## What changes were proposed in this pull request?
The leftover state from running a continuous processing streaming job should not affect later microbatch execution jobs. If a continuous processing job runs and the same thread gets reused for a microbatch execution job in the same environment, the microbatch job could get wrong answers because it can attempt to load the wrong version of the state.
## How was this patch tested?
New and existing unit tests
Closes#22386 from mukulmurthy/25399-streamthread.
Authored-by: Mukul Murthy <mukul.murthy@gmail.com>
Signed-off-by: Tathagata Das <tathagata.das1565@gmail.com>
## What changes were proposed in this pull request?
Correct some comparisons between unrelated types to what they seem to… have been trying to do
## How was this patch tested?
Existing tests.
Closes#22384 from srowen/SPARK-25398.
Authored-by: Sean Owen <sean.owen@databricks.com>
Signed-off-by: Sean Owen <sean.owen@databricks.com>
## What changes were proposed in this pull request?
In the PR, I propose new CSV option `emptyValue` and an update in the SQL Migration Guide which describes how to revert previous behavior when empty strings were not written at all. Since Spark 2.4, empty strings are saved as `""` to distinguish them from saved `null`s.
Closes#22234Closes#22367
## How was this patch tested?
It was tested by `CSVSuite` and new tests added in the PR #22234Closes#22389 from MaxGekk/csv-empty-value-master.
Lead-authored-by: Mario Molina <mmolimar@gmail.com>
Co-authored-by: Maxim Gekk <maxim.gekk@databricks.com>
Signed-off-by: hyukjinkwon <gurwls223@apache.org>
## What changes were proposed in this pull request?
It turns out it's a bug that a `DataSourceV2ScanExec` instance may be referred to in the execution plan multiple times. This bug is fixed by https://github.com/apache/spark/pull/22284 and now we have corrected SQL metrics for batch queries.
Thus we don't need the hack in `ProgressReporter` anymore, which fixes the same metrics problem for streaming queries.
## How was this patch tested?
existing tests
Closes#22380 from cloud-fan/followup.
Authored-by: Wenchen Fan <wenchen@databricks.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
## What changes were proposed in this pull request?
SPARK-21281 introduced a check for the inputs of `CreateStructLike` to be non-empty. This means that `struct()`, which was previously considered valid, now throws an Exception. This behavior change was introduced in 2.3.0. The change may break users' application on upgrade and it causes `VectorAssembler` to fail when an empty `inputCols` is defined.
The PR removes the added check making `struct()` valid again.
## How was this patch tested?
added UT
Closes#22373 from mgaido91/SPARK-25371.
Authored-by: Marco Gaido <marcogaido91@gmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
## What changes were proposed in this pull request?
In the Planner, we collect the placeholder which need to be substituted in the query execution plan and once we plan them, we substitute the placeholder with the effective plan.
In this second phase, we rely on the `==` comparison, ie. the `equals` method. This means that if two placeholder plans - which are different instances - have the same attributes (so that they are equal, according to the equal method) they are both substituted with their corresponding new physical plans. So, in such a situation, the first time we substitute both them with the first of the 2 new generated plan and the second time we substitute nothing.
This is usually of no harm for the execution of the query itself, as the 2 plans are identical. But since they are the same instance, now, the local variables are shared (which is unexpected). This causes issues for the metrics collected, as the same node is executed 2 times, so the metrics are accumulated 2 times, wrongly.
The PR proposes to use the `eq` method in checking which placeholder needs to be substituted,; thus in the previous situation, actually both the two different physical nodes which are created (one for each time the logical plan appears in the query plan) are used and the metrics are collected properly for each of them.
## How was this patch tested?
added UT
Closes#22284 from mgaido91/SPARK-25278.
Authored-by: Marco Gaido <marcogaido91@gmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
## What changes were proposed in this pull request?
This PR is to solve the CodeGen code generated by fast hash, and there is no need to apply for a block of memory for every new entry, because unsafeRow's memory can be reused.
## How was this patch tested?
the existed test cases.
Closes#21968 from heary-cao/updateNewMemory.
Authored-by: caoxuewen <cao.xuewen@zte.com.cn>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
## What changes were proposed in this pull request?
How to reproduce:
```scala
spark.sql("CREATE TABLE tbl(id long)")
spark.sql("INSERT OVERWRITE TABLE tbl VALUES 4")
spark.sql("CREATE VIEW view1 AS SELECT id FROM tbl")
spark.sql(s"INSERT OVERWRITE LOCAL DIRECTORY '/tmp/spark/parquet' " +
"STORED AS PARQUET SELECT ID FROM view1")
spark.read.parquet("/tmp/spark/parquet").schema
scala> spark.read.parquet("/tmp/spark/parquet").schema
res10: org.apache.spark.sql.types.StructType = StructType(StructField(id,LongType,true))
```
The schema should be `StructType(StructField(ID,LongType,true))` as we `SELECT ID FROM view1`.
This pr fix this issue.
## How was this patch tested?
unit tests
Closes#22359 from wangyum/SPARK-25313-FOLLOW-UP.
Authored-by: Yuming Wang <yumwang@ebay.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
## What changes were proposed in this pull request?
Apache Spark doesn't create Hive table with duplicated fields in both case-sensitive and case-insensitive mode. However, if Spark creates ORC files in case-sensitive mode first and create Hive table on that location, where it's created. In this situation, field resolution should fail in case-insensitive mode. Otherwise, we don't know which columns will be returned or filtered. Previously, SPARK-25132 fixed the same issue in Parquet.
Here is a simple example:
```
val data = spark.range(5).selectExpr("id as a", "id * 2 as A")
spark.conf.set("spark.sql.caseSensitive", true)
data.write.format("orc").mode("overwrite").save("/user/hive/warehouse/orc_data")
sql("CREATE TABLE orc_data_source (A LONG) USING orc LOCATION '/user/hive/warehouse/orc_data'")
spark.conf.set("spark.sql.caseSensitive", false)
sql("select A from orc_data_source").show
+---+
| A|
+---+
| 3|
| 2|
| 4|
| 1|
| 0|
+---+
```
See #22148 for more details about parquet data source reader.
## How was this patch tested?
Unit tests added.
Closes#22262 from seancxmao/SPARK-25175.
Authored-by: seancxmao <seancxmao@gmail.com>
Signed-off-by: Dongjoon Hyun <dongjoon@apache.org>
## What changes were proposed in this pull request?
How to reproduce:
```scala
val df1 = spark.createDataFrame(Seq(
(1, 1)
)).toDF("a", "b").withColumn("c", lit(null).cast("int"))
val df2 = df1.union(df1).withColumn("d", spark_partition_id).filter($"c".isNotNull)
df2.show
+---+---+----+---+
| a| b| c| d|
+---+---+----+---+
| 1| 1|null| 0|
| 1| 1|null| 1|
+---+---+----+---+
```
`filter($"c".isNotNull)` was transformed to `(null <=> c#10)` before https://github.com/apache/spark/pull/19201, but it is transformed to `(c#10 = null)` since https://github.com/apache/spark/pull/20155. This pr revert it to `(null <=> c#10)` to fix this issue.
## How was this patch tested?
unit tests
Closes#22368 from wangyum/SPARK-25368.
Authored-by: Yuming Wang <yumwang@ebay.com>
Signed-off-by: gatorsmile <gatorsmile@gmail.com>
## What changes were proposed in this pull request?
When running TPC-DS benchmarks on 2.4 release, npoggi and winglungngai saw more than 10% performance regression on the following queries: q67, q24a and q24b. After we applying the PR https://github.com/apache/spark/pull/22338, the performance regression still exists. If we revert the changes in https://github.com/apache/spark/pull/19222, npoggi and winglungngai found the performance regression was resolved. Thus, this PR is to revert the related changes for unblocking the 2.4 release.
In the future release, we still can continue the investigation and find out the root cause of the regression.
## How was this patch tested?
The existing test cases
Closes#22361 from gatorsmile/revertMemoryBlock.
Authored-by: gatorsmile <gatorsmile@gmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
## What changes were proposed in this pull request?
Add new optimization rule to eliminate unnecessary shuffling by flipping adjacent Window expressions.
## How was this patch tested?
Tested with unit tests, integration tests, and manual tests.
Closes#17899 from ptkool/adjacent_window_optimization.
Authored-by: ptkool <michael.styles@shopify.com>
Signed-off-by: gatorsmile <gatorsmile@gmail.com>
## What changes were proposed in this pull request?
In SharedSparkSession and TestHive, we need to disable the rule ConvertToLocalRelation for better test case coverage.
## How was this patch tested?
Identify the failures after excluding "ConvertToLocalRelation" rule.
Closes#22270 from dilipbiswal/SPARK-25267-final.
Authored-by: Dilip Biswal <dbiswal@us.ibm.com>
Signed-off-by: gatorsmile <gatorsmile@gmail.com>
## What changes were proposed in this pull request?
This pr removed the method `updateBytesReadWithFileSize` in `FileScanRDD` because it computes input metrics by file size supported in Hadoop 2.5 and earlier. The current Spark does not support the versions, so it causes wrong input metric numbers.
This is rework from #22232.
Closes#22232
## How was this patch tested?
Added tests in `FileBasedDataSourceSuite`.
Closes#22324 from maropu/pr22232-2.
Lead-authored-by: dujunling <dujunling@huawei.com>
Co-authored-by: Takeshi Yamamuro <yamamuro@apache.org>
Signed-off-by: Sean Owen <sean.owen@databricks.com>
This is not a perfect solution. It is designed to minimize complexity on the basis of solving problems.
It is effective for English, Chinese characters, Japanese, Korean and so on.
```scala
before:
+---+---------------------------+-------------+
|id |中国 |s2 |
+---+---------------------------+-------------+
|1 |ab |[a] |
|2 |null |[中国, abc] |
|3 |ab1 |[hello world]|
|4 |か行 きゃ(kya) きゅ(kyu) きょ(kyo) |[“中国] |
|5 |中国（你好）a |[“中（国）, 312] |
|6 |中国山(东)服务区 |[“中(国）] |
|7 |中国山东服务区 |[中(国)] |
|8 | |[中国] |
+---+---------------------------+-------------+
after:
+---+-----------------------------------+----------------+
|id |中国 |s2 |
+---+-----------------------------------+----------------+
|1 |ab |[a] |
|2 |null |[中国, abc] |
|3 |ab1 |[hello world] |
|4 |か行 きゃ(kya) きゅ(kyu) きょ(kyo) |[“中国] |
|5 |中国（你好）a |[“中（国）, 312]|
|6 |中国山(东)服务区 |[“中(国）] |
|7 |中国山东服务区 |[中(国)] |
|8 | |[中国] |
+---+-----------------------------------+----------------+
```
## What changes were proposed in this pull request?
When there are wide characters such as Chinese characters or Japanese characters in the data, the show method has a alignment problem.
Try to fix this problem.
## How was this patch tested?
(Please explain how this patch was tested. E.g. unit tests, integration tests, manual tests)
Please review http://spark.apache.org/contributing.html before opening a pull request.
Closes#22048 from xuejianbest/master.
Authored-by: xuejianbest <384329882@qq.com>
Signed-off-by: Sean Owen <sean.owen@databricks.com>
## What changes were proposed in this pull request?
In the PR, I propose to extended `to_json` and support any types as element types of input arrays. It should allow converting arrays of primitive types and arrays of arrays. For example:
```
select to_json(array('1','2','3'))
> ["1","2","3"]
select to_json(array(array(1,2,3),array(4)))
> [[1,2,3],[4]]
```
## How was this patch tested?
Added a couple sql tests for arrays of primitive type and of arrays. Also I added round trip test `from_json` -> `to_json`.
Closes#22226 from MaxGekk/to_json-array.
Authored-by: Maxim Gekk <maxim.gekk@databricks.com>
Signed-off-by: hyukjinkwon <gurwls223@apache.org>
## What changes were proposed in this pull request?
Let's see the follow example:
```
val location = "/tmp/t"
val df = spark.range(10).toDF("id")
df.write.format("parquet").saveAsTable("tbl")
spark.sql("CREATE VIEW view1 AS SELECT id FROM tbl")
spark.sql(s"CREATE TABLE tbl2(ID long) USING parquet location $location")
spark.sql("INSERT OVERWRITE TABLE tbl2 SELECT ID FROM view1")
println(spark.read.parquet(location).schema)
spark.table("tbl2").show()
```
The output column name in schema will be `id` instead of `ID`, thus the last query shows nothing from `tbl2`.
By enabling the debug message we can see that the output naming is changed from `ID` to `id`, and then the `outputColumns` in `InsertIntoHadoopFsRelationCommand` is changed in `RemoveRedundantAliases`.
**To guarantee correctness**, we should change the output columns from `Seq[Attribute]` to `Seq[String]` to avoid its names being replaced by optimizer.
I will fix project elimination related rules in https://github.com/apache/spark/pull/22311 after this one.
## How was this patch tested?
Unit test.
Closes#22320 from gengliangwang/fixOutputSchema.
Authored-by: Gengliang Wang <gengliang.wang@databricks.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
## What changes were proposed in this pull request?
An alternative fix for https://github.com/apache/spark/pull/21698
When Spark rerun tasks for an RDD, there are 3 different behaviors:
1. determinate. Always return the same result with same order when rerun.
2. unordered. Returns same data set in random order when rerun.
3. indeterminate. Returns different result when rerun.
Normally Spark doesn't need to care about it. Spark runs stages one by one, when a task is failed, just rerun it. Although the rerun task may return a different result, users will not be surprised.
However, Spark may rerun a finished stage when seeing fetch failures. When this happens, Spark needs to rerun all the tasks of all the succeeding stages if the RDD output is indeterminate, because the input of the succeeding stages has been changed.
If the RDD output is determinate, we only need to rerun the failed tasks of the succeeding stages, because the input doesn't change.
If the RDD output is unordered, it's same as determinate, because shuffle partitioner is always deterministic(round-robin partitioner is not a shuffle partitioner that extends `org.apache.spark.Partitioner`), so the reducers will still get the same input data set.
This PR fixed the failure handling for `repartition`, to avoid correctness issues.
For `repartition`, it applies a stateful map function to generate a round-robin id, which is order sensitive and makes the RDD's output indeterminate. When the stage contains `repartition` reruns, we must also rerun all the tasks of all the succeeding stages.
**future improvement:**
1. Currently we can't rollback and rerun a shuffle map stage, and just fail. We should fix it later. https://issues.apache.org/jira/browse/SPARK-25341
2. Currently we can't rollback and rerun a result stage, and just fail. We should fix it later. https://issues.apache.org/jira/browse/SPARK-25342
3. We should provide public API to allow users to tag the random level of the RDD's computing function.
## How is this pull request tested?
a new test case
Closes#22112 from cloud-fan/repartition.
Lead-authored-by: Wenchen Fan <wenchen@databricks.com>
Co-authored-by: Xingbo Jiang <xingbo.jiang@databricks.com>
Signed-off-by: Xiao Li <gatorsmile@gmail.com>
## What changes were proposed in this pull request?
This is a follow-up of #22313 and aim to ignore the micro benchmark test which takes over 2 minutes in Jenkins.
- https://amplab.cs.berkeley.edu/jenkins/view/Spark%20QA%20Test%20(Dashboard)/job/spark-master-test-sbt-hadoop-2.6/4939/consoleFull
## How was this patch tested?
The test case should be ignored in Jenkins.
```
[info] FilterPushdownBenchmark:
...
[info] - Pushdown benchmark with many filters !!! IGNORED !!!
```
Closes#22336 from dongjoon-hyun/SPARK-25306-2.
Authored-by: Dongjoon Hyun <dongjoon@apache.org>
Signed-off-by: Xiao Li <gatorsmile@gmail.com>
## What changes were proposed in this pull request?
This is a followup of https://github.com/apache/spark/pull/22259 .
Scala case class has a wide surface: apply, unapply, accessors, copy, etc.
In https://github.com/apache/spark/pull/22259 , we change the type of `UserDefinedFunction.inputTypes` from `Option[Seq[DataType]]` to `Option[Seq[Schema]]`. This breaks backward compatibility.
This PR changes the type back, and use a `var` to keep the new nullable info.
## How was this patch tested?
N/A
Closes#22319 from cloud-fan/revert.
Authored-by: Wenchen Fan <wenchen@databricks.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
## What changes were proposed in this pull request?
Revert SPARK-24863 (#21819) and SPARK-24748 (#21721) as per discussion in #21721. We will revisit them when the data source v2 APIs are out.
## How was this patch tested?
Jenkins
Closes#22334 from zsxwing/revert-SPARK-24863-SPARK-24748.
Authored-by: Shixiong Zhu <zsxwing@gmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
## What changes were proposed in this pull request?
In both ORC data sources, `createFilter` function has exponential time complexity due to its skewed filter tree generation. This PR aims to improve it by using new `buildTree` function.
**REPRODUCE**
```scala
// Create and read 1 row table with 1000 columns
sql("set spark.sql.orc.filterPushdown=true")
val selectExpr = (1 to 1000).map(i => s"id c$i")
spark.range(1).selectExpr(selectExpr: _*).write.mode("overwrite").orc("/tmp/orc")
print(s"With 0 filters, ")
spark.time(spark.read.orc("/tmp/orc").count)
// Increase the number of filters
(20 to 30).foreach { width =>
val whereExpr = (1 to width).map(i => s"c$i is not null").mkString(" and ")
print(s"With $width filters, ")
spark.time(spark.read.orc("/tmp/orc").where(whereExpr).count)
}
```
**RESULT**
```scala
With 0 filters, Time taken: 653 ms
With 20 filters, Time taken: 962 ms
With 21 filters, Time taken: 1282 ms
With 22 filters, Time taken: 1982 ms
With 23 filters, Time taken: 3855 ms
With 24 filters, Time taken: 6719 ms
With 25 filters, Time taken: 12669 ms
With 26 filters, Time taken: 25032 ms
With 27 filters, Time taken: 49585 ms
With 28 filters, Time taken: 98980 ms // over 1 min 38 seconds
With 29 filters, Time taken: 198368 ms // over 3 mins
With 30 filters, Time taken: 393744 ms // over 6 mins
```
**AFTER THIS PR**
```scala
With 0 filters, Time taken: 774 ms
With 20 filters, Time taken: 601 ms
With 21 filters, Time taken: 399 ms
With 22 filters, Time taken: 679 ms
With 23 filters, Time taken: 363 ms
With 24 filters, Time taken: 342 ms
With 25 filters, Time taken: 336 ms
With 26 filters, Time taken: 352 ms
With 27 filters, Time taken: 322 ms
With 28 filters, Time taken: 302 ms
With 29 filters, Time taken: 307 ms
With 30 filters, Time taken: 301 ms
```
## How was this patch tested?
Pass the Jenkins with newly added test cases.
Closes#22313 from dongjoon-hyun/SPARK-25306.
Authored-by: Dongjoon Hyun <dongjoon@apache.org>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
## What changes were proposed in this pull request?
Previously in `TakeOrderedAndProjectSuite` the SparkSession will not get recycled when the test suite finishes.
## How was this patch tested?
N/A
Closes#22330 from jiangxb1987/SPARK-19355.
Authored-by: Xingbo Jiang <xingbo.jiang@databricks.com>
Signed-off-by: Xiao Li <gatorsmile@gmail.com>
## What changes were proposed in this pull request?
Move the output verification of Explain test cases to a new suite ExplainSuite.
## How was this patch tested?
N/A
Closes#22300 from gatorsmile/test3200.
Authored-by: Xiao Li <gatorsmile@gmail.com>
Signed-off-by: Xiao Li <gatorsmile@gmail.com>
## What changes were proposed in this pull request?
Currently, filter pushdown will not work if Parquet schema and Hive metastore schema are in different letter cases even spark.sql.caseSensitive is false.
Like the below case:
```scala
spark.sparkContext.hadoopConfiguration.setInt("parquet.block.size", 8 * 1024 * 1024)
spark.range(1, 40 * 1024 * 1024, 1, 1).sortWithinPartitions("id").write.parquet("/tmp/t")
sql("CREATE TABLE t (ID LONG) USING parquet LOCATION '/tmp/t'")
sql("select * from t where id < 100L").write.csv("/tmp/id")
```
Although filter "ID < 100L" is generated by Spark, it fails to pushdown into parquet actually, Spark still does the full table scan when reading.
This PR provides a case-insensitive field resolution to make it work.
Before - "ID < 100L" fail to pushedown:
<img width="273" alt="screen shot 2018-08-23 at 10 08 26 pm" src="https://user-images.githubusercontent.com/2989575/44530558-40ef8b00-a721-11e8-8abc-7f97671590d3.png">
After - "ID < 100L" pushedown sucessfully:
<img width="267" alt="screen shot 2018-08-23 at 10 08 40 pm" src="https://user-images.githubusercontent.com/2989575/44530567-44831200-a721-11e8-8634-e9f664b33d39.png">
## How was this patch tested?
Added UTs.
Closes#22197 from yucai/SPARK-25207.
Authored-by: yucai <yyu1@ebay.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
## What changes were proposed in this pull request?
### For `SPARK-5775 read array from partitioned_parquet_with_key_and_complextypes`:
scala2.12
```
scala> (1 to 10).toString
res4: String = Range 1 to 10
```
scala2.11
```
scala> (1 to 10).toString
res2: String = Range(1, 2, 3, 4, 5, 6, 7, 8, 9, 10)
```
And
```
def prepareAnswer(answer: Seq[Row], isSorted: Boolean): Seq[Row] = {
val converted: Seq[Row] = answer.map(prepareRow)
if (!isSorted) converted.sortBy(_.toString()) else converted
}
```
sortBy `_.toString` is not a good idea.
### Other failures are caused by
```
Array(Int.box(1)).toSeq == Array(Double.box(1.0)).toSeq
```
It is false in 2.12.2 + and is true in 2.11.x , 2.12.0, 2.12.1
## How was this patch tested?
This is a patch on a specific unit test.
Closes#22264 from sadhen/SPARK25256.
Authored-by: 忍冬 <rendong@wacai.com>
Signed-off-by: Sean Owen <sean.owen@databricks.com>
## What changes were proposed in this pull request?
`JavaColumnExpressionSuite.java` was added and `org.apache.spark.sql.ColumnExpressionSuite#test("isInCollection: Java Collection")` was removed.
It provides native Java tests for the method `org.apache.spark.sql.Column#isInCollection`.
Closes#22253 from aai95/isInCollectionJavaTest.
Authored-by: aai95 <aai95@yandex.ru>
Signed-off-by: DB Tsai <d_tsai@apple.com>
## What changes were proposed in this pull request?
Introduced by #21320 and #11744
```
$ sbt
> ++2.12.6
> project sql
> compile
...
[error] [warn] spark/sql/core/src/main/scala/org/apache/spark/sql/execution/ProjectionOverSchema.scala:41: match may not be exhaustive.
[error] It would fail on the following inputs: (_, ArrayType(_, _)), (_, _)
[error] [warn] getProjection(a.child).map(p => (p, p.dataType)).map {
[error] [warn]
[error] [warn] spark/sql/core/src/main/scala/org/apache/spark/sql/execution/ProjectionOverSchema.scala:52: match may not be exhaustive.
[error] It would fail on the following input: (_, _)
[error] [warn] getProjection(child).map(p => (p, p.dataType)).map {
[error] [warn]
...
```
And
```
$ sbt
> ++2.12.6
> project hive
> testOnly *ParquetMetastoreSuite
...
[error] /Users/rendong/wdi/spark/sql/hive/src/test/scala/org/apache/spark/sql/hive/HiveSparkSubmitSuite.scala:22: object tools is not a member of package scala
[error] import scala.tools.nsc.Properties
[error] ^
[error] /Users/rendong/wdi/spark/sql/hive/src/test/scala/org/apache/spark/sql/hive/HiveSparkSubmitSuite.scala:146: not found: value Properties
[error] val version = Properties.versionNumberString match {
[error] ^
[error] two errors found
...
```
## How was this patch tested?
Existing tests.
Closes#22260 from sadhen/fix_exhaustive_match.
Authored-by: 忍冬 <rendong@wacai.com>
Signed-off-by: hyukjinkwon <gurwls223@apache.org>
## What changes were proposed in this pull request?
`fromId` is the child, and `toId` is the parent, see line 127 in `buildSparkPlanGraphNode` above.
The edges in Spark UI also go from child to parent.
## How was this patch tested?
Comment change only. Inspected code above. Inspected how the edges in Spark UI look like.
Closes#22268 from juliuszsompolski/sparkplangraphedgedoc.
Authored-by: Juliusz Sompolski <julek@databricks.com>
Signed-off-by: Xiao Li <gatorsmile@gmail.com>
## What changes were proposed in this pull request?
Alternative take on https://github.com/apache/spark/pull/22063 that does not introduce udfInternal.
Resolve issue with inferring func types in 2.12 by instead using info captured when UDF is registered -- capturing which types are nullable (i.e. not primitive)
## How was this patch tested?
Existing tests.
Closes#22259 from srowen/SPARK-25044.2.
Authored-by: Sean Owen <sean.owen@databricks.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
## What changes were proposed in this pull request?
This changes the calls of `toPandas()` and `createDataFrame()` to use the Arrow stream format, when Arrow is enabled. Previously, Arrow data was written to byte arrays where each chunk is an output of the Arrow file format. This was mainly due to constraints at the time, and caused some overhead by writing the schema/footer on each chunk of data and then having to read multiple Arrow file inputs and concat them together.
Using the Arrow stream format has improved these by increasing performance, lower memory overhead for the average case, and simplified the code. Here are the details of this change:
**toPandas()**
_Before:_
Spark internal rows are converted to Arrow file format, each group of records is a complete Arrow file which contains the schema and other metadata. Next a collect is done and an Array of Arrow files is the result. After that each Arrow file is sent to Python driver which then loads each file and concats them to a single Arrow DataFrame.
_After:_
Spark internal rows are converted to ArrowRecordBatches directly, which is the simplest Arrow component for IPC data transfers. The driver JVM then immediately starts serving data to Python as an Arrow stream, sending the schema first. It then starts a Spark job with a custom handler that sends Arrow RecordBatches to Python. Partitions arriving in order are sent immediately, and out-of-order partitions are buffered until the ones that precede it come in. This improves performance, simplifies memory usage on executors, and improves the average memory usage on the JVM driver. Since the order of partitions must be preserved, the worst case is that the first partition will be the last to arrive all data must be buffered in memory until then. This case is no worse that before when doing a full collect.
**createDataFrame()**
_Before:_
A Pandas DataFrame is split into parts and each part is made into an Arrow file. Then each file is prefixed by the buffer size and written to a temp file. The temp file is read and each Arrow file is parallelized as a byte array.
_After:_
A Pandas DataFrame is split into parts, then an Arrow stream is written to a temp file where each part is an ArrowRecordBatch. The temp file is read as a stream and the Arrow messages are examined. If the message is an ArrowRecordBatch, the data is saved as a byte array. After reading the file, each ArrowRecordBatch is parallelized as a byte array. This has slightly more processing than before because we must look each Arrow message to extract the record batches, but performance ends up a litle better. It is cleaner in the sense that IPC from Python to JVM is done over a single Arrow stream.
## How was this patch tested?
Added new unit tests for the additions to ArrowConverters in Scala, existing tests for Python.
## Performance Tests - toPandas
Tests run on a 4 node standalone cluster with 32 cores total, 14.04.1-Ubuntu and OpenJDK 8
measured wall clock time to execute `toPandas()` and took the average best time of 5 runs/5 loops each.
Test code
```python
df = spark.range(1 << 25, numPartitions=32).toDF("id").withColumn("x1", rand()).withColumn("x2", rand()).withColumn("x3", rand()).withColumn("x4", rand())
for i in range(5):
start = time.time()
_ = df.toPandas()
elapsed = time.time() - start
```
Current Master | This PR
---------------------|------------
5.803557 | 5.16207
5.409119 | 5.133671
5.493509 | 5.147513
5.433107 | 5.105243
5.488757 | 5.018685
Avg Master | Avg This PR
------------------|--------------
5.5256098 | 5.1134364
Speedup of **1.08060595**
## Performance Tests - createDataFrame
Tests run on a 4 node standalone cluster with 32 cores total, 14.04.1-Ubuntu and OpenJDK 8
measured wall clock time to execute `createDataFrame()` and get the first record. Took the average best time of 5 runs/5 loops each.
Test code
```python
def run():
pdf = pd.DataFrame(np.random.rand(10000000, 10))
spark.createDataFrame(pdf).first()
for i in range(6):
start = time.time()
run()
elapsed = time.time() - start
gc.collect()
print("Run %d: %f" % (i, elapsed))
```
Current Master | This PR
--------------------|----------
6.234608 | 5.665641
6.32144 | 5.3475
6.527859 | 5.370803
6.95089 | 5.479151
6.235046 | 5.529167
Avg Master | Avg This PR
---------------|----------------
6.4539686 | 5.4784524
Speedup of **1.178064192**
## Memory Improvements
**toPandas()**
The most significant improvement is reduction of the upper bound space complexity in the JVM driver. Before, the entire dataset was collected in the JVM first before sending it to Python. With this change, as soon as a partition is collected, the result handler immediately sends it to Python, so the upper bound is the size of the largest partition. Also, using the Arrow stream format is more efficient because the schema is written once per stream, followed by record batches. The schema is now only send from driver JVM to Python. Before, multiple Arrow file formats were used that each contained the schema. This duplicated schema was created in the executors, sent to the driver JVM, and then Python where all but the first one received are discarded.
I verified the upper bound limit by running a test that would collect data that would exceed the amount of driver JVM memory available. Using these settings on a standalone cluster:
```
spark.driver.memory 1g
spark.executor.memory 5g
spark.sql.execution.arrow.enabled true
spark.sql.execution.arrow.fallback.enabled false
spark.sql.execution.arrow.maxRecordsPerBatch 0
spark.driver.maxResultSize 2g
```
Test code:
```python
from pyspark.sql.functions import rand
df = spark.range(1 << 25, numPartitions=32).toDF("id").withColumn("x1", rand()).withColumn("x2", rand()).withColumn("x3", rand())
df.toPandas()
```
This makes total data size of 33554432×8×4 = 1073741824
With the current master, it fails with OOM but passes using this PR.
**createDataFrame()**
No significant change in memory except that using the stream format instead of separate file formats avoids duplicated the schema, similar to toPandas above. The process of reading the stream and parallelizing the batches does cause the record batch message metadata to be copied, but it's size is insignificant.
Closes#21546 from BryanCutler/arrow-toPandas-stream-SPARK-23030.
Authored-by: Bryan Cutler <cutlerb@gmail.com>
Signed-off-by: hyukjinkwon <gurwls223@apache.org>
## What changes were proposed in this pull request?
Support Filter in ConvertToLocalRelation, similar to how Project works.
Additionally, in Optimizer, run ConvertToLocalRelation earlier to simplify the plan. This is good for very short queries which often are queries on local relations.
## How was this patch tested?
New test. Manual benchmark.
Author: Bogdan Raducanu <bogdan@databricks.com>
Author: Shixiong Zhu <zsxwing@gmail.com>
Author: Yinan Li <ynli@google.com>
Author: Li Jin <ice.xelloss@gmail.com>
Author: s71955 <sujithchacko.2010@gmail.com>
Author: DB Tsai <d_tsai@apple.com>
Author: jaroslav chládek <mastermism@gmail.com>
Author: Huangweizhe <huangweizhe@bbdservice.com>
Author: Xiangrui Meng <meng@databricks.com>
Author: hyukjinkwon <gurwls223@apache.org>
Author: Kent Yao <yaooqinn@hotmail.com>
Author: caoxuewen <cao.xuewen@zte.com.cn>
Author: liuxian <liu.xian3@zte.com.cn>
Author: Adam Bradbury <abradbury@users.noreply.github.com>
Author: Jose Torres <torres.joseph.f+github@gmail.com>
Author: Yuming Wang <yumwang@ebay.com>
Author: Liang-Chi Hsieh <viirya@gmail.com>
Closes#22205 from bogdanrdc/local-relation-filter.
## What changes were proposed in this pull request?
This adds `spark.executor.pyspark.memory` to configure Python's address space limit, [`resource.RLIMIT_AS`](https://docs.python.org/3/library/resource.html#resource.RLIMIT_AS). Limiting Python's address space allows Python to participate in memory management. In practice, we see fewer cases of Python taking too much memory because it doesn't know to run garbage collection. This results in YARN killing fewer containers. This also improves error messages so users know that Python is consuming too much memory:
```
File "build/bdist.linux-x86_64/egg/package/library.py", line 265, in fe_engineer
fe_eval_rec.update(f(src_rec_prep, mat_rec_prep))
File "build/bdist.linux-x86_64/egg/package/library.py", line 163, in fe_comp
comparisons = EvaluationUtils.leven_list_compare(src_rec_prep.get(item, []), mat_rec_prep.get(item, []))
File "build/bdist.linux-x86_64/egg/package/evaluationutils.py", line 25, in leven_list_compare
permutations = sorted(permutations, reverse=True)
MemoryError
```
The new pyspark memory setting is used to increase requested YARN container memory, instead of sharing overhead memory between python and off-heap JVM activity.
## How was this patch tested?
Tested memory limits in our YARN cluster and verified that MemoryError is thrown.
Author: Ryan Blue <blue@apache.org>
Closes#21977 from rdblue/SPARK-25004-add-python-memory-limit.
## What changes were proposed in this pull request?
In the PR, I propose to not perform recursive parallel listening of files in the `scanPartitions` method because it can cause a deadlock. Instead of that I propose to do `scanPartitions` in parallel for top level partitions only.
## How was this patch tested?
I extended an existing test to trigger the deadlock.
Author: Maxim Gekk <maxim.gekk@databricks.com>
Closes#22233 from MaxGekk/fix-recover-partitions.
## What changes were proposed in this pull request?
This PR implements the possibility of the user to override the maximum number of buckets when saving to a table.
Currently the limit is a hard-coded 100k, which might be insufficient for large workloads.
A new configuration entry is proposed: `spark.sql.bucketing.maxBuckets`, which defaults to the previous 100k.
## How was this patch tested?
Added unit tests in the following spark.sql test suites:
- CreateTableAsSelectSuite
- BucketedWriteSuite
Author: Fernando Pereira <fernando.pereira@epfl.ch>
Closes#21087 from ferdonline/enh/configurable_bucket_limit.
## What changes were proposed in this pull request?
The PR excludes Python UDFs filters in FileSourceStrategy so that they don't ExtractPythonUDF rule to throw exception. It doesn't make sense to pass Python UDF filters in FileSourceStrategy anyway because they cannot be used as push down filters.
## How was this patch tested?
Add a new regression test
Closes#22104 from icexelloss/SPARK-24721-udf-filter.
Authored-by: Li Jin <ice.xelloss@gmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
## What changes were proposed in this pull request?
this pr add a configuration parameter to configure the capacity of fast aggregation.
Performance comparison:
```
Java HotSpot(TM) 64-Bit Server VM 1.8.0_60-b27 on Windows 7 6.1
Intel64 Family 6 Model 94 Stepping 3, GenuineIntel
Aggregate w multiple keys: Best/Avg Time(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------
fasthash = default 5612 / 5882 3.7 267.6 1.0X
fasthash = config 3586 / 3595 5.8 171.0 1.6X
```
## How was this patch tested?
the existed test cases.
Closes#21931 from heary-cao/FastHashCapacity.
Authored-by: caoxuewen <cao.xuewen@zte.com.cn>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
## What changes were proposed in this pull request?
This is based on the discussion https://github.com/apache/spark/pull/16677/files#r212805327.
As SQL standard doesn't mandate that a nested order by followed by a limit has to respect that ordering clause, this patch removes the `child.outputOrdering` check.
## How was this patch tested?
Unit tests.
Closes#22239 from viirya/improve-global-limit-parallelism-followup.
Authored-by: Liang-Chi Hsieh <viirya@gmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
## What changes were proposed in this pull request?
Improved the documentation for the datetime functions in `org.apache.spark.sql.functions` by adding details about the supported column input types, the column return type, behaviour on invalid input, supporting examples and clarifications.
## How was this patch tested?
Manually testing each of the datetime functions with different input to ensure that the corresponding Javadoc/Scaladoc matches the behaviour of the function. Successfully ran the `unidoc` SBT process.
Closes#20901 from abradbury/SPARK-23792.
Authored-by: Adam Bradbury <abradbury@users.noreply.github.com>
Signed-off-by: Sean Owen <sean.owen@databricks.com>
## What changes were proposed in this pull request?
This PR generates the code that to refer a `StructType` generated in the scala code instead of generating `StructType` in Java code.
The original code has two issues.
1. Avoid to used the field name such as `key.name`
1. Support complicated schema (e.g. nested DataType)
At first, [the JIRA entry](https://issues.apache.org/jira/browse/SPARK-25178) proposed to change the generated field name of the keySchema / valueSchema to a dummy name in `RowBasedHashMapGenerator` and `VectorizedHashMapGenerator.scala`. This proposal can addresse issue 1.
Ueshin suggested an approach to refer to a `StructType` generated in the scala code using `ctx.addReferenceObj()`. This approach can address issues 1 and 2. Finally, this PR uses this approach.
## How was this patch tested?
Existing UTs
Closes#22187 from kiszk/SPARK-25178.
Authored-by: Kazuaki Ishizaki <ishizaki@jp.ibm.com>
Signed-off-by: Takuya UESHIN <ueshin@databricks.com>
(Link to Jira: https://issues.apache.org/jira/browse/SPARK-4502)
_N.B. This is a restart of PR #16578 which includes a subset of that code. Relevant review comments from that PR should be considered incorporated by reference. Please avoid duplication in review by reviewing that PR first. The summary below is an edited copy of the summary of the previous PR._
## What changes were proposed in this pull request?
One of the hallmarks of a column-oriented data storage format is the ability to read data from a subset of columns, efficiently skipping reads from other columns. Spark has long had support for pruning unneeded top-level schema fields from the scan of a parquet file. For example, consider a table, `contacts`, backed by parquet with the following Spark SQL schema:
```
root
|-- name: struct
| |-- first: string
| |-- last: string
|-- address: string
```
Parquet stores this table's data in three physical columns: `name.first`, `name.last` and `address`. To answer the query
```SQL
select address from contacts
```
Spark will read only from the `address` column of parquet data. However, to answer the query
```SQL
select name.first from contacts
```
Spark will read `name.first` and `name.last` from parquet.
This PR modifies Spark SQL to support a finer-grain of schema pruning. With this patch, Spark reads only the `name.first` column to answer the previous query.
### Implementation
There are two main components of this patch. First, there is a `ParquetSchemaPruning` optimizer rule for gathering the required schema fields of a `PhysicalOperation` over a parquet file, constructing a new schema based on those required fields and rewriting the plan in terms of that pruned schema. The pruned schema fields are pushed down to the parquet requested read schema. `ParquetSchemaPruning` uses a new `ProjectionOverSchema` extractor for rewriting a catalyst expression in terms of a pruned schema.
Second, the `ParquetRowConverter` has been patched to ensure the ordinals of the parquet columns read are correct for the pruned schema. `ParquetReadSupport` has been patched to address a compatibility mismatch between Spark's built in vectorized reader and the parquet-mr library's reader.
### Limitation
Among the complex Spark SQL data types, this patch supports parquet column pruning of nested sequences of struct fields only.
## How was this patch tested?
Care has been taken to ensure correctness and prevent regressions. A more advanced version of this patch incorporating optimizations for rewriting queries involving aggregations and joins has been running on a production Spark cluster at VideoAmp for several years. In that time, one bug was found and fixed early on, and we added a regression test for that bug.
We forward-ported this patch to Spark master in June 2016 and have been running this patch against Spark 2.x branches on ad-hoc clusters since then.
Closes#21320 from mallman/spark-4502-parquet_column_pruning-foundation.
Lead-authored-by: Michael Allman <msa@allman.ms>
Co-authored-by: Adam Jacques <adam@technowizardry.net>
Co-authored-by: Michael Allman <michael@videoamp.com>
Signed-off-by: Xiao Li <gatorsmile@gmail.com>
## What changes were proposed in this pull request?
Dataset.apply calls dataset.deserializer (to provide an early error) which ends up calling the full Analyzer on the deserializer. This can take tens of milliseconds, depending on how big the plan is.
Since Dataset.apply is called for many Dataset operations such as Dataset.where it can be a significant overhead for short queries.
According to a comment in the PR that introduced this check, we can at least remove this check for DataFrames: https://github.com/apache/spark/pull/20402#discussion_r164338267
## How was this patch tested?
Existing tests + manual benchmark
Author: Bogdan Raducanu <bogdan@databricks.com>
Closes#22201 from bogdanrdc/deserializer-fix.
## What changes were proposed in this pull request?
**Problem statement**
load data command with hdfs file paths consists of wild card strings like * are not working
eg:
"load data inpath 'hdfs://hacluster/user/ext* into table t1"
throws Analysis exception while executing this query
**Analysis -**
Currently fs.exists() API which is used for path validation in load command API cannot resolve the path with wild card pattern, To mitigate this problem i am using globStatus() API another api which can resolve the paths with hdfs supported wildcards like *,? etc(inline with hive wildcard support).
**Improvement identified as part of this issue -**
Currently system wont support wildcard character to be used for folder level path in a local file system. This PR has handled this scenario, the same globStatus API will unify the validation logic of local and non local file systems, this will ensure the behavior consistency between the hdfs and local file path in load command.
with this improvement user will be able to use a wildcard character in folder level path of a local file system in load command inline with hive behaviour, in older versions user can use wildcards only in file path of the local file system if they use in folder path system use to give an error by mentioning that not supported.
eg: load data local inpath '/localfilesystem/folder* into table t1
## How was this patch tested?
a) Manually tested by executing test-cases in HDFS yarn cluster. Reports is been attached in below section.
b) Existing test-case can verify the impact and functionality for local file path scenarios
c) A test-case is been added for verifying the functionality when wild card is been used in folder level path of a local file system
## Test Results
Note: all ip's were updated to localhost for security reasons.
HDFS path details
```
vm1:/opt/ficlient # hadoop fs -ls /user/data/sujith1
Found 2 items
-rw-r--r-- 3 shahid hadoop 4802 2018-03-26 15:45 /user/data/sujith1/typeddata60.txt
-rw-r--r-- 3 shahid hadoop 4883 2018-03-26 15:45 /user/data/sujith1/typeddata61.txt
vm1:/opt/ficlient # hadoop fs -ls /user/data/sujith2
Found 2 items
-rw-r--r-- 3 shahid hadoop 4802 2018-03-26 15:45 /user/data/sujith2/typeddata60.txt
-rw-r--r-- 3 shahid hadoop 4883 2018-03-26 15:45 /user/data/sujith2/typeddata61.txt
```
positive scenario by specifying complete file path to know about record size
```
0: jdbc:hive2://localhost:22550/default> create table wild_spark (time timestamp, name string, isright boolean, datetoday date, num binary, height double, score float, decimaler decimal(10,0), id tinyint, age int, license bigint, length smallint) row format delimited fields terminated by ',';
+---------+--+
| Result |
+---------+--+
+---------+--+
No rows selected (1.217 seconds)
0: jdbc:hive2://localhost:22550/default> load data inpath '/user/data/sujith1/typeddata60.txt' into table wild_spark;
+---------+--+
| Result |
+---------+--+
+---------+--+
No rows selected (4.236 seconds)
0: jdbc:hive2://localhost:22550/default> load data inpath '/user/data/sujith1/typeddata61.txt' into table wild_spark;
+---------+--+
| Result |
+---------+--+
+---------+--+
No rows selected (0.602 seconds)
0: jdbc:hive2://localhost:22550/default> select count(*) from wild_spark;
+-----------+--+
| count(1) |
+-----------+--+
| 121 |
+-----------+--+
1 row selected (18.529 seconds)
0: jdbc:hive2://localhost:22550/default>
```
With wild card character in file path
```
0: jdbc:hive2://localhost:22550/default> create table spark_withWildChar (time timestamp, name string, isright boolean, datetoday date, num binary, height double, score float, decimaler decimal(10,0), id tinyint, age int, license bigint, length smallint) row format delimited fields terminated by ',';
+---------+--+
| Result |
+---------+--+
+---------+--+
No rows selected (0.409 seconds)
0: jdbc:hive2://localhost:22550/default> load data inpath '/user/data/sujith1/type*' into table spark_withWildChar;
+---------+--+
| Result |
+---------+--+
+---------+--+
No rows selected (1.502 seconds)
0: jdbc:hive2://localhost:22550/default> select count(*) from spark_withWildChar;
+-----------+--+
| count(1) |
+-----------+--+
| 121 |
+-----------+--+
```
with ? wild card scenario
```
0: jdbc:hive2://localhost:22550/default> create table spark_withWildChar_DiffChar (time timestamp, name string, isright boolean, datetoday date, num binary, height double, score float, decimaler decimal(10,0), id tinyint, age int, license bigint, length smallint) row format delimited fields terminated by ',';
+---------+--+
| Result |
+---------+--+
+---------+--+
No rows selected (0.489 seconds)
0: jdbc:hive2://localhost:22550/default> load data inpath '/user/data/sujith1/?ypeddata60.txt' into table spark_withWildChar_DiffChar;
+---------+--+
| Result |
+---------+--+
+---------+--+
No rows selected (1.152 seconds)
0: jdbc:hive2://localhost:22550/default> load data inpath '/user/data/sujith1/?ypeddata61.txt' into table spark_withWildChar_DiffChar;
+---------+--+
| Result |
+---------+--+
+---------+--+
No rows selected (0.644 seconds)
0: jdbc:hive2://localhost:22550/default> select count(*) from spark_withWildChar_DiffChar;
+-----------+--+
| count(1) |
+-----------+--+
| 121 |
+-----------+--+
1 row selected (16.078 seconds)
```
with folder level wild card scenario
```
0: jdbc:hive2://localhost:22550/default> create table spark_withWildChar_folderlevel (time timestamp, name string, isright boolean, datetoday date, num binary, height double, score float, decimaler decimal(10,0), id tinyint, age int, license bigint, length smallint) row format delimited fields terminated by ',';
+---------+--+
| Result |
+---------+--+
+---------+--+
No rows selected (0.489 seconds)
0: jdbc:hive2://localhost:22550/default> load data inpath '/user/data/suji*/*' into table spark_withWildChar_folderlevel;
+---------+--+
| Result |
+---------+--+
+---------+--+
No rows selected (1.152 seconds)
0: jdbc:hive2://localhost:22550/default> select count(*) from spark_withWildChar_folderlevel;
+-----------+--+
| count(1) |
+-----------+--+
| 242 |
+-----------+--+
1 row selected (16.078 seconds)
```
Negative scenario invalid path
```
0: jdbc:hive2://localhost:22550/default> load data inpath '/user/data/sujiinvalid*/*' into table spark_withWildChar_folder;
Error: org.apache.spark.sql.AnalysisException: LOAD DATA input path does not exist: /user/data/sujiinvalid*/*; (state=,code=0)
0: jdbc:hive2://localhost:22550/default>
```
Hive Test results- file level
```
0: jdbc:hive2://localhost:21066/> create table hive_withWildChar_files (time timestamp, name string, isright boolean, datetoday date, num binary, height double, score float, decimaler decimal(10,0), id tinyint, age int, license bigint, length smallint) stored as TEXTFILE;
No rows affected (0.723 seconds)
0: jdbc:hive2://localhost:21066/> load data inpath '/user/data/sujith1/type*' into table hive_withWildChar_files;
INFO : Loading data to table default.hive_withwildchar_files from hdfs://hacluster/user/sujith1/type*
No rows affected (0.682 seconds)
0: jdbc:hive2://localhost:21066/> select count(*) from hive_withWildChar_files;
+------+--+
| _c0 |
+------+--+
| 121 |
+------+--+
1 row selected (50.832 seconds)
```
Hive Test results- folder level
```
0: jdbc:hive2://localhost:21066/> create table hive_withWildChar_folder (time timestamp, name string, isright boolean, datetoday date, num binary, height double, score float, decimaler decimal(10,0), id tinyint, age int, license bigint, length smallint) stored as TEXTFILE;
No rows affected (0.459 seconds)
0: jdbc:hive2://localhost:21066/> load data inpath '/user/data/suji*/*' into table hive_withWildChar_folder;
INFO : Loading data to table default.hive_withwildchar_folder from hdfs://hacluster/user/data/suji*/*
No rows affected (0.76 seconds)
0: jdbc:hive2://localhost:21066/> select count(*) from hive_withWildChar_folder;
+------+--+
| _c0 |
+------+--+
| 242 |
+------+--+
1 row selected (46.483 seconds)
```
Closes#20611 from sujith71955/master_wldcardsupport.
Lead-authored-by: s71955 <sujithchacko.2010@gmail.com>
Co-authored-by: sujith71955 <sujithchacko.2010@gmail.com>
Signed-off-by: hyukjinkwon <gurwls223@apache.org>
## What changes were proposed in this pull request?
Fix a race in the rate source tests. We need a better way of testing restart behavior.
## How was this patch tested?
unit test
Closes#22191 from jose-torres/racetest.
Authored-by: Jose Torres <torres.joseph.f+github@gmail.com>
Signed-off-by: Tathagata Das <tathagata.das1565@gmail.com>
## What changes were proposed in this pull request?
Casting to `DecimalType` is not always needed to force nullable.
If the decimal type to cast is wider than original type, or only truncating or precision loss, the casted value won't be `null`.
## How was this patch tested?
Added and modified tests.
Closes#22200 from ueshin/issues/SPARK-25208/cast_nullable_decimal.
Authored-by: Takuya UESHIN <ueshin@databricks.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
## What changes were proposed in this pull request?
[SPARK-25126] (https://issues.apache.org/jira/browse/SPARK-25126)
reports loading a large number of orc files consumes a lot of memory
in both 2.0 and 2.3. The issue is caused by creating a Reader for every
orc file in order to infer the schema.
In OrFileOperator.ReadSchema, a Reader is created for every file
although only the first valid one is used. This uses significant
amount of memory when there `paths` have a lot of files. In 2.3
a different code path (OrcUtils.readSchema) is used for inferring
schema for orc files. This commit changes both functions to create
Reader lazily.
## How was this patch tested?
Pass the Jenkins with a newly added test case by dongjoon-hyun
Closes#22157 from raofu/SPARK-25126.
Lead-authored-by: Rao Fu <rao@coupang.com>
Co-authored-by: Dongjoon Hyun <dongjoon@apache.org>
Co-authored-by: Rao Fu <raofu04@gmail.com>
Signed-off-by: hyukjinkwon <gurwls223@apache.org>
## What changes were proposed in this pull request?
VectorizedParquetRecordReader::initializeInternal rebuilds the column list and path list once for each column. Therefore, it indirectly iterates 2\*colCount\*colCount times for each parquet file.
This inefficiency impacts jobs that read parquet-backed tables with many columns and many files. Jobs that read tables with few columns or few files are not impacted.
This PR changes initializeInternal so that it builds each list only once.
I ran benchmarks on my laptop with 1 worker thread, running this query:
<pre>
sql("select * from parquet_backed_table where id1 = 1").collect
</pre>
There are roughly one matching row for every 425 rows, and the matching rows are sprinkled pretty evenly throughout the table (that is, every page for column <code>id1</code> has at least one matching row).
6000 columns, 1 million rows, 67 32M files:
master | branch | improvement
-------|---------|-----------
10.87 min | 6.09 min | 44%
6000 columns, 1 million rows, 23 98m files:
master | branch | improvement
-------|---------|-----------
7.39 min | 5.80 min | 21%
600 columns 10 million rows, 67 32M files:
master | branch | improvement
-------|---------|-----------
1.95 min | 1.96 min | -0.5%
60 columns, 100 million rows, 67 32M files:
master | branch | improvement
-------|---------|-----------
0.55 min | 0.55 min | 0%
## How was this patch tested?
- sql unit tests
- pyspark-sql tests
Closes#22188 from bersprockets/SPARK-25164.
Authored-by: Bruce Robbins <bersprockets@gmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
## What changes were proposed in this pull request?
This pr proposed to show RDD/relation names in RDD/Hive table scan nodes.
This change made these names show up in the webUI and explain results.
For example;
```
scala> sql("CREATE TABLE t(c1 int) USING hive")
scala> sql("INSERT INTO t VALUES(1)")
scala> spark.table("t").explain()
== Physical Plan ==
Scan hive default.t [c1#8], HiveTableRelation `default`.`t`, org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe, [c1#8]
^^^^^^^^^^^
```
<img width="212" alt="spark-pr-hive" src="https://user-images.githubusercontent.com/692303/44501013-51264c80-a6c6-11e8-94f8-0704aee83bb6.png">
Closes#20226
## How was this patch tested?
Added tests in `DataFrameSuite`, `DatasetSuite`, and `HiveExplainSuite`
Closes#22153 from maropu/pr20226.
Lead-authored-by: Takeshi Yamamuro <yamamuro@apache.org>
Co-authored-by: Tejas Patil <tejasp@fb.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
## What changes were proposed in this pull request?
They depend on internal Expression APIs. Let's see how far we can get without it.
## How was this patch tested?
Just some code removal. There's no existing tests as far as I can tell so it's easy to remove.
Closes#22185 from rxin/SPARK-25127.
Authored-by: Reynold Xin <rxin@databricks.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
## What changes were proposed in this pull request?
The race condition that caused test failure is between 2 threads.
- The MicrobatchExecution thread that processes inputs to produce answers and then generates progress events.
- The test thread that generates some input data, checked the answer and then verified the query generated progress event.
The synchronization structure between these threads is as follows
1. MicrobatchExecution thread, in every batch, does the following in order.
a. Processes batch input to generate answer.
b. Signals `awaitProgressLockCondition` to wake up threads waiting for progress using `awaitOffset`
c. Generates progress event
2. Test execution thread
a. Calls `awaitOffset` to wait for progress, which waits on `awaitProgressLockCondition`.
b. As soon as `awaitProgressLockCondition` is signaled, it would move on the in the test to check answer.
c. Finally, it would verify the last generated progress event.
What can happen is the following sequence of events: 2a -> 1a -> 1b -> 2b -> 2c -> 1c.
In other words, the progress event may be generated after the test tries to verify it.
The solution has two steps.
1. Signal the waiting thread after the progress event has been generated, that is, after `finishTrigger()`.
2. Increase the timeout of `awaitProgressLockCondition.await(100 ms)` to a large value.
This latter is to ensure that test thread for keeps waiting on `awaitProgressLockCondition`until the MicroBatchExecution thread explicitly signals it. With the existing small timeout of 100ms the following sequence can occur.
- MicroBatchExecution thread updates committed offsets
- Test thread waiting on `awaitProgressLockCondition` accidentally times out after 100 ms, finds that the committed offsets have been updated, therefore returns from `awaitOffset` and moves on to the progress event tests.
- MicroBatchExecution thread then generates progress event and signals. But the test thread has already attempted to verify the event and failed.
By increasing the timeout to large (e.g., `streamingTimeoutMs = 60 seconds`, similar to `awaitInitialization`), this above type of race condition is also avoided.
## How was this patch tested?
Ran locally many times.
Closes#22182 from tdas/SPARK-25184.
Authored-by: Tathagata Das <tathagata.das1565@gmail.com>
Signed-off-by: Tathagata Das <tathagata.das1565@gmail.com>
