## What changes were proposed in this pull request?
Previously range operator could not be interrupted. For example, using DAGScheduler.cancelStage(...) on a query with range might have been ineffective.
This change adds periodic checks of TaskContext.isInterrupted to codegen version, and InterruptibleOperator to non-codegen version.
I benchmarked the performance of codegen version on a sample query `spark.range(1000L * 1000 * 1000 * 10).count()` and there is no measurable difference.
## How was this patch tested?
Adds a unit test.
Author: Ala Luszczak <ala@databricks.com>
Closes#16872 from ala/SPARK-19514b.
## What changes were proposed in this pull request?
SPARK-19265 had made table relation cache general; this follow-up aims to make `tableRelationCache`'s maximum size configurable.
In order to do sanity-check, this patch also adds a `checkValue()` method to `TypedConfigBuilder`.
## How was this patch tested?
new test case: `test("conf entry: checkValue()")`
Author: Liwei Lin <lwlin7@gmail.com>
Closes#16736 from lw-lin/conf.
## What changes were proposed in this pull request?
- Remove support for Hadoop 2.5 and earlier
- Remove reflection and code constructs only needed to support multiple versions at once
- Update docs to reflect newer versions
- Remove older versions' builds and profiles.
## How was this patch tested?
Existing tests
Author: Sean Owen <sowen@cloudera.com>
Closes#16810 from srowen/SPARK-19464.
## What changes were proposed in this pull request?
when csv infer schema, it does not use user defined csvoptions to parse the field, such as `inf`, `-inf` which are should be parsed to DoubleType
this pr add `options.nanValue`, `options.negativeInf`, `options.positiveIn` to check if the field is a DoubleType
## How was this patch tested?
unit test added
Author: windpiger <songjun@outlook.com>
Closes#16834 from windpiger/fixinferInfSchemaCsv.
## What changes were proposed in this pull request?
addBatch method in Sink trait is supposed to be a synchronous method to coordinate with the fault-tolerance design in StreamingExecution (being different with the compute() method in DStream)
We need to add more notes in the comments of this method to remind the developers
## How was this patch tested?
existing tests
Author: CodingCat <zhunansjtu@gmail.com>
Closes#16840 from CodingCat/SPARK-19499.
## What changes were proposed in this pull request?
`mapGroupsWithState` is a new API for arbitrary stateful operations in Structured Streaming, similar to `DStream.mapWithState`
*Requirements*
- Users should be able to specify a function that can do the following
- Access the input row corresponding to a key
- Access the previous state corresponding to a key
- Optionally, update or remove the state
- Output any number of new rows (or none at all)
*Proposed API*
```
// ------------ New methods on KeyValueGroupedDataset ------------
class KeyValueGroupedDataset[K, V] {
// Scala friendly
def mapGroupsWithState[S: Encoder, U: Encoder](func: (K, Iterator[V], KeyedState[S]) => U)
def flatMapGroupsWithState[S: Encode, U: Encoder](func: (K, Iterator[V], KeyedState[S]) => Iterator[U])
// Java friendly
def mapGroupsWithState[S, U](func: MapGroupsWithStateFunction[K, V, S, R], stateEncoder: Encoder[S], resultEncoder: Encoder[U])
def flatMapGroupsWithState[S, U](func: FlatMapGroupsWithStateFunction[K, V, S, R], stateEncoder: Encoder[S], resultEncoder: Encoder[U])
}
// ------------------- New Java-friendly function classes -------------------
public interface MapGroupsWithStateFunction<K, V, S, R> extends Serializable {
R call(K key, Iterator<V> values, state: KeyedState<S>) throws Exception;
}
public interface FlatMapGroupsWithStateFunction<K, V, S, R> extends Serializable {
Iterator<R> call(K key, Iterator<V> values, state: KeyedState<S>) throws Exception;
}
// ---------------------- Wrapper class for state data ----------------------
trait State[S] {
def exists(): Boolean
def get(): S // throws Exception is state does not exist
def getOption(): Option[S]
def update(newState: S): Unit
def remove(): Unit // exists() will be false after this
}
```
Key Semantics of the State class
- The state can be null.
- If the state.remove() is called, then state.exists() will return false, and getOption will returm None.
- After that state.update(newState) is called, then state.exists() will return true, and getOption will return Some(...).
- None of the operations are thread-safe. This is to avoid memory barriers.
*Usage*
```
val stateFunc = (word: String, words: Iterator[String, runningCount: KeyedState[Long]) => {
val newCount = words.size + runningCount.getOption.getOrElse(0L)
runningCount.update(newCount)
(word, newCount)
}
dataset // type is Dataset[String]
.groupByKey[String](w => w) // generates KeyValueGroupedDataset[String, String]
.mapGroupsWithState[Long, (String, Long)](stateFunc) // returns Dataset[(String, Long)]
```
## How was this patch tested?
New unit tests.
Author: Tathagata Das <tathagata.das1565@gmail.com>
Closes#16758 from tdas/mapWithState.
### What changes were proposed in this pull request?
Prior to Spark 2.1, the option names are case sensitive for all the formats. Since Spark 2.1, the option key names become case insensitive except the format `Text` and `LibSVM `. This PR is to fix these issues.
Also, add a check to know whether the input option vector type is legal for `LibSVM`.
### How was this patch tested?
Added test cases
Author: gatorsmile <gatorsmile@gmail.com>
Closes#16737 from gatorsmile/libSVMTextOptions.
## What changes were proposed in this pull request?
This pull request makes SQLConf slightly more extensible by removing the visibility limitations on the build* functions.
## How was this patch tested?
N/A - there are no logic changes and everything should be covered by existing unit tests.
Author: Reynold Xin <rxin@databricks.com>
Closes#16835 from rxin/SPARK-19495.
## What changes were proposed in this pull request?
This pull request adds two new user facing functions:
- `to_date` which accepts an expression and a format and returns a date.
- `to_timestamp` which accepts an expression and a format and returns a timestamp.
For example, Given a date in format: `2016-21-05`. (YYYY-dd-MM)
### Date Function
*Previously*
```
to_date(unix_timestamp(lit("2016-21-05"), "yyyy-dd-MM").cast("timestamp"))
```
*Current*
```
to_date(lit("2016-21-05"), "yyyy-dd-MM")
```
### Timestamp Function
*Previously*
```
unix_timestamp(lit("2016-21-05"), "yyyy-dd-MM").cast("timestamp")
```
*Current*
```
to_timestamp(lit("2016-21-05"), "yyyy-dd-MM")
```
### Tasks
- [X] Add `to_date` to Scala Functions
- [x] Add `to_date` to Python Functions
- [x] Add `to_date` to SQL Functions
- [X] Add `to_timestamp` to Scala Functions
- [x] Add `to_timestamp` to Python Functions
- [x] Add `to_timestamp` to SQL Functions
- [x] Add function to R
## How was this patch tested?
- [x] Add Functions to `DateFunctionsSuite`
- Test new `ParseToTimestamp` Expression (*not necessary*)
- Test new `ParseToDate` Expression (*not necessary*)
- [x] Add test for R
- [x] Add test for Python in test.py
Please review http://spark.apache.org/contributing.html before opening a pull request.
Author: anabranch <wac.chambers@gmail.com>
Author: Bill Chambers <bill@databricks.com>
Author: anabranch <bill@databricks.com>
Closes#16138 from anabranch/SPARK-16609.
## What changes were proposed in this pull request?
This change introduces a new metric "number of generated rows". It is used exclusively for Range, which is a leaf in the query tree, yet doesn't read any input data, and therefore cannot report "recordsRead".
Additionally the way in which the metrics are reported by the JIT-compiled version of Range was changed. Previously, it was immediately reported that all the records were produced. This could be confusing for a user monitoring execution progress in the UI. Now, the metric is updated gradually.
In order to avoid negative impact on Range performance, the code generation was reworked. The values are now produced in batches in the tighter inner loop, while the metrics are updated in the outer loop.
The change also contains a number of unit tests, which should help ensure the correctness of metrics for various input sources.
## How was this patch tested?
Unit tests.
Author: Ala Luszczak <ala@databricks.com>
Closes#16829 from ala/SPARK-19447.
## What changes were proposed in this pull request?
This PR refactors CSV schema inference path to be consistent with JSON data source and moves some filtering codes having the similar/same logics into `CSVUtils`.
It makes the methods in classes have consistent arguments with JSON ones. (this PR renames `.../json/InferSchema.scala` → `.../json/JsonInferSchema.scala`)
`CSVInferSchema` and `JsonInferSchema`
``` scala
private[csv] object CSVInferSchema {
...
def infer(
csv: Dataset[String],
caseSensitive: Boolean,
options: CSVOptions): StructType = {
...
```
``` scala
private[sql] object JsonInferSchema {
...
def infer(
json: RDD[String],
columnNameOfCorruptRecord: String,
configOptions: JSONOptions): StructType = {
...
```
These allow schema inference from `Dataset[String]` directly, meaning the similar functionalities that use `JacksonParser`/`JsonInferSchema` for JSON can be easily implemented by `UnivocityParser`/`CSVInferSchema` for CSV.
This completes refactoring CSV datasource and they are now pretty consistent.
## How was this patch tested?
Existing tests should cover this and
```
./dev/change-scala-version.sh 2.10
./build/mvn -Pyarn -Phadoop-2.4 -Dscala-2.10 -DskipTests clean package
```
Author: hyukjinkwon <gurwls223@gmail.com>
Closes#16680 from HyukjinKwon/SPARK-16101-schema-inference.
## What changes were proposed in this pull request?
```
Caused by: java.lang.IllegalArgumentException: Wrong FS: s3a://**************/checkpoint/7b2231a3-d845-4740-bfa3-681850e5987f/metadata, expected: file:///
at org.apache.hadoop.fs.FileSystem.checkPath(FileSystem.java:649)
at org.apache.hadoop.fs.RawLocalFileSystem.pathToFile(RawLocalFileSystem.java:82)
at org.apache.hadoop.fs.RawLocalFileSystem.deprecatedGetFileStatus(RawLocalFileSystem.java:606)
at org.apache.hadoop.fs.RawLocalFileSystem.getFileLinkStatusInternal(RawLocalFileSystem.java:824)
at org.apache.hadoop.fs.RawLocalFileSystem.getFileStatus(RawLocalFileSystem.java:601)
at org.apache.hadoop.fs.FilterFileSystem.getFileStatus(FilterFileSystem.java:421)
at org.apache.hadoop.fs.FileSystem.exists(FileSystem.java:1426)
at org.apache.spark.sql.execution.streaming.StreamMetadata$.read(StreamMetadata.scala:51)
at org.apache.spark.sql.execution.streaming.StreamExecution.<init>(StreamExecution.scala:100)
at org.apache.spark.sql.streaming.StreamingQueryManager.createQuery(StreamingQueryManager.scala:232)
at org.apache.spark.sql.streaming.StreamingQueryManager.startQuery(StreamingQueryManager.scala:269)
at org.apache.spark.sql.streaming.DataStreamWriter.start(DataStreamWriter.scala:262)
```
Can easily replicate on spark standalone cluster by providing checkpoint location uri scheme anything other than "file://" and not overriding in config.
WorkAround --conf spark.hadoop.fs.defaultFS=s3a://somebucket or set it in sparkConf or spark-default.conf
## How was this patch tested?
existing ut
Author: uncleGen <hustyugm@gmail.com>
Closes#16815 from uncleGen/SPARK-19407.
## What changes were proposed in this pull request?
The current way of resolving `InsertIntoTable` and `CreateTable` is convoluted: sometimes we replace them with concrete implementation commands during analysis, sometimes during planning phase.
And the error checking logic is also a mess: we may put it in extended analyzer rules, or extended checking rules, or `CheckAnalysis`.
This PR simplifies the data source analysis:
1. `InsertIntoTable` and `CreateTable` are always unresolved and need to be replaced by concrete implementation commands during analysis.
2. The error checking logic is mainly in 2 rules: `PreprocessTableCreation` and `PreprocessTableInsertion`.
## How was this patch tested?
existing test.
Author: Wenchen Fan <wenchen@databricks.com>
Closes#16269 from cloud-fan/ddl.
## What changes were proposed in this pull request?
We've already upgraded parquet-mr to 1.8.2. This PR does some further cleanup by removing a workaround of PARQUET-686 and a hack due to PARQUET-363 and PARQUET-278. All three Parquet issues are fixed in parquet-mr 1.8.2.
## How was this patch tested?
Existing unit tests.
Author: Cheng Lian <lian@databricks.com>
Closes#16791 from liancheng/parquet-1.8.2-cleanup.
### What changes were proposed in this pull request?
So far, we allow users to create a table with an empty schema: `CREATE TABLE tab1`. This could break many code paths if we enable it. Thus, we should follow Hive to block it.
For Hive serde tables, some serde libraries require the specified schema and record it in the metastore. To get the list, we need to check `hive.serdes.using.metastore.for.schema,` which contains a list of serdes that require user-specified schema. The default values are
- org.apache.hadoop.hive.ql.io.orc.OrcSerde
- org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe
- org.apache.hadoop.hive.serde2.columnar.ColumnarSerDe
- org.apache.hadoop.hive.serde2.dynamic_type.DynamicSerDe
- org.apache.hadoop.hive.serde2.MetadataTypedColumnsetSerDe
- org.apache.hadoop.hive.serde2.columnar.LazyBinaryColumnarSerDe
- org.apache.hadoop.hive.ql.io.parquet.serde.ParquetHiveSerDe
- org.apache.hadoop.hive.serde2.lazybinary.LazyBinarySerDe
### How was this patch tested?
Added test cases for both Hive and data source tables
Author: gatorsmile <gatorsmile@gmail.com>
Closes#16636 from gatorsmile/fixEmptyTableSchema.
## What changes were proposed in this pull request?
This PR proposes to
- remove unused `findTightestCommonType` in `TypeCoercion` as suggested in https://github.com/apache/spark/pull/16777#discussion_r99283834
- rename `findTightestCommonTypeOfTwo ` to `findTightestCommonType`.
- fix comments accordingly
The usage was removed while refactoring/fixing in several JIRAs such as SPARK-16714, SPARK-16735 and SPARK-16646
## How was this patch tested?
Existing tests.
Author: hyukjinkwon <gurwls223@gmail.com>
Closes#16786 from HyukjinKwon/SPARK-19446.
## What changes were proposed in this pull request?
There is a metadata introduced before to mark the optional columns in merged Parquet schema for filter predicate pushdown. As we upgrade to Parquet 1.8.2 which includes the fix for the pushdown of optional columns, we don't need this metadata now.
## How was this patch tested?
Jenkins tests.
Please review http://spark.apache.org/contributing.html before opening a pull request.
Author: Liang-Chi Hsieh <viirya@gmail.com>
Closes#16756 from viirya/remove-optional-metadata.
## What changes were proposed in this pull request?
1, add the multi-cols support based on current private api
2, add the multi-cols support to pyspark
## How was this patch tested?
unit tests
Author: Zheng RuiFeng <ruifengz@foxmail.com>
Author: Ruifeng Zheng <ruifengz@foxmail.com>
Closes#12135 from zhengruifeng/quantile4multicols.
## What changes were proposed in this pull request?
This PR deduplicates arguments, `url` and `table` in `JdbcUtils` with `JDBCOptions`.
It avoids to use duplicated arguments, for example, as below:
from
```scala
val jdbcOptions = new JDBCOptions(url, table, map)
JdbcUtils.saveTable(ds, url, table, jdbcOptions)
```
to
```scala
val jdbcOptions = new JDBCOptions(url, table, map)
JdbcUtils.saveTable(ds, jdbcOptions)
```
## How was this patch tested?
Running unit test in `JdbcSuite`/`JDBCWriteSuite`
Building with Scala 2.10 as below:
```
./dev/change-scala-version.sh 2.10
./build/mvn -Pyarn -Phadoop-2.4 -Dscala-2.10 -DskipTests clean package
```
Author: hyukjinkwon <gurwls223@gmail.com>
Closes#16753 from HyukjinKwon/SPARK-19296.
## What changes were proposed in this pull request?
This PR proposes three things as below:
- Support LaTex inline-formula, `\( ... \)` in Scala API documentation
It seems currently,
```
\( ... \)
```
are rendered as they are, for example,
<img width="345" alt="2017-01-30 10 01 13" src="https://cloud.githubusercontent.com/assets/6477701/22423960/ab37d54a-e737-11e6-9196-4f6229c0189c.png">
It seems mistakenly more backslashes were added.
- Fix warnings Scaladoc/Javadoc generation
This PR fixes t two types of warnings as below:
```
[warn] .../spark/sql/catalyst/src/main/scala/org/apache/spark/sql/Row.scala:335: Could not find any member to link for "UnsupportedOperationException".
[warn] /**
[warn] ^
```
```
[warn] .../spark/sql/core/src/main/scala/org/apache/spark/sql/internal/VariableSubstitution.scala:24: Variable var undefined in comment for class VariableSubstitution in class VariableSubstitution
[warn] * `${var}`, `${system:var}` and `${env:var}`.
[warn] ^
```
- Fix Javadoc8 break
```
[error] .../spark/mllib/target/java/org/apache/spark/ml/PredictionModel.java:7: error: reference not found
[error] * E.g., {link VectorUDT} for vector features.
[error] ^
[error] .../spark/mllib/target/java/org/apache/spark/ml/PredictorParams.java:12: error: reference not found
[error] * E.g., {link VectorUDT} for vector features.
[error] ^
[error] .../spark/mllib/target/java/org/apache/spark/ml/Predictor.java:10: error: reference not found
[error] * E.g., {link VectorUDT} for vector features.
[error] ^
[error] .../spark/sql/hive/target/java/org/apache/spark/sql/hive/HiveAnalysis.java:5: error: reference not found
[error] * Note that, this rule must be run after {link PreprocessTableInsertion}.
[error] ^
```
## How was this patch tested?
Manually via `sbt unidoc` and `jeykil build`.
Author: hyukjinkwon <gurwls223@gmail.com>
Closes#16741 from HyukjinKwon/warn-and-break.
## What changes were proposed in this pull request?
In StructuredStreaming, if a new trigger was skipped because no new data arrived, we suddenly report nothing for the metrics `stateOperator`. We could however easily report the metrics from `lastExecution` to ensure continuity of metrics.
## How was this patch tested?
Regression test in `StreamingQueryStatusAndProgressSuite`
Author: Burak Yavuz <brkyvz@gmail.com>
Closes#16716 from brkyvz/state-agg.
## What changes were proposed in this pull request?
After https://github.com/apache/spark/pull/16552 , `CreateHiveTableAsSelectCommand` becomes very similar to `CreateDataSourceTableAsSelectCommand`, and we can further simplify it by only creating table in the table-not-exist branch.
This PR also adds hive provider checking in DataStream reader/writer, which is missed in #16552
## How was this patch tested?
N/A
Author: Wenchen Fan <wenchen@databricks.com>
Closes#16693 from cloud-fan/minor.
## What changes were proposed in this pull request?
This pr added a variable for a UDF name in `ScalaUDF`.
Then, if the variable filled, `DataFrame#explain` prints the name.
## How was this patch tested?
Added a test in `UDFSuite`.
Author: Takeshi YAMAMURO <linguin.m.s@gmail.com>
Closes#16707 from maropu/SPARK-19338.
## What changes were proposed in this pull request?
As of Spark 2.1, Spark SQL assumes the machine timezone for datetime manipulation, which is bad if users are not in the same timezones as the machines, or if different users have different timezones.
We should introduce a session local timezone setting that is used for execution.
An explicit non-goal is locale handling.
### Semantics
Setting the session local timezone means that the timezone-aware expressions listed below should use the timezone to evaluate values, and also it should be used to convert (cast) between string and timestamp or between timestamp and date.
- `CurrentDate`
- `CurrentBatchTimestamp`
- `Hour`
- `Minute`
- `Second`
- `DateFormatClass`
- `ToUnixTimestamp`
- `UnixTimestamp`
- `FromUnixTime`
and below are implicitly timezone-aware through cast from timestamp to date:
- `DayOfYear`
- `Year`
- `Quarter`
- `Month`
- `DayOfMonth`
- `WeekOfYear`
- `LastDay`
- `NextDay`
- `TruncDate`
For example, if you have timestamp `"2016-01-01 00:00:00"` in `GMT`, the values evaluated by some of timezone-aware expressions are:
```scala
scala> val df = Seq(new java.sql.Timestamp(1451606400000L)).toDF("ts")
df: org.apache.spark.sql.DataFrame = [ts: timestamp]
scala> df.selectExpr("cast(ts as string)", "year(ts)", "month(ts)", "dayofmonth(ts)", "hour(ts)", "minute(ts)", "second(ts)").show(truncate = false)
+-------------------+----------------------+-----------------------+----------------------------+--------+----------+----------+
|ts |year(CAST(ts AS DATE))|month(CAST(ts AS DATE))|dayofmonth(CAST(ts AS DATE))|hour(ts)|minute(ts)|second(ts)|
+-------------------+----------------------+-----------------------+----------------------------+--------+----------+----------+
|2016-01-01 00:00:00|2016 |1 |1 |0 |0 |0 |
+-------------------+----------------------+-----------------------+----------------------------+--------+----------+----------+
```
whereas setting the session local timezone to `"PST"`, they are:
```scala
scala> spark.conf.set("spark.sql.session.timeZone", "PST")
scala> df.selectExpr("cast(ts as string)", "year(ts)", "month(ts)", "dayofmonth(ts)", "hour(ts)", "minute(ts)", "second(ts)").show(truncate = false)
+-------------------+----------------------+-----------------------+----------------------------+--------+----------+----------+
|ts |year(CAST(ts AS DATE))|month(CAST(ts AS DATE))|dayofmonth(CAST(ts AS DATE))|hour(ts)|minute(ts)|second(ts)|
+-------------------+----------------------+-----------------------+----------------------------+--------+----------+----------+
|2015-12-31 16:00:00|2015 |12 |31 |16 |0 |0 |
+-------------------+----------------------+-----------------------+----------------------------+--------+----------+----------+
```
Notice that even if you set the session local timezone, it affects only in `DataFrame` operations, neither in `Dataset` operations, `RDD` operations nor in `ScalaUDF`s. You need to properly handle timezone by yourself.
### Design of the fix
I introduced an analyzer to pass session local timezone to timezone-aware expressions and modified DateTimeUtils to take the timezone argument.
## How was this patch tested?
Existing tests and added tests for timezone aware expressions.
Author: Takuya UESHIN <ueshin@happy-camper.st>
Closes#16308 from ueshin/issues/SPARK-18350.
## What changes were proposed in this pull request?
In CachedTableSuite, we are not setting up the test data at the beginning. Some tests fail while trying to run individually. When running the entire suite they run fine.
Here are some of the tests that fail -
- test("SELECT star from cached table")
- test("Self-join cached")
As part of this simplified a couple of tests by calling a support method to count the number of
InMemoryRelations.
## How was this patch tested?
Ran the failing tests individually.
Author: Dilip Biswal <dbiswal@us.ibm.com>
Closes#16688 from dilipbiswal/cachetablesuite_simple.
## What changes were proposed in this pull request?
Similar to SPARK-15165, codegen is in danger of arbitrary code injection. The root cause is how variable names are created by codegen.
In GenerateExec#codeGenAccessor, a variable name is created like as follows.
```
val value = ctx.freshName(name)
```
The variable `value` is named based on the value of the variable `name` and the value of `name` is from schema given by users so an attacker can attack with queries like as follows.
```
SELECT inline(array(cast(struct(1) AS struct<`=new Object() { {f();} public void f() {throw new RuntimeException("This exception is injected.");} public int x;}.x`:int>)))
```
In the example above, a RuntimeException is thrown but an attacker can replace it with arbitrary code.
## How was this patch tested?
Added a new test case.
Author: Kousuke Saruta <sarutak@oss.nttdata.co.jp>
Closes#16681 from sarutak/SPARK-19334.
## What changes were proposed in this pull request?
After [SPARK-19107](https://issues.apache.org/jira/browse/SPARK-19107), we now can treat hive as a data source and create hive tables with DataFrameWriter and Catalog. However, the support is not completed, there are still some cases we do not support.
This PR implement:
DataFrameWriter.saveAsTable work with hive format with append mode
## How was this patch tested?
unit test added
Author: windpiger <songjun@outlook.com>
Closes#16552 from windpiger/saveAsTableWithHiveAppend.
## What changes were proposed in this pull request?
As adaptive query execution may change the number of partitions in different batches, it may break streaming queries. Hence, we should disallow this feature in Structured Streaming.
## How was this patch tested?
`test("SPARK-19268: Adaptive query execution should be disallowed")`.
Author: Shixiong Zhu <shixiong@databricks.com>
Closes#16683 from zsxwing/SPARK-19268.
## What changes were proposed in this pull request?
Currently, running the codes in Java
```java
spark.udf().register("inc", new UDF1<Long, Long>() {
Override
public Long call(Long i) {
return i + 1;
}
}, DataTypes.LongType);
spark.range(10).toDF("x").createOrReplaceTempView("tmp");
Row result = spark.sql("SELECT inc(x) FROM tmp GROUP BY inc(x)").head();
Assert.assertEquals(7, result.getLong(0));
```
fails as below:
```
org.apache.spark.sql.AnalysisException: expression 'tmp.`x`' is neither present in the group by, nor is it an aggregate function. Add to group by or wrap in first() (or first_value) if you don't care which value you get.;;
Aggregate [UDF(x#19L)], [UDF(x#19L) AS UDF(x)#23L]
+- SubqueryAlias tmp, `tmp`
+- Project [id#16L AS x#19L]
+- Range (0, 10, step=1, splits=Some(8))
at org.apache.spark.sql.catalyst.analysis.CheckAnalysis$class.failAnalysis(CheckAnalysis.scala:40)
at org.apache.spark.sql.catalyst.analysis.Analyzer.failAnalysis(Analyzer.scala:57)
```
The root cause is because we were creating the function every time when it needs to build as below:
```scala
scala> def inc(i: Int) = i + 1
inc: (i: Int)Int
scala> (inc(_: Int)).hashCode
res15: Int = 1231799381
scala> (inc(_: Int)).hashCode
res16: Int = 2109839984
scala> (inc(_: Int)) == (inc(_: Int))
res17: Boolean = false
```
This seems leading to the comparison failure between `ScalaUDF`s created from Java UDF API, for example, in `Expression.semanticEquals`.
In case of Scala one, it seems already fine.
Both can be tested easily as below if any reviewer is more comfortable with Scala:
```scala
val df = Seq((1, 10), (2, 11), (3, 12)).toDF("x", "y")
val javaUDF = new UDF1[Int, Int] {
override def call(i: Int): Int = i + 1
}
// spark.udf.register("inc", javaUDF, IntegerType) // Uncomment this for Java API
// spark.udf.register("inc", (i: Int) => i + 1) // Uncomment this for Scala API
df.createOrReplaceTempView("tmp")
spark.sql("SELECT inc(y) FROM tmp GROUP BY inc(y)").show()
```
## How was this patch tested?
Unit test in `JavaUDFSuite.java` and `./dev/lint-java`.
Author: hyukjinkwon <gurwls223@gmail.com>
Closes#16553 from HyukjinKwon/SPARK-9435.
## What changes were proposed in this pull request?
Hive will expand the view text, so it needs 2 fields: originalText and viewText. Since we don't expand the view text, but only add table properties, perhaps only a single field `viewText` is enough in CatalogTable.
This PR brought in the following changes:
1. Remove the param `viewOriginalText` from `CatalogTable`;
2. Update the output of command `DescribeTableCommand`.
## How was this patch tested?
Tested by exsiting test cases, also updated the failed test cases.
Author: jiangxingbo <jiangxb1987@gmail.com>
Closes#16679 from jiangxb1987/catalogTable.
## What changes were proposed in this pull request?
To implement DDL commands, we added several analyzer rules in sql/hive module to analyze DDL related plans. However, our `Analyzer` currently only have one extending interface: `extendedResolutionRules`, which defines extra rules that will be run together with other rules in the resolution batch, and doesn't fit DDL rules well, because:
1. DDL rules may do some checking and normalization, but we may do it many times as the resolution batch will run rules again and again, until fixed point, and it's hard to tell if a DDL rule has already done its checking and normalization. It's fine because DDL rules are idempotent, but it's bad for analysis performance
2. some DDL rules may depend on others, and it's pretty hard to write `if` conditions to guarantee the dependencies. It will be good if we have a batch which run rules in one pass, so that we can guarantee the dependencies by rules order.
This PR adds a new extending interface in `Analyzer`: `postHocResolutionRules`, which defines rules that will be run only once in a batch runs right after the resolution batch.
## How was this patch tested?
existing tests
Author: Wenchen Fan <wenchen@databricks.com>
Closes#16645 from cloud-fan/analyzer.
## What changes were proposed in this pull request?
when we append data to a existed partitioned datasource table, the InsertIntoHadoopFsRelationCommand.getCustomPartitionLocations currently
return the same location with Hive default, it should return None.
## How was this patch tested?
Author: windpiger <songjun@outlook.com>
Closes#16642 from windpiger/appendSchema.
## What changes were proposed in this pull request?
As I pointed out in https://github.com/apache/spark/pull/15807#issuecomment-259143655 , the current subexpression elimination framework has a problem, it always evaluates all common subexpressions at the beginning, even they are inside conditional expressions and may not be accessed.
Ideally we should implement it like scala lazy val, so we only evaluate it when it gets accessed at lease once. https://github.com/apache/spark/issues/15837 tries this approach, but it seems too complicated and may introduce performance regression.
This PR simply stops common subexpression elimination for conditional expressions, with some cleanup.
## How was this patch tested?
regression test
Author: Wenchen Fan <wenchen@databricks.com>
Closes#16659 from cloud-fan/codegen.
### What changes were proposed in this pull request?
It is weird to create Hive source tables when using InMemoryCatalog. We are unable to operate it. This PR is to block users to create Hive source tables.
### How was this patch tested?
Fixed the test cases
Author: gatorsmile <gatorsmile@gmail.com>
Closes#16587 from gatorsmile/blockHiveTable.
## What changes were proposed in this pull request?
This PR refactors CSV read path to be consistent with JSON data source. It makes the methods in classes have consistent arguments with JSON ones.
`UnivocityParser` and `JacksonParser`
``` scala
private[csv] class UnivocityParser(
schema: StructType,
requiredSchema: StructType,
options: CSVOptions) extends Logging {
...
def parse(input: String): Seq[InternalRow] = {
...
```
``` scala
class JacksonParser(
schema: StructType,
columnNameOfCorruptRecord: String,
options: JSONOptions) extends Logging {
...
def parse(input: String): Option[InternalRow] = {
...
```
These allow parsing an iterator (`String` to `InternalRow`) as below for both JSON and CSV:
```scala
iter.flatMap(parser.parse)
```
## How was this patch tested?
Existing tests should cover this.
Author: hyukjinkwon <gurwls223@gmail.com>
Closes#16669 from HyukjinKwon/SPARK-16101-read.
## What changes were proposed in this pull request?
For data source tables, we will always reorder the specified table schema, or the query in CTAS, to put partition columns at the end. e.g. `CREATE TABLE t(a int, b int, c int, d int) USING parquet PARTITIONED BY (d, b)` will create a table with schema `<a, c, d, b>`
Hive serde tables don't have this problem before, because its CREATE TABLE syntax specifies data schema and partition schema individually.
However, after we unifed the CREATE TABLE syntax, Hive serde table also need to do the reorder. This PR puts the reorder logic in a analyzer rule, which works with both data source tables and Hive serde tables.
## How was this patch tested?
new regression test
Author: Wenchen Fan <wenchen@databricks.com>
Closes#16655 from cloud-fan/schema.
## What changes were proposed in this pull request?
JDBC read is failing with NPE due to missing null value check for array data type if the source table has null values in the array type column. For null values Resultset.getArray() returns null.
This PR adds null safe check to the Resultset.getArray() value before invoking method on the Array object.
## How was this patch tested?
Updated the PostgresIntegration test suite to test null values. Ran docker integration tests on my laptop.
Author: sureshthalamati <suresh.thalamati@gmail.com>
Closes#15192 from sureshthalamati/jdbc_array_null_fix-SPARK-14536.
## What changes were proposed in this pull request?
This PR refactors CSV write path to be consistent with JSON data source.
This PR makes the methods in classes have consistent arguments with JSON ones.
- `UnivocityGenerator` and `JacksonGenerator`
``` scala
private[csv] class UnivocityGenerator(
schema: StructType,
writer: Writer,
options: CSVOptions = new CSVOptions(Map.empty[String, String])) {
...
def write ...
def close ...
def flush ...
```
``` scala
private[sql] class JacksonGenerator(
schema: StructType,
writer: Writer,
options: JSONOptions = new JSONOptions(Map.empty[String, String])) {
...
def write ...
def close ...
def flush ...
```
- This PR also makes the classes put in together in a consistent manner with JSON.
- `CsvFileFormat`
``` scala
CsvFileFormat
CsvOutputWriter
```
- `JsonFileFormat`
``` scala
JsonFileFormat
JsonOutputWriter
```
## How was this patch tested?
Existing tests should cover this.
Author: hyukjinkwon <gurwls223@gmail.com>
Closes#16496 from HyukjinKwon/SPARK-16101-write.
## What changes were proposed in this pull request?
There is a race condition when stopping StateStore which makes `StateStoreSuite.maintenance` flaky. `StateStore.stop` doesn't wait for the running task to finish, and an out-of-date task may fail `doMaintenance` and cancel the new task. Here is a reproducer: dde1b5b106
This PR adds MaintenanceTask to eliminate the race condition.
## How was this patch tested?
Jenkins
Author: Shixiong Zhu <shixiong@databricks.com>
Author: Tathagata Das <tathagata.das1565@gmail.com>
Closes#16627 from zsxwing/SPARK-19267.
## What changes were proposed in this pull request?
When we query a table with a filter on partitioned columns, we will push the partition filter to the metastore to get matched partitions directly.
In `HiveExternalCatalog.listPartitionsByFilter`, we assume the column names in partition filter are already normalized and we don't need to consider case sensitivity. However, `HiveTableScanExec` doesn't follow this assumption. This PR fixes it.
## How was this patch tested?
new regression test
Author: Wenchen Fan <wenchen@databricks.com>
Closes#16647 from cloud-fan/bug.
## What changes were proposed in this pull request?
This PR refactors the code generation part to get data from `ColumnarVector` and `ColumnarBatch` by using a trait `ColumnarBatchScan` for ease of reuse. This is because this part will be reused by several components (e.g. parquet reader, Dataset.cache, and others) since `ColumnarBatch` will be first citizen.
This PR is a part of https://github.com/apache/spark/pull/15219. In advance, this PR makes the code generation for `ColumnarVector` and `ColumnarBatch` reuseable as a trait. In general, this is very useful for other components from the reuseability view, too.
## How was this patch tested?
tested existing test suites
Author: Kazuaki Ishizaki <ishizaki@jp.ibm.com>
Closes#15467 from kiszk/columnarrefactor.
## What changes were proposed in this pull request?
The initial shouldFilterOut() method invocation filter the root path name(table name in the intial call) and remove if it contains _. I moved the check one level below, so it first list files/directories in the given root path and then apply filter.
(Please fill in changes proposed in this fix)
## How was this patch tested?
Added new test case for this scenario
(Please explain how this patch was tested. E.g. unit tests, integration tests, manual tests)
(If this patch involves UI changes, please attach a screenshot; otherwise, remove this)
Please review http://spark.apache.org/contributing.html before opening a pull request.
Author: jayadevanmurali <jayadevan.m@tcs.com>
Author: jayadevan <jayadevan.m@tcs.com>
Closes#16635 from jayadevanmurali/branch-0.1-SPARK-19059.
## What changes were proposed in this pull request?
We have a table relation plan cache in `HiveMetastoreCatalog`, which caches a lot of things: file status, resolved data source, inferred schema, etc.
However, it doesn't make sense to limit this cache with hive support, we should move it to SQL core module so that users can use this cache without hive support.
It can also reduce the size of `HiveMetastoreCatalog`, so that it's easier to remove it eventually.
main changes:
1. move the table relation cache to `SessionCatalog`
2. `SessionCatalog.lookupRelation` will return `SimpleCatalogRelation` and the analyzer will convert it to `LogicalRelation` or `MetastoreRelation` later, then `HiveSessionCatalog` doesn't need to override `lookupRelation` anymore
3. `FindDataSourceTable` will read/write the table relation cache.
## How was this patch tested?
existing tests.
Author: Wenchen Fan <wenchen@databricks.com>
Closes#16621 from cloud-fan/plan-cache.
## What changes were proposed in this pull request?
We should call `StateStore.abort()` when there should be any error before the store is committed.
## How was this patch tested?
Manually.
Author: Liwei Lin <lwlin7@gmail.com>
Closes#16547 from lw-lin/append-filter.
## What changes were proposed in this pull request?
On CREATE/ALTER a view, it's no longer needed to generate a SQL text string from the LogicalPlan, instead we store the SQL query text、the output column names of the query plan, and current database to CatalogTable. Permanent views created by this approach can be resolved by current view resolution approach.
The main advantage includes:
1. If you update an underlying view, the current view also gets updated;
2. That gives us a change to get ride of SQL generation for operators.
Major changes of this PR:
1. Generate the view-specific properties(e.g. view default database, view query output column names) during permanent view creation and store them as properties in the CatalogTable;
2. Update the commands `CreateViewCommand` and `AlterViewAsCommand`, get rid of SQL generation from them.
## How was this patch tested?
Existing tests.
Author: jiangxingbo <jiangxb1987@gmail.com>
Closes#16613 from jiangxb1987/view-write-path.
## What changes were proposed in this pull request?
Inserting data into Hive tables has its own implementation that is distinct from data sources: `InsertIntoHiveTable`, `SparkHiveWriterContainer` and `SparkHiveDynamicPartitionWriterContainer`.
Note that one other major difference is that data source tables write directly to the final destination without using some staging directory, and then Spark itself adds the partitions/tables to the catalog. Hive tables actually write to some staging directory, and then call Hive metastore's loadPartition/loadTable function to load those data in. So we still need to keep `InsertIntoHiveTable` to put this special logic. In the future, we should think of writing to the hive table location directly, so that we don't need to call `loadTable`/`loadPartition` at the end and remove `InsertIntoHiveTable`.
This PR removes `SparkHiveWriterContainer` and `SparkHiveDynamicPartitionWriterContainer`, and create a `HiveFileFormat` to implement the write logic. In the future, we should also implement the read logic in `HiveFileFormat`.
## How was this patch tested?
existing tests
Author: Wenchen Fan <wenchen@databricks.com>
Closes#16517 from cloud-fan/insert-hive.
## What changes were proposed in this pull request?
Added outer_explode, outer_posexplode, outer_inline functions and expressions.
Some bug fixing in GenerateExec.scala for CollectionGenerator. Previously it was not correctly handling the case of outer with empty collections, only with nulls.
## How was this patch tested?
New tests added to GeneratorFunctionSuite
Author: Bogdan Raducanu <bogdan.rdc@gmail.com>
Closes#16608 from bogdanrdc/SPARK-13721.
## What changes were proposed in this pull request?
In append mode, we check whether the schema of the write is compatible with the schema of the existing data. It can be a significant performance issue in cloud environment to find the existing schema for files. This patch removes the check.
Note that for catalog tables, we always do the check, as discussed in https://github.com/apache/spark/pull/16339#discussion_r96208357
## How was this patch tested?
N/A
Closes#16339.
Author: Reynold Xin <rxin@databricks.com>
Closes#16622 from rxin/SPARK-18917.
