a path parameter. Either remove the path option, or call save() without the parameter. To ignore this check, set 'spark.sql.legacy.pathOptionBehavior.enabled' to 'true'.;
```
The user can restore the previous behavior by setting `spark.sql.legacy.pathOptionBehavior.enabled` to `true`.
### How was this patch tested?
Added new tests.
Closes#29543 from imback82/path_option.
Authored-by: Terry Kim <yuminkim@gmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
Use `InSet` expression to fix data issue when pruning DPP on non-atomic type. for example:
```scala
spark.range(1000)
.select(col("id"), col("id").as("k"))
.write
.partitionBy("k")
.format("parquet")
.mode("overwrite")
.saveAsTable("df1");
spark.range(100)
.select(col("id"), col("id").as("k"))
.write
.partitionBy("k")
.format("parquet")
.mode("overwrite")
.saveAsTable("df2")
spark.sql("set spark.sql.optimizer.dynamicPartitionPruning.fallbackFilterRatio=2")
spark.sql("set spark.sql.optimizer.dynamicPartitionPruning.reuseBroadcastOnly=false")
spark.sql("SELECT df1.id, df2.k FROM df1 JOIN df2 ON struct(df1.k) = struct(df2.k) AND df2.id < 2").show
```
It should return two records, but it returns empty.
### Why are the changes needed?
Fix data issue
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Add new unit test.
Closes#29475 from wangyum/SPARK-32659.
Authored-by: Yuming Wang <yumwang@ebay.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
Instead of deleting the data, we can move the data to trash.
Based on the configuration provided by the user it will be deleted permanently from the trash.
### Why are the changes needed?
Instead of directly deleting the data, we can provide flexibility to move data to the trash and then delete it permanently.
### Does this PR introduce _any_ user-facing change?
Yes, After truncate table the data is not permanently deleted now.
It is first moved to the trash and then after the given time deleted permanently;
### How was this patch tested?
new UTs added
Closes#29387 from Udbhav30/tuncateTrash.
Authored-by: Udbhav30 <u.agrawal30@gmail.com>
Signed-off-by: Dongjoon Hyun <dongjoon@apache.org>
### What changes were proposed in this pull request?
This PR regenerates the golden explain file based on the fix: https://github.com/apache/spark/pull/29537
### Why are the changes needed?
Eliminates the personal related information (e.g., local directories) in the explain plan.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Checked manually.
Closes#29546 from Ngone51/follow-up-gen-golden-file.
Authored-by: yi.wu <yi.wu@databricks.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
Spark's CSV source can optionally ignore lines starting with a comment char. Some code paths check to see if it's set before applying comment logic (i.e. not set to default of `\0`), but many do not, including the one that passes the option to Univocity. This means that rows beginning with a null char were being treated as comments even when 'disabled'.
### Why are the changes needed?
To avoid dropping rows that start with a null char when this is not requested or intended. See JIRA for an example.
### Does this PR introduce _any_ user-facing change?
Nothing beyond the effect of the bug fix.
### How was this patch tested?
Existing tests plus new test case.
Closes#29516 from srowen/SPARK-32614.
Authored-by: Sean Owen <srowen@gmail.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
1. Extract `SQLQueryTestSuite.replaceNotIncludedMsg` to `PlanTest`.
2. Reuse `replaceNotIncludedMsg` to normalize the explain plan that generated in `PlanStabilitySuite`.
### Why are the changes needed?
This's a follow-up of https://github.com/apache/spark/pull/29270.
Eliminates the personal related information (e.g., local directories) in the explain plan.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Updated test.
Closes#29537 from Ngone51/follow-up-plan-stablity.
Authored-by: yi.wu <yi.wu@databricks.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
…er operations"
### What changes were proposed in this pull request?
This reverts commit 510a1656e6.
### Why are the changes needed?
see https://github.com/apache/spark/pull/29204#discussion_r475716547
### Does this PR introduce _any_ user-facing change?
NO
### How was this patch tested?
pass ci tools
Closes#29531 from yaooqinn/revert.
Authored-by: Kent Yao <yaooqinn@hotmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
There is a bug in the way the optimizer rule in `SimplifyExtractValueOps` is currently written in master branch which yields incorrect results in scenarios like the following:
```
sql("SELECT CAST(NULL AS struct<a:int,b:int>) struct_col")
.select($"struct_col".withField("d", lit(4)).getField("d").as("d"))
// currently returns this:
+---+
|d |
+---+
|4 |
+---+
// when in fact it should return this:
+----+
|d |
+----+
|null|
+----+
```
The changes in this PR will fix this bug.
### Why are the changes needed?
To fix the aforementioned bug. Optimizer rules should improve the performance of the query but yield exactly the same results.
### Does this PR introduce _any_ user-facing change?
Yes, this bug will no longer occur.
That said, this isn't something to be concerned about as this bug was introduced in Spark 3.1 and Spark 3.1 has yet to be released.
### How was this patch tested?
Unit tests were added. Jenkins must pass them.
Closes#29522 from fqaiser94/SPARK-32641.
Authored-by: fqaiser94@gmail.com <fqaiser94@gmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
This PR proposes to fix ORC predicate pushdown under case-insensitive analysis case. The field names in pushed down predicates don't need to match in exact letter case with physical field names in ORC files, if we enable case-insensitive analysis.
This is re-submitted for #29457. Because #29457 has a hive-1.2 error and there were some tests failed with hive-1.2 profile at the same time, #29457 was reverted to unblock others.
### Why are the changes needed?
Currently ORC predicate pushdown doesn't work with case-insensitive analysis. A predicate "a < 0" cannot pushdown to ORC file with field name "A" under case-insensitive analysis.
But Parquet predicate pushdown works with this case. We should make ORC predicate pushdown work with case-insensitive analysis too.
### Does this PR introduce _any_ user-facing change?
Yes, after this PR, under case-insensitive analysis, ORC predicate pushdown will work.
### How was this patch tested?
Unit tests.
Closes#29530 from viirya/fix-orc-pushdown.
Authored-by: Liang-Chi Hsieh <viirya@gmail.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
This PR enhances `Catalog.createTable()` to allow users to set the table's description. This corresponds to the following SQL syntax:
```sql
CREATE TABLE ...
COMMENT 'this is a fancy table';
```
### Why are the changes needed?
This brings the Scala/Python catalog APIs a bit closer to what's already possible via SQL.
### Does this PR introduce any user-facing change?
Yes, it adds a new parameter to `Catalog.createTable()`.
### How was this patch tested?
Existing unit tests:
```sh
./python/run-tests \
--python-executables python3.7 \
--testnames 'pyspark.sql.tests.test_catalog,pyspark.sql.tests.test_context'
```
```
$ ./build/sbt
testOnly org.apache.spark.sql.internal.CatalogSuite org.apache.spark.sql.CachedTableSuite org.apache.spark.sql.hive.MetastoreDataSourcesSuite org.apache.spark.sql.hive.execution.HiveDDLSuite
```
Closes#27908 from nchammas/SPARK-31000-table-description.
Authored-by: Nicholas Chammas <nicholas.chammas@liveramp.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
Added Java docs for Long data types in the Row class.
### Why are the changes needed?
The Long datatype is somehow missing in Row.scala's `apply` and `get` methods.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
Existing UTs.
Closes#29534 from yeshengm/docs-fix.
Authored-by: Yesheng Ma <kimi.ysma@gmail.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
Change to use `dataTypes.foreach` instead of get the element use specified index in `def this(dataTypes: Seq[DataType]) `constructor of `SpecificInternalRow` because the random access performance is unsatisfactory if the input argument not a `IndexSeq`.
This pr followed srowen's advice.
### Why are the changes needed?
I found that SPARK-32550 had some negative impact on performance, the typical cases is "deterministic cardinality estimation" in `HyperLogLogPlusPlusSuite` when rsd is 0.001, we found the code that is significantly slower is line 41 in `HyperLogLogPlusPlusSuite`： `new SpecificInternalRow(hll.aggBufferAttributes.map(_.dataType)) `
08b951b1cb/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/expressions/aggregate/HyperLogLogPlusPlusSuite.scala (L40-L44)
The size of "hll.aggBufferAttributes" in this case is 209716, the results of comparison before and after spark-32550 merged are as follows, The unit is ns:
| After SPARK-32550 createBuffer | After SPARK-32550 end to end | Before SPARK-32550 createBuffer | Before SPARK-32550 end to end
-- | -- | -- | -- | --
rsd 0.001, n 1000 | 52715513243 | 53004810687 | 195807999 | 773977677
rsd 0.001, n 5000 | 51881246165 | 52519358215 | 13689949 | 249974855
rsd 0.001, n 10000 | 52234282788 | 52374639172 | 14199071 | 183452846
rsd 0.001, n 50000 | 55503517122 | 55664035449 | 15219394 | 584477125
rsd 0.001, n 100000 | 51862662845 | 52116774177 | 19662834 | 166483678
rsd 0.001, n 500000 | 51619226715 | 52183189526 | 178048012 | 16681330
rsd 0.001, n 1000000 | 54861366981 | 54976399142 | 226178708 | 18826340
rsd 0.001, n 5000000 | 52023602143 | 52354615149 | 388173579 | 15446409
rsd 0.001, n 10000000 | 53008591660 | 53601392304 | 533454460 | 16033032
### Does this PR introduce _any_ user-facing change?
no
### How was this patch tested?
`mvn test -pl sql/catalyst -DwildcardSuites=org.apache.spark.sql.catalyst.expressions.aggregate.HyperLogLogPlusPlusSuite -Dtest=none`
**Before**:
```
Run completed in 8 minutes, 18 seconds.
Total number of tests run: 5
Suites: completed 2, aborted 0
Tests: succeeded 5, failed 0, canceled 0, ignored 0, pending 0
```
**After**
```
Run completed in 7 seconds, 65 milliseconds.
Total number of tests run: 5
Suites: completed 2, aborted 0
Tests: succeeded 5, failed 0, canceled 0, ignored 0, pending 0
```
Closes#29529 from LuciferYang/revert-spark-32550.
Authored-by: yangjie01 <yangjie01@baidu.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
This PR proposes to make the behavior consistent for the `path` option when loading dataframes with a single path (e.g, `option("path", path).format("parquet").load(path)` vs. `option("path", path).parquet(path)`) by disallowing `path` option to coexist with `load`'s path parameters.
### Why are the changes needed?
The current behavior is inconsistent:
```scala
scala> Seq(1).toDF.write.mode("overwrite").parquet("/tmp/test")
scala> spark.read.option("path", "/tmp/test").format("parquet").load("/tmp/test").show
+-----+
|value|
+-----+
| 1|
+-----+
scala> spark.read.option("path", "/tmp/test").parquet("/tmp/test").show
+-----+
|value|
+-----+
| 1|
| 1|
+-----+
```
### Does this PR introduce _any_ user-facing change?
Yes, now if the `path` option is specified along with `load`'s path parameters, it would fail:
```scala
scala> Seq(1).toDF.write.mode("overwrite").parquet("/tmp/test")
scala> spark.read.option("path", "/tmp/test").format("parquet").load("/tmp/test").show
org.apache.spark.sql.AnalysisException: There is a path option set and load() is called with path parameters. Either remove the path option or move it into the load() parameters.;
at org.apache.spark.sql.DataFrameReader.verifyPathOptionDoesNotExist(DataFrameReader.scala:310)
at org.apache.spark.sql.DataFrameReader.load(DataFrameReader.scala:232)
... 47 elided
scala> spark.read.option("path", "/tmp/test").parquet("/tmp/test").show
org.apache.spark.sql.AnalysisException: There is a path option set and load() is called with path parameters. Either remove the path option or move it into the load() parameters.;
at org.apache.spark.sql.DataFrameReader.verifyPathOptionDoesNotExist(DataFrameReader.scala:310)
at org.apache.spark.sql.DataFrameReader.load(DataFrameReader.scala:250)
at org.apache.spark.sql.DataFrameReader.parquet(DataFrameReader.scala:778)
at org.apache.spark.sql.DataFrameReader.parquet(DataFrameReader.scala:756)
... 47 elided
```
The user can restore the previous behavior by setting `spark.sql.legacy.pathOptionBehavior.enabled` to `true`.
### How was this patch tested?
Added a test
Closes#29328 from imback82/dfw_option.
Authored-by: Terry Kim <yuminkim@gmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
For broadcast hash join and shuffled hash join, whenever the build side hashed relation turns out to be empty. We don't need to execute stream side plan at all, and can return an empty iterator (for inner join and left semi join), because we know for sure that none of stream side rows can be outputted as there's no match.
### Why are the changes needed?
A very minor optimization for rare use case, but in case build side turns out to be empty, we can leverage it to short-cut stream side to save CPU and IO.
Example broadcast hash join query similar to `JoinBenchmark` with empty hashed relation:
```
def broadcastHashJoinLongKey(): Unit = {
val N = 20 << 20
val M = 1 << 16
val dim = broadcast(spark.range(0).selectExpr("id as k", "cast(id as string) as v"))
codegenBenchmark("Join w long", N) {
val df = spark.range(N).join(dim, (col("id") % M) === col("k"))
assert(df.queryExecution.sparkPlan.find(_.isInstanceOf[BroadcastHashJoinExec]).isDefined)
df.noop()
}
}
```
Comparing wall clock time for enabling and disabling this PR (for non-codegen code path). Seeing like 8x improvement.
```
Java HotSpot(TM) 64-Bit Server VM 1.8.0_181-b13 on Mac OS X 10.15.4
Intel(R) Core(TM) i9-9980HK CPU 2.40GHz
Join w long: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
Join PR disabled 637 646 12 32.9 30.4 1.0X
Join PR enabled 77 78 2 271.8 3.7 8.3X
```
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Added unit test in `JoinSuite`.
Closes#29484 from c21/empty-relation.
Authored-by: Cheng Su <chengsu@fb.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
This PR fixes a bug of #29406. #29406 partially pushes down data filter even if it mixed in partition filters. But in some cases partition columns might be in data columns too. It will possibly push down a predicate with partition column to datasource.
### Why are the changes needed?
The test "org.apache.spark.sql.hive.orc.HiveOrcHadoopFsRelationSuite.save()/load() - partitioned table - simple queries - partition columns in data" is currently failed with hive-1.2 profile in master branch.
```
[info] - save()/load() - partitioned table - simple queries - partition columns in data *** FAILED *** (1 second, 457 milliseconds)
[info] java.util.NoSuchElementException: key not found: p1
[info] at scala.collection.immutable.Map$Map2.apply(Map.scala:138)
[info] at org.apache.spark.sql.hive.orc.OrcFilters$.buildLeafSearchArgument(OrcFilters.scala:250)
[info] at org.apache.spark.sql.hive.orc.OrcFilters$.convertibleFiltersHelper$1(OrcFilters.scala:143)
[info] at org.apache.spark.sql.hive.orc.OrcFilters$.$anonfun$convertibleFilters$4(OrcFilters.scala:146)
[info] at scala.collection.TraversableLike.$anonfun$flatMap$1(TraversableLike.scala:245)
[info] at scala.collection.IndexedSeqOptimized.foreach(IndexedSeqOptimized.scala:36)
[info] at scala.collection.IndexedSeqOptimized.foreach$(IndexedSeqOptimized.scala:33)
[info] at scala.collection.mutable.WrappedArray.foreach(WrappedArray.scala:38)
[info] at scala.collection.TraversableLike.flatMap(TraversableLike.scala:245)
[info] at scala.collection.TraversableLike.flatMap$(TraversableLike.scala:242)
[info] at scala.collection.AbstractTraversable.flatMap(Traversable.scala:108)
[info] at org.apache.spark.sql.hive.orc.OrcFilters$.convertibleFilters(OrcFilters.scala:145)
[info] at org.apache.spark.sql.hive.orc.OrcFilters$.createFilter(OrcFilters.scala:83)
[info] at org.apache.spark.sql.hive.orc.OrcFileFormat.buildReader(OrcFileFormat.scala:142)
```
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Unit test.
Closes#29526 from viirya/SPARK-32352-followup.
Authored-by: Liang-Chi Hsieh <viirya@gmail.com>
Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
### What changes were proposed in this pull request?
Copy to master branch the unit test added for branch-2.4(https://github.com/apache/spark/pull/29430).
### Why are the changes needed?
The unit test will pass at master branch, indicating that issue reported in https://issues.apache.org/jira/browse/SPARK-32609 is already fixed at master branch. But adding this unit test for future possible failure catch.
### Does this PR introduce _any_ user-facing change?
no.
### How was this patch tested?
sbt test run
Closes#29435 from mingjialiu/master.
Authored-by: mingjial <mingjial@google.com>
Signed-off-by: Dongjoon Hyun <dongjoon@apache.org>
### What changes were proposed in this pull request?
Removing unused DELETE_ACTION in FileStreamSinkLog.
### Why are the changes needed?
DELETE_ACTION is not used nowhere in the code.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
Tests where not added, because code was removed.
Closes#29505 from michal-wieleba/SPARK-32648.
Authored-by: majsiel <majsiel@gmail.com>
Signed-off-by: Dongjoon Hyun <dongjoon@apache.org>
### What changes were proposed in this pull request?
Override `def get: Date` in `DaysWritable` use the `daysToMillis(int d)` from the parent class `DateWritable` instead of `long daysToMillis(int d, boolean doesTimeMatter)`.
### Why are the changes needed?
It fixes failures of `HiveSerDeReadWriteSuite` with the profile `hive-1.2`. In that case, the parent class `DateWritable` has different implementation before the commit to Hive da3ed68eda. In particular, `get()` calls `new Date(daysToMillis(daysSinceEpoch))` instead of overrided `def get(doesTimeMatter: Boolean): Date` in the child class. The `get()` method returns wrong result `1970-01-01` because it uses not updated `daysSinceEpoch`.
### Does this PR introduce _any_ user-facing change?
Yes.
### How was this patch tested?
By running the test suite `HiveSerDeReadWriteSuite`:
```
$ build/sbt -Phive-1.2 -Phadoop-2.7 "test:testOnly org.apache.spark.sql.hive.execution.HiveSerDeReadWriteSuite"
```
and
```
$ build/sbt -Phive-2.3 -Phadoop-2.7 "test:testOnly org.apache.spark.sql.hive.execution.HiveSerDeReadWriteSuite"
```
Closes#29523 from MaxGekk/insert-date-into-hive-table-1.2.
Authored-by: Max Gekk <max.gekk@gmail.com>
Signed-off-by: Liang-Chi Hsieh <viirya@gmail.com>
### What changes were proposed in this pull request?
As mentioned in https://github.com/apache/spark/pull/29428#issuecomment-678735163 by viirya ,
fix bug in UT, since in script transformation no-serde mode, output of decimal is same in both hive-1.2/hive-2.3
### Why are the changes needed?
FIX UT
### Does this PR introduce _any_ user-facing change?
NO
### How was this patch tested?
EXISTED UT
Closes#29520 from AngersZhuuuu/SPARK-32608-FOLLOW.
Authored-by: angerszhu <angers.zhu@gmail.com>
Signed-off-by: Liang-Chi Hsieh <viirya@gmail.com>
### What changes were proposed in this pull request?
This reverts commit e277ef1a83.
### Why are the changes needed?
Because master and branch-3.0 both have few tests failed under hive-1.2 profile. And the PR #29457 missed a change in hive-1.2 code that causes compilation error. So it will make debugging the failed tests harder. I'd like revert #29457 first to unblock it.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
Unit test
Closes#29519 from viirya/revert-SPARK-32646.
Authored-by: Liang-Chi Hsieh <viirya@gmail.com>
Signed-off-by: Liang-Chi Hsieh <viirya@gmail.com>
### What changes were proposed in this pull request?
Some Code refine.
1. rename EmptyHashedRelationWithAllNullKeys to HashedRelationWithAllNullKeys.
2. simplify generated code for BHJ NAAJ.
### Why are the changes needed?
Refine code and naming to avoid confusing understanding.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Existing test.
Closes#29503 from leanken/leanken-SPARK-32678.
Authored-by: xuewei.linxuewei <xuewei.linxuewei@alibaba-inc.com>
Signed-off-by: Liang-Chi Hsieh <viirya@gmail.com>
### What changes were proposed in this pull request?
The purpose of this pr is to resolve [SPARK-32526](https://issues.apache.org/jira/browse/SPARK-32526), all remaining failed cases are fixed.
The main change of this pr as follow:
- Change of `ExecutorAllocationManager.scala` for core module compilation in Scala 2.13, it's a blocking problem
- Change `Seq[_]` to `scala.collection.Seq[_]` refer to failed cases
- Added different expected plan of `Test 4: Star with several branches` of StarJoinCostBasedReorderSuite for Scala 2.13 because the candidates plans:
```
Join Inner, (d1_pk#5 = f1_fk1#0)
:- Join Inner, (f1_fk2#1 = d2_pk#8)
: :- Join Inner, (f1_fk3#2 = d3_pk#11)
```
and
```
Join Inner, (f1_fk2#1 = d2_pk#8)
:- Join Inner, (d1_pk#5 = f1_fk1#0)
: :- Join Inner, (f1_fk3#2 = d3_pk#11)
```
have same cost `Cost(200,9200)`, but `HashMap` is rewritten in scala 2.13 and The order of iterations leads to different results.
This pr fix test cases as follow:
- LiteralExpressionSuite (1 FAILED -> PASS)
- StarJoinCostBasedReorderSuite ( 1 FAILED-> PASS)
- ObjectExpressionsSuite( 2 FAILED-> PASS)
- ScalaReflectionSuite （1 FAILED-> PASS）
- RowEncoderSuite (10 FAILED-> PASS)
- ExpressionEncoderSuite (ABORTED-> PASS)
### Why are the changes needed?
We need to support a Scala 2.13 build.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
<!--
- Scala 2.12: Pass the Jenkins or GitHub Action
- Scala 2.13: Do the following:
```
dev/change-scala-version.sh 2.13
mvn clean install -DskipTests -pl sql/catalyst -Pscala-2.13 -am
mvn test -pl sql/catalyst -Pscala-2.13
```
**Before**
```
Tests: succeeded 4035, failed 17, canceled 0, ignored 6, pending 0
*** 1 SUITE ABORTED ***
*** 15 TESTS FAILED ***
```
**After**
```
Tests: succeeded 4338, failed 0, canceled 0, ignored 6, pending 0
All tests passed.
```
Closes#29434 from LuciferYang/sql-catalyst-tests.
Authored-by: yangjie01 <yangjie01@baidu.com>
Signed-off-by: Sean Owen <srowen@gmail.com>
### What changes were proposed in this pull request?
Rephrase the description for some operations to make it clearer.
### Why are the changes needed?
Add more detail in the document.
### Does this PR introduce _any_ user-facing change?
No, document only.
### How was this patch tested?
Document only.
Closes#29269 from xuanyuanking/SPARK-31792-follow.
Authored-by: Yuanjian Li <yuanjian.li@databricks.com>
Signed-off-by: Jungtaek Lim (HeartSaVioR) <kabhwan.opensource@gmail.com>
## What changes were proposed in this pull request?
This fixed SPARK-32672 a data corruption. Essentially the BooleanBitSet CompressionScheme would miss nulls at the end of a CompressedBatch. The values would then default to false.
### Why are the changes needed?
It fixes data corruption
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
I manually tested it against the original issue that was producing errors for me. I also added in a unit test.
Closes#29506 from revans2/SPARK-32672.
Authored-by: Robert (Bobby) Evans <bobby@apache.org>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
Fix typo for docs, log messages and comments
### Why are the changes needed?
typo fix to increase readability
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
manual test has been performed to test the updated
Closes#29443 from brandonJY/spell-fix-doc.
Authored-by: Brandon Jiang <Brandon.jiang.a@outlook.com>
Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
### What changes were proposed in this pull request?
Add document to `ExpressionEvalHelper`, and ask people to explore all the cases that can lead to null results (including null in struct fields, array elements and map values).
This PR also fixes `ComplexTypeSuite.GetArrayStructFields` to explore all the null cases.
### Why are the changes needed?
It happened several times that we hit correctness bugs caused by wrong expression nullability. When writing unit tests, we usually don't test the nullability flag directly, and it's too late to add such tests for all expressions.
In https://github.com/apache/spark/pull/22375, we extended the expression test framework, which checks the nullability flag when the expected result/field/element is null.
This requires the test cases to explore all the cases that can lead to null results
### Does this PR introduce _any_ user-facing change?
no
### How was this patch tested?
I reverted 5d296ed39e locally, and `ComplexTypeSuite` can catch the bug.
Closes#29493 from cloud-fan/small.
Authored-by: Wenchen Fan <wenchen@databricks.com>
Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
### What changes were proposed in this pull request?
This PR proposes to fix ORC predicate pushdown under case-insensitive analysis case. The field names in pushed down predicates don't need to match in exact letter case with physical field names in ORC files, if we enable case-insensitive analysis.
### Why are the changes needed?
Currently ORC predicate pushdown doesn't work with case-insensitive analysis. A predicate "a < 0" cannot pushdown to ORC file with field name "A" under case-insensitive analysis.
But Parquet predicate pushdown works with this case. We should make ORC predicate pushdown work with case-insensitive analysis too.
### Does this PR introduce _any_ user-facing change?
Yes, after this PR, under case-insensitive analysis, ORC predicate pushdown will work.
### How was this patch tested?
Unit tests.
Closes#29457 from viirya/fix-orc-pushdown.
Authored-by: Liang-Chi Hsieh <viirya@gmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
Hive no serde mode when column less then output specified column, it will pad null value to it, spark should do this also.
```
hive> SELECT TRANSFORM(a, b)
> ROW FORMAT DELIMITED
> FIELDS TERMINATED BY '|'
> LINES TERMINATED BY '\n'
> NULL DEFINED AS 'NULL'
> USING 'cat' as (a string, b string, c string, d string)
> ROW FORMAT DELIMITED
> FIELDS TERMINATED BY '|'
> LINES TERMINATED BY '\n'
> NULL DEFINED AS 'NULL'
> FROM (
> select 1 as a, 2 as b
> ) tmp ;
OK
1 2 NULL NULL
Time taken: 24.626 seconds, Fetched: 1 row(s)
```
### Why are the changes needed?
Keep save behavior with hive data.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
Added UT
Closes#29500 from AngersZhuuuu/SPARK-32667.
Authored-by: angerszhu <angers.zhu@gmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
This PR reverts https://github.com/apache/spark/pull/27860 to downgrade Janino, as the new version has a bug.
### Why are the changes needed?
The symptom is about NaN comparison. For code below
```
if (double_value <= 0.0) {
...
} else {
...
}
```
If `double_value` is NaN, `NaN <= 0.0` is false and we should go to the else branch. However, current Spark goes to the if branch and causes correctness issues like SPARK-32640.
One way to fix it is:
```
boolean cond = double_value <= 0.0;
if (cond) {
...
} else {
...
}
```
I'm not familiar with Janino so I don't know what's going on there.
### Does this PR introduce _any_ user-facing change?
Yes, fix correctness bugs.
### How was this patch tested?
a new test
Closes#29495 from cloud-fan/revert.
Authored-by: Wenchen Fan <wenchen@databricks.com>
Signed-off-by: Dongjoon Hyun <dongjoon@apache.org>
### What changes were proposed in this pull request?
This is a followup of https://github.com/apache/spark/pull/29469
Instead of passing the physical plan to the fallbacked v1 source directly and skipping analysis, optimization, planning altogether, this PR proposes to pass the optimized plan.
### Why are the changes needed?
It's a bit risky to pass the physical plan directly. When the fallbacked v1 source applies more operations to the input DataFrame, it will re-apply the post-planning physical rules like `CollapseCodegenStages`, `InsertAdaptiveSparkPlan`, etc., which is very tricky.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
existing test suite with some new tests
Closes#29489 from cloud-fan/follow.
Authored-by: Wenchen Fan <wenchen@databricks.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
In case that a last `SQLQueryTestSuite` test run is killed, it will fail in a next run because of a following reason:
```
[info] org.apache.spark.sql.SQLQueryTestSuite *** ABORTED *** (17 seconds, 483 milliseconds)
[info] org.apache.spark.sql.AnalysisException: Can not create the managed table('`testdata`'). The associated location('file:/Users/maropu/Repositories/spark/spark-master/sql/core/spark-warehouse/org.apache.spark.sql.SQLQueryTestSuite/testdata') already exists.;
[info] at org.apache.spark.sql.catalyst.catalog.SessionCatalog.validateTableLocation(SessionCatalog.scala:355)
[info] at org.apache.spark.sql.execution.command.CreateDataSourceTableAsSelectCommand.run(createDataSourceTables.scala:170)
[info] at org.apache.spark.sql.execution.command.DataWritingCommandExec.sideEffectResult$lzycompute(commands.scala:108)
```
This PR intends to add code to deletes orphan directories under a warehouse dir in `SQLQueryTestSuite` before creating test tables.
### Why are the changes needed?
To improve test convenience
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
Manually checked
Closes#29488 from maropu/DeleteDirs.
Authored-by: Takeshi Yamamuro <yamamuro@apache.org>
Signed-off-by: Dongjoon Hyun <dongjoon@apache.org>
### What changes were proposed in this pull request?
Scrip Transform no-serde (`ROW FORMAT DELIMITED`) mode `LINE TERMINNATED BY `
only support `\n`.
Tested in hive :
Hive 1.1
Hive 2.3.7
### Why are the changes needed?
Strictly limit the use method to ensure the accuracy of data
### Does this PR introduce _any_ user-facing change?
User use Scrip Transform no-serde (ROW FORMAT DELIMITED) mode with `LINE TERMINNATED BY `
not equal `'\n'`. will throw error
### How was this patch tested?
Added UT
Closes#29438 from AngersZhuuuu/SPARK-32607.
Authored-by: angerszhu <angers.zhu@gmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
In many operations on CompactibleFileStreamLog reads a metadata log file and materializes all entries into memory. As the nature of the compact operation, CompactibleFileStreamLog may have a huge compact log file with bunch of entries included, and for now they're just monotonically increasing, which means the amount of memory to materialize also grows incrementally. This leads pressure on GC.
This patch proposes to streamline the logic on file stream source and sink whenever possible to avoid memory issue. To make this possible we have to break the existing behavior of excluding entries - now the `compactLogs` method is called with all entries, which forces us to materialize all entries into memory. This is hopefully no effect on end users, because only file stream sink has a condition to exclude entries, and the condition has been never true. (DELETE_ACTION has been never set.)
Based on the observation, this patch also changes the existing UT a bit which simulates the situation where "A" file is added, and another batch marks the "A" file as deleted. This situation simply doesn't work with the change, but as I mentioned earlier it hasn't been used. (I'm not sure the UT is from the actual run. I guess not.)
### Why are the changes needed?
The memory issue (OOME) is reported by both JIRA issue and user mailing list.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
* Existing UTs
* Manual test done
The manual test leverages the simple apps which continuously writes the file stream sink metadata log.
bea7680e4c
The test is configured to have a batch metadata log file at 1.9M (10,000 entries) whereas other Spark configuration is set to the default. (compact interval = 10) The app runs as driver, and the heap memory on driver is set to 3g.
> before the patch
<img width="1094" alt="Screen Shot 2020-06-23 at 3 37 44 PM" src="https://user-images.githubusercontent.com/1317309/85375841-d94f3480-b571-11ea-817b-c6b48b34888a.png">
It only ran for 40 mins, with the latest compact batch file size as 1.3G. The process struggled with GC, and after some struggling, it threw OOME.
> after the patch
<img width="1094" alt="Screen Shot 2020-06-23 at 3 53 29 PM" src="https://user-images.githubusercontent.com/1317309/85375901-eff58b80-b571-11ea-837e-30d107f677f9.png">
It sustained 2 hours run (manually stopped as it's expected to run more), with the latest compact batch file size as 2.2G. The actual memory usage didn't even go up to 1.2G, and be cleaned up soon without outstanding GC activity.
Closes#28904 from HeartSaVioR/SPARK-30462.
Authored-by: Jungtaek Lim (HeartSaVioR) <kabhwan.opensource@gmail.com>
Signed-off-by: Dongjoon Hyun <dongjoon@apache.org>
### What changes were proposed in this pull request?
Address the [#comment](https://github.com/apache/spark/pull/28840#discussion_r471172006).
### Why are the changes needed?
Make code robust.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
ut.
Closes#29453 from ulysses-you/SPARK-31999-FOLLOWUP.
Authored-by: ulysses <youxiduo@weidian.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
This PR introduces a LogicalNode AlreadyPlanned, and related physical plan and preparation rule.
With the DataSourceV2 write operations, we have a way to fallback to the V1 writer APIs using InsertableRelation. The gross part is that we're in physical land, but the InsertableRelation takes a logical plan, so we have to pass the logical plans to these physical nodes, and then potentially go through re-planning. This re-planning can cause issues for an already optimized plan.
A useful primitive could be specifying that a plan is ready for execution through a logical node AlreadyPlanned. This would wrap a physical plan, and then we can go straight to execution.
### Why are the changes needed?
To avoid having a physical plan that is disconnected from the physical plan that is being executed in V1WriteFallback execution. When a physical plan node executes a logical plan, the inner query is not connected to the running physical plan. The physical plan that actually runs is not visible through the Spark UI and its metrics are not exposed. In some cases, the EXPLAIN plan doesn't show it.
### Does this PR introduce _any_ user-facing change?
Nope
### How was this patch tested?
V1FallbackWriterSuite tests that writes still work
Closes#29469 from brkyvz/alreadyAnalyzed2.
Lead-authored-by: Burak Yavuz <brkyvz@gmail.com>
Co-authored-by: Burak Yavuz <burak@databricks.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
When CSV/JSON datasources infer schema (e.g, `def inferSchema(files: Seq[FileStatus])`, they use the `files` along with the original options. `files` in `inferSchema` could have been deduced from the "path" option if the option was present, so this can cause issues (e.g., reading more data, listing the path again) since the "path" option is **added** to the `files`.
### Why are the changes needed?
The current behavior can cause the following issue:
```scala
class TestFileFilter extends PathFilter {
override def accept(path: Path): Boolean = path.getParent.getName != "p=2"
}
val path = "/tmp"
val df = spark.range(2)
df.write.json(path + "/p=1")
df.write.json(path + "/p=2")
val extraOptions = Map(
"mapred.input.pathFilter.class" -> classOf[TestFileFilter].getName,
"mapreduce.input.pathFilter.class" -> classOf[TestFileFilter].getName
)
// This works fine.
assert(spark.read.options(extraOptions).json(path).count == 2)
// The following with "path" option fails with the following:
// assertion failed: Conflicting directory structures detected. Suspicious paths
// file:/tmp
// file:/tmp/p=1
assert(spark.read.options(extraOptions).format("json").option("path", path).load.count() === 2)
```
### Does this PR introduce _any_ user-facing change?
Yes, the above failure doesn't happen and you get the consistent experience when you use `spark.read.csv(path)` or `spark.read.format("csv").option("path", path).load`.
### How was this patch tested?
Updated existing tests.
Closes#29437 from imback82/path_bug.
Authored-by: Terry Kim <yuminkim@gmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
For SQL
```
SELECT TRANSFORM(a, b, c)
ROW FORMAT DELIMITED
FIELDS TERMINATED BY ','
LINES TERMINATED BY '\n'
NULL DEFINED AS 'null'
USING 'cat' AS (a, b, c)
ROW FORMAT DELIMITED
FIELDS TERMINATED BY ','
LINES TERMINATED BY '\n'
NULL DEFINED AS 'NULL'
FROM testData
```
The correct
TOK_TABLEROWFORMATFIELD should be `, `nut actually ` ','`
TOK_TABLEROWFORMATLINES should be `\n` but actually` '\n'`
### Why are the changes needed?
Fix string value format
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
Added UT
Closes#29428 from AngersZhuuuu/SPARK-32608.
Authored-by: angerszhu <angers.zhu@gmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
Change maps in two constructors of SpecificInternalRow to while loops.
### Why are the changes needed?
This was originally noticed with https://github.com/apache/spark/pull/29353 and https://github.com/apache/spark/pull/29354 and will have impacts on performance of reading ORC and Avro files. Ran AvroReadBenchmarks with the new cases of nested and array'd structs in https://github.com/apache/spark/pull/29352. Haven't run benchmarks for ORC but can do that if needed.
**Before:**
```
Nested Struct Scan: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
Nested Struct 74674 75319 912 0.0 142429.1 1.0X
Array of Struct Scan: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
Array of Structs 34193 34339 206 0.0 65217.9 1.0X
```
**After:**
```
Nested Struct Scan: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
Nested Struct 48451 48619 237 0.0 92413.2 1.0X
Array of Struct Scan: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
Array of Structs 18518 18683 234 0.0 35319.6 1.0X
```
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
Ran AvroReadBenchmarks with the new cases of nested and array'd structs in https://github.com/apache/spark/pull/29352.
Closes#29366 from msamirkhan/spark-32550.
Lead-authored-by: Samir Khan <muhammad.samir.khan@gmail.com>
Co-authored-by: skhan04 <samirkhan@verizonmedia.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
Update `ObjectSerializerPruning.alignNullTypeInIf`, to consider the isNull check generated in `RowEncoder`, which is `Invoke(inputObject, "isNullAt", BooleanType, Literal(index) :: Nil)`.
### Why are the changes needed?
Query fails if we don't fix this bug, due to type mismatch in `If`.
### Does this PR introduce _any_ user-facing change?
Yes, the failed query can run after this fix.
### How was this patch tested?
new tests
Closes#29467 from cloud-fan/bug.
Authored-by: Wenchen Fan <wenchen@databricks.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
During working on SPARK-25557, we found that ORC predicate pushdown doesn't have case-sensitivity test. This PR proposes to add case-sensitivity test for ORC predicate pushdown.
### Why are the changes needed?
Increasing test coverage for ORC predicate pushdown.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
Pass Jenkins tests.
Closes#29427 from viirya/SPARK-25557-followup3.
Authored-by: Liang-Chi Hsieh <viirya@gmail.com>
Signed-off-by: Dongjoon Hyun <dongjoon@apache.org>
### What changes were proposed in this pull request?
Found java.util.NoSuchElementException in UT log of AdaptiveQueryExecSuite. During AQE, when sub-plan changed, LiveExecutionData is using the new sub-plan SQLMetrics to override the old ones, But in the final aggregateMetrics, since the plan was updated, the old metrics will throw NoSuchElementException when it try to match with the new metricTypes. To sum up, we need to filter out those outdated metrics to avoid throwing java.util.NoSuchElementException, which cause SparkUI SQL Tab abnormally rendered.
### Why are the changes needed?
SQL Metrics is not correct for some AQE cases, and it break SparkUI SQL Tab when it comes to NAAJ rewritten to LocalRelation case.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
* Added case in SQLAppStatusListenerSuite.
* Run AdaptiveQueryExecSuite with no "java.util.NoSuchElementException".
* Validation on Spark Web UI
Closes#29431 from leanken/leanken-SPARK-32615.
Authored-by: xuewei.linxuewei <xuewei.linxuewei@alibaba-inc.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
This PR proposes to detect possible regression inside `SparkPlan`. To achieve this goal, this PR added a base test suite called `PlanStabilitySuite`. The basic workflow of this test suite is similar to `SQLQueryTestSuite`. It also uses `SPARK_GENERATE_GOLDEN_FILES` to decide whether it should regenerate the golden files or compare to the golden result for each input query. The difference is, `PlanStabilitySuite` uses the serialized explain result(.txt format) of the `SparkPlan` as the output of a query, instead of the data result.
And since `SparkPlan` is non-deterministic for various reasons, e.g., expressions ids changes, expression order changes, we'd reduce the plan to a simplified version that only contains node names and references. And we only identify those important nodes, e.g., `Exchange`, `SubqueryExec`, in the simplified plan.
And we'd reuse TPC-DS queries(v1.4, v2.7, modified) to test plans' stability. Currently, one TPC-DS query can only have one corresponding simplified golden plan.
This PR also did a few refactor, which extracts `TPCDSBase` from `TPCDSQuerySuite`. So, `PlanStabilitySuite` can use the TPC-DS queries as well.
### Why are the changes needed?
Nowadays, Spark is getting more and more complex. Any changes might cause regression unintentionally. Spark already has some benchmark to catch the performance regression. But, yet, it doesn't have a way to detect the regression inside `SparkPlan`. It would be good if we could detect the possible regression early during the compile phase before the runtime phase.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Added `PlanStabilitySuite` and it's subclasses.
Closes#29270 from Ngone51/plan-stable.
Authored-by: yi.wu <yi.wu@databricks.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
Add support for full outer join inside shuffled hash join. Currently if the query is a full outer join, we only use sort merge join as the physical operator. However it can be CPU and IO intensive in case input table is large for sort merge join. Shuffled hash join on the other hand saves the sort CPU and IO compared to sort merge join, especially when table is large.
This PR implements the full outer join as followed:
* Process rows from stream side by looking up hash relation, and mark the matched rows from build side by:
* for joining with unique key, a `BitSet` is used to record matched rows from build side (`key index` to represent each row)
* for joining with non-unique key, a `HashSet[Long]` is used to record matched rows from build side (`key index` + `value index` to represent each row).
`key index` is defined as the index into key addressing array `longArray` in `BytesToBytesMap`.
`value index` is defined as the iterator index of values for same key.
* Process rows from build side by iterating hash relation, and filter out rows from build side being looked up already (done in `ShuffledHashJoinExec.fullOuterJoin`)
For context, this PR was originally implemented as followed (up to commit e3322766d4):
1. Construct hash relation from build side, with extra boolean value at the end of row to track look up information (done in `ShuffledHashJoinExec.buildHashedRelation` and `UnsafeHashedRelation.apply`).
2. Process rows from stream side by looking up hash relation, and mark the matched rows from build side be looked up (done in `ShuffledHashJoinExec.fullOuterJoin`).
3. Process rows from build side by iterating hash relation, and filter out rows from build side being looked up already (done in `ShuffledHashJoinExec.fullOuterJoin`).
See discussion of pros and cons between these two approaches [here](https://github.com/apache/spark/pull/29342#issuecomment-672275450), [here](https://github.com/apache/spark/pull/29342#issuecomment-672288194) and [here](https://github.com/apache/spark/pull/29342#issuecomment-672640531).
TODO: codegen for full outer shuffled hash join can be implemented in another followup PR.
### Why are the changes needed?
As implementation in this PR, full outer shuffled hash join will have overhead to iterate build side twice (once for building hash map, and another for outputting non-matching rows), and iterate stream side once. However, full outer sort merge join needs to iterate both sides twice, and sort the large table can be more CPU and IO intensive. So full outer shuffled hash join can be more efficient than sort merge join when stream side is much more larger than build side.
For example query below, full outer SHJ saved 30% wall clock time compared to full outer SMJ.
```
def shuffleHashJoin(): Unit = {
val N: Long = 4 << 22
withSQLConf(
SQLConf.SHUFFLE_PARTITIONS.key -> "2",
SQLConf.AUTO_BROADCASTJOIN_THRESHOLD.key -> "20000000") {
codegenBenchmark("shuffle hash join", N) {
val df1 = spark.range(N).selectExpr(s"cast(id as string) as k1")
val df2 = spark.range(N / 10).selectExpr(s"cast(id * 10 as string) as k2")
val df = df1.join(df2, col("k1") === col("k2"), "full_outer")
df.noop()
}
}
}
```
```
Running benchmark: shuffle hash join
Running case: shuffle hash join off
Stopped after 2 iterations, 16602 ms
Running case: shuffle hash join on
Stopped after 5 iterations, 31911 ms
Java HotSpot(TM) 64-Bit Server VM 1.8.0_181-b13 on Mac OS X 10.15.4
Intel(R) Core(TM) i9-9980HK CPU 2.40GHz
shuffle hash join: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
shuffle hash join off 7900 8301 567 2.1 470.9 1.0X
shuffle hash join on 6250 6382 95 2.7 372.5 1.3X
```
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Added unit test in `JoinSuite.scala`, `AbstractBytesToBytesMapSuite.java` and `HashedRelationSuite.scala`.
Closes#29342 from c21/full-outer-shj.
Authored-by: Cheng Su <chengsu@fb.com>
Signed-off-by: Takeshi Yamamuro <yamamuro@apache.org>
