## What changes were proposed in this pull request?
This pull-request exclusively includes the class splitting feature described in #16648. When code for a given class would grow beyond 1600k bytes, a private, nested sub-class is generated into which subsequent functions are inlined. Additional sub-classes are generated as the code threshold is met subsequent times. This code includes 3 changes:
1. Includes helper maps, lists, and functions for keeping track of sub-classes during code generation (included in the `CodeGenerator` class). These helper functions allow nested classes and split functions to be initialized/declared/inlined to the appropriate locations in the various projection classes.
2. Changes `addNewFunction` to return a string to support instances where a split function is inlined to a nested class and not the outer class (and so must be invoked using the class-qualified name). Uses of `addNewFunction` throughout the codebase are modified so that the returned name is properly used.
3. Removes instances of the `this` keyword when used on data inside generated classes. All state declared in the outer class is by default global and accessible to the nested classes. However, if a reference to global state in a nested class is prepended with the `this` keyword, it would attempt to reference state belonging to the nested class (which would not exist), rather than the correct variable belonging to the outer class.
## How was this patch tested?
Added a test case to the `GeneratedProjectionSuite` that increases the number of columns tested in various projections to a threshold that would previously have triggered a `JaninoRuntimeException` for the Constant Pool.
Note: This PR does not address the second Constant Pool issue with code generation (also mentioned in #16648): excess global mutable state. A second PR may be opened to resolve that issue.
Author: ALeksander Eskilson <alek.eskilson@cerner.com>
Closes#18075 from bdrillard/class_splitting_only.
## What changes were proposed in this pull request?
It is really painful to not have configs in logical plan and expressions. We had to add all sorts of hacks (e.g. pass SQLConf explicitly in functions). This patch exposes SQLConf in logical plan, using a thread local variable and a getter closure that's set once there is an active SparkSession.
The implementation is a bit of a hack, since we didn't anticipate this need in the beginning (config was only exposed in physical plan). The implementation is described in `SQLConf.get`.
In terms of future work, we should follow up to clean up CBO (remove the need for passing in config).
## How was this patch tested?
Updated relevant tests for constraint propagation.
Author: Reynold Xin <rxin@databricks.com>
Closes#18299 from rxin/SPARK-21092.
## What changes were proposed in this pull request?
Since `stack` function generates a table with nullable columns, it should allow mixed null values.
```scala
scala> sql("select stack(3, 1, 2, 3)").printSchema
root
|-- col0: integer (nullable = true)
scala> sql("select stack(3, 1, 2, null)").printSchema
org.apache.spark.sql.AnalysisException: cannot resolve 'stack(3, 1, 2, NULL)' due to data type mismatch: Argument 1 (IntegerType) != Argument 3 (NullType); line 1 pos 7;
```
## How was this patch tested?
Pass the Jenkins with a new test case.
Author: Dongjoon Hyun <dongjoon@apache.org>
Closes#17251 from dongjoon-hyun/SPARK-19910.
## What changes were proposed in this pull request?
This patch fixes a bug that can cause NullPointerException in LikeSimplification, when the pattern for like is null.
## How was this patch tested?
Added a new unit test case in LikeSimplificationSuite.
Author: Reynold Xin <rxin@databricks.com>
Closes#18273 from rxin/SPARK-21059.
The PR contains a tiny change to fix the way Spark parses string literals into timestamps. Currently, some timestamps that contain nanoseconds are corrupted during the conversion from internal UTF8Strings into the internal representation of timestamps.
Consider the following example:
```
spark.sql("SELECT cast('2015-01-02 00:00:00.000000001' as TIMESTAMP)").show(false)
+------------------------------------------------+
|CAST(2015-01-02 00:00:00.000000001 AS TIMESTAMP)|
+------------------------------------------------+
|2015-01-02 00:00:00.000001 |
+------------------------------------------------+
```
The fix was tested with existing tests. Also, there is a new test to cover cases that did not work previously.
Author: aokolnychyi <anton.okolnychyi@sap.com>
Closes#18252 from aokolnychyi/spark-17914.
## What changes were proposed in this pull request?
add test case to MathExpressionsSuite as #17906
## How was this patch tested?
unit test cases
Author: liuxian <liu.xian3@zte.com.cn>
Closes#18082 from 10110346/wip-lx-0524.
## What changes were proposed in this pull request?
Currently, hive's stats are read into `CatalogStatistics`, while spark's stats are also persisted through `CatalogStatistics`. As a result, hive's stats can be unexpectedly propagated into spark' stats.
For example, for a catalog table, we read stats from hive, e.g. "totalSize" and put it into `CatalogStatistics`. Then, by using "ALTER TABLE" command, we will store the stats in `CatalogStatistics` into metastore as spark's stats (because we don't know whether it's from spark or not). But spark's stats should be only generated by "ANALYZE" command. This is unexpected from this command.
Secondly, now that we have spark's stats in metastore, after inserting new data, although hive updated "totalSize" in metastore, we still cannot get the right `sizeInBytes` in `CatalogStatistics`, because we respect spark's stats (should not exist) over hive's stats.
A running example is shown in [JIRA](https://issues.apache.org/jira/browse/SPARK-21031).
To fix this, we add a new method `alterTableStats` to store spark's stats, and let `alterTable` keep existing stats.
## How was this patch tested?
Added new tests.
Author: Zhenhua Wang <wzh_zju@163.com>
Closes#18248 from wzhfy/separateHiveStats.
## What changes were proposed in this pull request?
add more datatype for some unit tests
## How was this patch tested?
unit tests
Author: liuxian <liu.xian3@zte.com.cn>
Closes#17880 from 10110346/wip_lx_0506.
### What changes were proposed in this pull request?
The precision and scale of decimal values are wrong when the input is BigDecimal between -1.0 and 1.0.
The BigDecimal's precision is the digit count starts from the leftmost nonzero digit based on the [JAVA's BigDecimal definition](https://docs.oracle.com/javase/7/docs/api/java/math/BigDecimal.html). However, our Decimal decision follows the database decimal standard, which is the total number of digits, including both to the left and the right of the decimal point. Thus, this PR is to fix the issue by doing the conversion.
Before this PR, the following queries failed:
```SQL
select 1 > 0.0001
select floor(0.0001)
select ceil(0.0001)
```
### How was this patch tested?
Added test cases.
Author: Xiao Li <gatorsmile@gmail.com>
Closes#18244 from gatorsmile/bigdecimal.
### What changes were proposed in this pull request?
Currently, the unquoted string of a function identifier is being used as the function identifier in the function registry. This could cause the incorrect the behavior when users use `.` in the function names. This PR is to take the `FunctionIdentifier` as the identifier in the function registry.
- Add one new function `createOrReplaceTempFunction` to `FunctionRegistry`
```Scala
final def createOrReplaceTempFunction(name: String, builder: FunctionBuilder): Unit
```
### How was this patch tested?
Add extra test cases to verify the inclusive bug fixes.
Author: Xiao Li <gatorsmile@gmail.com>
Author: gatorsmile <gatorsmile@gmail.com>
Closes#18142 from gatorsmile/fuctionRegistry.
## What changes were proposed in this pull request?
Removed a duplicate case in "SPARK-20854: select hint syntax with expressions"
## How was this patch tested?
Existing tests.
Author: Bogdan Raducanu <bogdan@databricks.com>
Closes#18217 from bogdanrdc/SPARK-20854-2.
## What changes were proposed in this pull request?
`HintInfo.isBroadcastable` is actually not an accurate name, it's used to force the planner to broadcast a plan no matter what the data size is, via the hint mechanism. I think `forceBroadcast` is a better name.
And `isBroadcastable` only have 2 possible values: `Some(true)` and `None`, so we can just use boolean type for it.
## How was this patch tested?
existing tests.
Author: Wenchen Fan <wenchen@databricks.com>
Closes#18189 from cloud-fan/stats.
## What changes were proposed in this pull request?
SQL hint syntax:
* support expressions such as strings, numbers, etc. instead of only identifiers as it is currently.
* support multiple hints, which was missing compared to the DataFrame syntax.
DataFrame API:
* support any parameters in DataFrame.hint instead of just strings
## How was this patch tested?
Existing tests. New tests in PlanParserSuite. New suite DataFrameHintSuite.
Author: Bogdan Raducanu <bogdan@databricks.com>
Closes#18086 from bogdanrdc/SPARK-20854.
## What changes were proposed in this pull request?
Add build-int SQL function - UUID.
## How was this patch tested?
unit tests
Author: Yuming Wang <wgyumg@gmail.com>
Closes#18136 from wangyum/SPARK-20910.
## What changes were proposed in this pull request?
We changed the parser to reject unaliased subqueries in the FROM clause in SPARK-20690. However, the error message that we now give isn't very helpful:
scala> sql("""SELECT x FROM (SELECT 1 AS x)""")
org.apache.spark.sql.catalyst.parser.ParseException:
mismatched input 'FROM' expecting {<EOF>, 'WHERE', 'GROUP', 'ORDER', 'HAVING', 'LIMIT', 'LATERAL', 'WINDOW', 'UNION', 'EXCEPT', 'MINUS', 'INTERSECT', 'SORT', 'CLUSTER', 'DISTRIBUTE'}(line 1, pos 9)
We should modify the parser to throw a more clear error for such queries:
scala> sql("""SELECT x FROM (SELECT 1 AS x)""")
org.apache.spark.sql.catalyst.parser.ParseException:
The unaliased subqueries in the FROM clause are not supported.(line 1, pos 14)
## How was this patch tested?
Modified existing tests to reflect this change.
Author: Liang-Chi Hsieh <viirya@gmail.com>
Closes#18141 from viirya/SPARK-20916.
## What changes were proposed in this pull request?
Add build-int SQL function - DAYOFWEEK
## How was this patch tested?
unit tests
Author: Yuming Wang <wgyumg@gmail.com>
Closes#18134 from wangyum/SPARK-20909.
## What changes were proposed in this pull request?
This PR adds built-in SQL function `(REPLACE(<string_expression>, <search_string> [, <replacement_string>])`
`REPLACE()` return that string that is replaced all occurrences with given string.
## How was this patch tested?
added new test suites
Author: Kazuaki Ishizaki <ishizaki@jp.ibm.com>
Closes#18047 from kiszk/SPARK-20750.
## What changes were proposed in this pull request?
See class doc of `ConstantPropagation` for the approach used.
## How was this patch tested?
- Added unit tests
Author: Tejas Patil <tejasp@fb.com>
Closes#17993 from tejasapatil/SPARK-20758_const_propagation.
## What changes were proposed in this pull request?
This pr added parsing rules to support table column aliases in FROM clause.
## How was this patch tested?
Added tests in `PlanParserSuite`, `SQLQueryTestSuite`, and `PlanParserSuite`.
Author: Takeshi Yamamuro <yamamuro@apache.org>
Closes#18079 from maropu/SPARK-20841.
### What changes were proposed in this pull request?
In Cache manager, the plan matching should ignore Hint.
```Scala
val df1 = spark.range(10).join(broadcast(spark.range(10)))
df1.cache()
spark.range(10).join(spark.range(10)).explain()
```
The output plan of the above query shows that the second query is not using the cached data of the first query.
```
BroadcastNestedLoopJoin BuildRight, Inner
:- *Range (0, 10, step=1, splits=2)
+- BroadcastExchange IdentityBroadcastMode
+- *Range (0, 10, step=1, splits=2)
```
After the fix, the plan becomes
```
InMemoryTableScan [id#20L, id#23L]
+- InMemoryRelation [id#20L, id#23L], true, 10000, StorageLevel(disk, memory, deserialized, 1 replicas)
+- BroadcastNestedLoopJoin BuildRight, Inner
:- *Range (0, 10, step=1, splits=2)
+- BroadcastExchange IdentityBroadcastMode
+- *Range (0, 10, step=1, splits=2)
```
### How was this patch tested?
Added a test.
Author: Xiao Li <gatorsmile@gmail.com>
Closes#18131 from gatorsmile/HintCache.
## What changes were proposed in this pull request?
spark-sql>SELECT ceil(cast(12345.1233 as float));
spark-sql>12345
For this case, the result we expected is `12346`
spark-sql>SELECT floor(cast(-12345.1233 as float));
spark-sql>-12345
For this case, the result we expected is `-12346`
Because in `Ceil` or `Floor`, `inputTypes` has no FloatType, so it is converted to LongType.
## How was this patch tested?
After the modification:
spark-sql>SELECT ceil(cast(12345.1233 as float));
spark-sql>12346
spark-sql>SELECT floor(cast(-12345.1233 as float));
spark-sql>-12346
Author: liuxian <liu.xian3@zte.com.cn>
Closes#18103 from 10110346/wip-lx-0525-1.
## What changes were proposed in this pull request?
Add built-in SQL function `CH[A]R`:
For `CHR(bigint|double n)`, returns the ASCII character having the binary equivalent to `n`. If n is larger than 256 the result is equivalent to CHR(n % 256)
## How was this patch tested?
unit tests
Author: Yuming Wang <wgyumg@gmail.com>
Closes#18019 from wangyum/SPARK-20748.
## What changes were proposed in this pull request?
It is reported that there is performance downgrade when applying ML pipeline for dataset with many columns but few rows.
A big part of the performance downgrade comes from some operations (e.g., `select`) on DataFrame/Dataset which re-create new DataFrame/Dataset with a new `LogicalPlan`. The cost can be ignored in the usage of SQL, normally.
However, it's not rare to chain dozens of pipeline stages in ML. When the query plan grows incrementally during running those stages, the total cost spent on re-creation of DataFrame grows too. In particular, the `Analyzer` will go through the big query plan even most part of it is analyzed.
By eliminating part of the cost, the time to run the example code locally is reduced from about 1min to about 30 secs.
In particular, the time applying the pipeline locally is mostly spent on calling transform of the 137 `Bucketizer`s. Before the change, each call of `Bucketizer`'s transform can cost about 0.4 sec. So the total time spent on all `Bucketizer`s' transform is about 50 secs. After the change, each call only costs about 0.1 sec.
<del>We also make `boundEnc` as lazy variable to reduce unnecessary running time.</del>
### Performance improvement
The codes and datasets provided by Barry Becker to re-produce this issue and benchmark can be found on the JIRA.
Before this patch: about 1 min
After this patch: about 20 secs
## How was this patch tested?
Existing tests.
Please review http://spark.apache.org/contributing.html before opening a pull request.
Author: Liang-Chi Hsieh <viirya@gmail.com>
Closes#17770 from viirya/SPARK-20392.
## What changes were proposed in this pull request?
This is a follow-up to SPARK-20857 to move the broadcast hint from Statistics into a new HintInfo class, so we can be more flexible in adding new hints in the future.
## How was this patch tested?
Updated test cases to reflect the change.
Author: Reynold Xin <rxin@databricks.com>
Closes#18087 from rxin/SPARK-20867.
## What changes were proposed in this pull request?
This patch renames BroadcastHint to ResolvedHint (and Hint to UnresolvedHint) so the hint framework is more generic and would allow us to introduce other hint types in the future without introducing new hint nodes.
## How was this patch tested?
Updated test cases.
Author: Reynold Xin <rxin@databricks.com>
Closes#18072 from rxin/SPARK-20857.
## What changes were proposed in this pull request?
As srowen pointed in 609ba5f2b9 (commitcomment-22221259), the previous tests are not proper.
This follow-up is going to fix the tests.
## 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#18048 from viirya/SPARK-20399-follow-up.
### What changes were proposed in this pull request?
After we adding a new field `stats` into `CatalogTable`, we should not expose Hive-specific Stats metadata to `MetastoreRelation`. It complicates all the related codes. It also introduces a bug in `SHOW CREATE TABLE`. The statistics-related table properties should be skipped by `SHOW CREATE TABLE`, since it could be incorrect in the newly created table. See the Hive JIRA: https://issues.apache.org/jira/browse/HIVE-13792
Also fix the issue to fill Hive-generated RowCounts to our stats.
This PR is to handle Hive-specific Stats metadata in `HiveClientImpl`.
### How was this patch tested?
Added a few test cases.
Author: Xiao Li <gatorsmile@gmail.com>
Closes#14971 from gatorsmile/showCreateTableNew.
### What changes were proposed in this pull request?
Currently, we have a bug when we specify `IF NOT EXISTS` in `INSERT OVERWRITE` data source tables. For example, given a query:
```SQL
INSERT OVERWRITE TABLE $tableName partition (b=2, c=3) IF NOT EXISTS SELECT 9, 10
```
we will get the following error:
```
unresolved operator 'InsertIntoTable Relation[a#425,d#426,b#427,c#428] parquet, Map(b -> Some(2), c -> Some(3)), true, true;;
'InsertIntoTable Relation[a#425,d#426,b#427,c#428] parquet, Map(b -> Some(2), c -> Some(3)), true, true
+- Project [cast(9#423 as int) AS a#429, cast(10#424 as int) AS d#430]
+- Project [9 AS 9#423, 10 AS 10#424]
+- OneRowRelation$
```
This PR is to fix the issue to follow the behavior of Hive serde tables
> INSERT OVERWRITE will overwrite any existing data in the table or partition unless IF NOT EXISTS is provided for a partition
### How was this patch tested?
Modified an existing test case
Author: gatorsmile <gatorsmile@gmail.com>
Closes#18050 from gatorsmile/insertPartitionIfNotExists.
## What changes were proposed in this pull request?
spark-sql>SELECT ceil(1234567890123456);
1234567890123456
spark-sql>SELECT ceil(12345678901234567);
12345678901234568
spark-sql>SELECT ceil(123456789012345678);
123456789012345680
when the length of the getText is greater than 16. long to double will be precision loss.
but mysql handle the value is ok.
mysql> SELECT ceil(1234567890123456);
+------------------------+
| ceil(1234567890123456) |
+------------------------+
| 1234567890123456 |
+------------------------+
1 row in set (0.00 sec)
mysql> SELECT ceil(12345678901234567);
+-------------------------+
| ceil(12345678901234567) |
+-------------------------+
| 12345678901234567 |
+-------------------------+
1 row in set (0.00 sec)
mysql> SELECT ceil(123456789012345678);
+--------------------------+
| ceil(123456789012345678) |
+--------------------------+
| 123456789012345678 |
+--------------------------+
1 row in set (0.00 sec)
## How was this patch tested?
Supplement the unit test.
Author: caoxuewen <cao.xuewen@zte.com.cn>
Closes#18016 from heary-cao/ceil_long.
## What changes were proposed in this pull request?
spark-sql>select month("1582-09-28");
spark-sql>10
For this case, the expected result is 9, but it is 10.
spark-sql>select day("1582-04-18");
spark-sql>28
For this case, the expected result is 18, but it is 28.
when the date before "1582-10-04", the function of `month` and `day` return the value which is not we expected.
## How was this patch tested?
unit tests
Author: liuxian <liu.xian3@zte.com.cn>
Closes#17997 from 10110346/wip_lx_0516.
## What changes were proposed in this pull request?
GenerateUnsafeProjection.writeStructToBuffer() did not honor the assumption that the caller must make sure that a value is not null before using the getter. This could lead to various errors. This change fixes that behavior.
Example of code generated before:
```scala
/* 059 */ final UTF8String fieldName = value.getUTF8String(0);
/* 060 */ if (value.isNullAt(0)) {
/* 061 */ rowWriter1.setNullAt(0);
/* 062 */ } else {
/* 063 */ rowWriter1.write(0, fieldName);
/* 064 */ }
```
Example of code generated now:
```scala
/* 060 */ boolean isNull1 = value.isNullAt(0);
/* 061 */ UTF8String value1 = isNull1 ? null : value.getUTF8String(0);
/* 062 */ if (isNull1) {
/* 063 */ rowWriter1.setNullAt(0);
/* 064 */ } else {
/* 065 */ rowWriter1.write(0, value1);
/* 066 */ }
```
## How was this patch tested?
Adds GenerateUnsafeProjectionSuite.
Author: Ala Luszczak <ala@databricks.com>
Closes#18030 from ala/fix-generate-unsafe-projection.
## What changes were proposed in this pull request?
In the previous approach we used `aliasMap` to link an `Attribute` to the expression with potentially the form `f(a, b)`, but we only searched the `expressions` and `children.expressions` for this, which is not enough when an `Alias` may lies deep in the logical plan. In that case, we can't generate the valid equivalent constraint classes and thus we fail at preventing the recursive deductions.
We fix this problem by collecting all `Alias`s from the logical plan.
## How was this patch tested?
No additional test case is added, but do modified one test case to cover this situation.
Author: Xingbo Jiang <xingbo.jiang@databricks.com>
Closes#18020 from jiangxb1987/inferConstrants.
## What changes were proposed in this pull request?
We add missing attributes into Filter in Analyzer. But we shouldn't do it through subqueries like this:
select 1 from (select 1 from onerow t1 LIMIT 1) where t1.c1=1
This query works in current codebase. However, the outside where clause shouldn't be able to refer `t1.c1` attribute.
The root cause is we allow subqueries in FROM have no alias names previously, it is confusing and isn't supported by various databases such as MySQL, Postgres, Oracle. We shouldn't support it too.
## 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#17935 from viirya/SPARK-20690.
## What changes were proposed in this pull request?
Hive allows inserting data to bucketed table without guaranteeing bucketed and sorted-ness based on these two configs : `hive.enforce.bucketing` and `hive.enforce.sorting`.
What does this PR achieve ?
- Spark will disallow users from writing outputs to hive bucketed tables by default (given that output won't adhere with Hive's semantics).
- IF user still wants to write to hive bucketed table, the only resort is to use `hive.enforce.bucketing=false` and `hive.enforce.sorting=false` which means user does NOT care about bucketing guarantees.
Changes done in this PR:
- Extract table's bucketing information in `HiveClientImpl`
- While writing table info to metastore, `HiveClientImpl` now populates the bucketing information in the hive `Table` object
- `InsertIntoHiveTable` allows inserts to bucketed table only if both `hive.enforce.bucketing` and `hive.enforce.sorting` are `false`
Ability to create bucketed tables will enable adding test cases to Spark while I add more changes related to hive bucketing support. Design doc for hive hive bucketing support : https://docs.google.com/document/d/1a8IDh23RAkrkg9YYAeO51F4aGO8-xAlupKwdshve2fc/edit#
## How was this patch tested?
- Added test for creating bucketed and sorted table.
- Added test to ensure that INSERTs fail if strict bucket / sort is enforced
- Added test to ensure that INSERTs can go through if strict bucket / sort is NOT enforced
- Added test to validate that bucketing information shows up in output of DESC FORMATTED
- Added test to ensure that `SHOW CREATE TABLE` works for hive bucketed tables
Author: Tejas Patil <tejasp@fb.com>
Closes#17644 from tejasapatil/SPARK-17729_create_bucketed_table.
## What changes were proposed in this pull request?
This pr added a new Optimizer rule to combine nested Concat. The master supports a pipeline operator '||' to concatenate strings in #17711 (This pr is follow-up). Since the parser currently generates nested Concat expressions, the optimizer needs to combine the nested expressions.
## How was this patch tested?
Added tests in `CombineConcatSuite` and `SQLQueryTestSuite`.
Author: Takeshi Yamamuro <yamamuro@apache.org>
Closes#17970 from maropu/SPARK-20730.
## What changes were proposed in this pull request?
This PR proposes three things as below:
- Use casting rules to a timestamp in `to_timestamp` by default (it was `yyyy-MM-dd HH:mm:ss`).
- Support single argument for `to_timestamp` similarly with APIs in other languages.
For example, the one below works
```
import org.apache.spark.sql.functions._
Seq("2016-12-31 00:12:00.00").toDF("a").select(to_timestamp(col("a"))).show()
```
prints
```
+----------------------------------------+
|to_timestamp(`a`, 'yyyy-MM-dd HH:mm:ss')|
+----------------------------------------+
| 2016-12-31 00:12:00|
+----------------------------------------+
```
whereas this does not work in SQL.
**Before**
```
spark-sql> SELECT to_timestamp('2016-12-31 00:12:00');
Error in query: Invalid number of arguments for function to_timestamp; line 1 pos 7
```
**After**
```
spark-sql> SELECT to_timestamp('2016-12-31 00:12:00');
2016-12-31 00:12:00
```
- Related document improvement for SQL function descriptions and other API descriptions accordingly.
**Before**
```
spark-sql> DESCRIBE FUNCTION extended to_date;
...
Usage: to_date(date_str, fmt) - Parses the `left` expression with the `fmt` expression. Returns null with invalid input.
Extended Usage:
Examples:
> SELECT to_date('2016-12-31', 'yyyy-MM-dd');
2016-12-31
```
```
spark-sql> DESCRIBE FUNCTION extended to_timestamp;
...
Usage: to_timestamp(timestamp, fmt) - Parses the `left` expression with the `format` expression to a timestamp. Returns null with invalid input.
Extended Usage:
Examples:
> SELECT to_timestamp('2016-12-31', 'yyyy-MM-dd');
2016-12-31 00:00:00.0
```
**After**
```
spark-sql> DESCRIBE FUNCTION extended to_date;
...
Usage:
to_date(date_str[, fmt]) - Parses the `date_str` expression with the `fmt` expression to
a date. Returns null with invalid input. By default, it follows casting rules to a date if
the `fmt` is omitted.
Extended Usage:
Examples:
> SELECT to_date('2009-07-30 04:17:52');
2009-07-30
> SELECT to_date('2016-12-31', 'yyyy-MM-dd');
2016-12-31
```
```
spark-sql> DESCRIBE FUNCTION extended to_timestamp;
...
Usage:
to_timestamp(timestamp[, fmt]) - Parses the `timestamp` expression with the `fmt` expression to
a timestamp. Returns null with invalid input. By default, it follows casting rules to
a timestamp if the `fmt` is omitted.
Extended Usage:
Examples:
> SELECT to_timestamp('2016-12-31 00:12:00');
2016-12-31 00:12:00
> SELECT to_timestamp('2016-12-31', 'yyyy-MM-dd');
2016-12-31 00:00:00
```
## How was this patch tested?
Added tests in `datetime.sql`.
Author: hyukjinkwon <gurwls223@gmail.com>
Closes#17901 from HyukjinKwon/to_timestamp_arg.
## What changes were proposed in this pull request?
spark-sql>select bround(12.3, 2);
spark-sql>NULL
For this case, the expected result is 12.3, but it is null.
So ,when the second parameter is bigger than "decimal.scala", the result is not we expected.
"round" function has the same problem. This PR can solve the problem for both of them.
## How was this patch tested?
unit test cases in MathExpressionsSuite and MathFunctionsSuite
Author: liuxian <liu.xian3@zte.com.cn>
Closes#17906 from 10110346/wip_lx_0509.
## What changes were proposed in this pull request?
The new SQL parser is introduced into Spark 2.0. All string literals are unescaped in parser. Seems it bring an issue regarding the regex pattern string.
The following codes can reproduce it:
val data = Seq("\u0020\u0021\u0023", "abc")
val df = data.toDF()
// 1st usage: works in 1.6
// Let parser parse pattern string
val rlike1 = df.filter("value rlike '^\\x20[\\x20-\\x23]+$'")
// 2nd usage: works in 1.6, 2.x
// Call Column.rlike so the pattern string is a literal which doesn't go through parser
val rlike2 = df.filter($"value".rlike("^\\x20[\\x20-\\x23]+$"))
// In 2.x, we need add backslashes to make regex pattern parsed correctly
val rlike3 = df.filter("value rlike '^\\\\x20[\\\\x20-\\\\x23]+$'")
Follow the discussion in #17736, this patch adds a config to fallback to 1.6 string literal parsing and mitigate migration issue.
## 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#17887 from viirya/add-config-fallback-string-parsing.
## What changes were proposed in this pull request?
This pr added parsing rules to support aliases in table value functions.
The previous pr (#17666) has been reverted because of the regression. This new pr fixed the regression and add tests in `SQLQueryTestSuite`.
## How was this patch tested?
Added tests in `PlanParserSuite` and `SQLQueryTestSuite`.
Author: Takeshi Yamamuro <yamamuro@apache.org>
Closes#17928 from maropu/SPARK-20311-3.
## What changes were proposed in this pull request?
In filter estimation, we update column stats for those columns in filter condition. However, if the number of rows decreases after the filter (i.e. the overall selectivity is less than 1), we need to update (scale down) the number of distinct values (NDV) for all columns, no matter they are in filter conditions or not.
This pr also fixes the inconsistency of rounding mode for ndv and rowCount.
## How was this patch tested?
Added new tests.
Author: wangzhenhua <wangzhenhua@huawei.com>
Closes#17918 from wzhfy/scaleDownNdvAfterFilter.
## What changes were proposed in this pull request?
The query
```
SELECT 1 FROM (SELECT COUNT(*) WHERE FALSE) t1
```
should return a single row of output because the subquery is an aggregate without a group-by and thus should return a single row. However, Spark incorrectly returns zero rows.
This is caused by SPARK-16208 / #13906, a patch which added an optimizer rule to propagate EmptyRelation through operators. The logic for handling aggregates is wrong: it checks whether aggregate expressions are non-empty for deciding whether the output should be empty, whereas it should be checking grouping expressions instead:
An aggregate with non-empty grouping expression will return one output row per group. If the input to the grouped aggregate is empty then all groups will be empty and thus the output will be empty. It doesn't matter whether the aggregation output columns include aggregate expressions since that won't affect the number of output rows.
If the grouping expressions are empty, however, then the aggregate will always produce a single output row and thus we cannot propagate the EmptyRelation.
The current implementation is incorrect and also misses an optimization opportunity by not propagating EmptyRelation in the case where a grouped aggregate has aggregate expressions (in other words, `SELECT COUNT(*) from emptyRelation GROUP BY x` would _not_ be optimized to `EmptyRelation` in the old code, even though it safely could be).
This patch resolves this issue by modifying `PropagateEmptyRelation` to consider only the presence/absence of grouping expressions, not the aggregate functions themselves, when deciding whether to propagate EmptyRelation.
## How was this patch tested?
- Added end-to-end regression tests in `SQLQueryTest`'s `group-by.sql` file.
- Updated unit tests in `PropagateEmptyRelationSuite`.
Author: Josh Rosen <joshrosen@databricks.com>
Closes#17929 from JoshRosen/fix-PropagateEmptyRelation.
## What changes were proposed in this pull request?
This pr added parsing rules to support aliases in table value functions.
## How was this patch tested?
Added tests in `PlanParserSuite`.
Author: Takeshi Yamamuro <yamamuro@apache.org>
Closes#17666 from maropu/SPARK-20311.
## What changes were proposed in this pull request?
Fix build warnings primarily related to Breeze 0.13 operator changes, Java style problems
## How was this patch tested?
Existing tests
Author: Sean Owen <sowen@cloudera.com>
Closes#17803 from srowen/SPARK-20523.
## What changes were proposed in this pull request?
A fix for the same problem was made in #17693 but ignored `JsonToStructs`. This PR uses the same fix for `JsonToStructs`.
## How was this patch tested?
Regression test
Author: Burak Yavuz <brkyvz@gmail.com>
Closes#17826 from brkyvz/SPARK-20549.
## What changes were proposed in this pull request?
Add support for the SQL standard distinct predicate to SPARK SQL.
```
<expression> IS [NOT] DISTINCT FROM <expression>
```
## How was this patch tested?
Tested using unit tests, integration tests, manual tests.
Author: ptkool <michael.styles@shopify.com>
Closes#17764 from ptkool/is_not_distinct_from.
## What changes were proposed in this pull request?
Currently, when the type string is invalid, it looks printing empty parentheses. This PR proposes a small improvement in an error message by removing it in the parse as below:
```scala
spark.range(1).select($"col".cast("aa"))
```
**Before**
```
org.apache.spark.sql.catalyst.parser.ParseException:
DataType aa() is not supported.(line 1, pos 0)
== SQL ==
aa
^^^
```
**After**
```
org.apache.spark.sql.catalyst.parser.ParseException:
DataType aa is not supported.(line 1, pos 0)
== SQL ==
aa
^^^
```
## How was this patch tested?
Unit tests in `DataTypeParserSuite`.
Author: hyukjinkwon <gurwls223@gmail.com>
Closes#17784 from HyukjinKwon/SPARK-20492.
## What changes were proposed in this pull request?
Relax the requirement that a `TimeZoneAwareExpression` has to have its `timeZoneId` set to be considered resolved.
With this change, a `Cast` (which is a `TimeZoneAwareExpression`) can be considered resolved if the `(fromType, toType)` combination doesn't require time zone information.
Also de-relaxed test cases in `CastSuite` so Casts in that test suite don't get a default`timeZoneId = Option("GMT")`.
## How was this patch tested?
Ran the de-relaxed`CastSuite` and it's passing. Also ran the SQL unit tests and they're passing too.
Author: Kris Mok <kris.mok@databricks.com>
Closes#17777 from rednaxelafx/fix-catalyst-cast-timezone.
## What changes were proposed in this pull request?
change to using Jackson's `com.fasterxml.jackson.core.JsonFactory`
public JsonParser createParser(String content)
## How was this patch tested?
existing unit tests
Please review http://spark.apache.org/contributing.html before opening a pull request.
Author: Eric Wasserman <ericw@sgn.com>
Closes#17693 from ewasserman/SPARK-20314.
## What changes were proposed in this pull request?
This PR avoids an exception in the case where `scala.math.BigInt` has a value that does not fit into long value range (e.g. `Long.MAX_VALUE+1`). When we run the following code by using the current Spark, the following exception is thrown.
This PR keeps the value using `BigDecimal` if we detect such an overflow case by catching `ArithmeticException`.
Sample program:
```
case class BigIntWrapper(value:scala.math.BigInt)```
spark.createDataset(BigIntWrapper(scala.math.BigInt("10000000000000000002"))::Nil).show
```
Exception:
```
Error while encoding: java.lang.ArithmeticException: BigInteger out of long range
staticinvoke(class org.apache.spark.sql.types.Decimal$, DecimalType(38,0), apply, assertnotnull(assertnotnull(input[0, org.apache.spark.sql.BigIntWrapper, true])).value, true) AS value#0
java.lang.RuntimeException: Error while encoding: java.lang.ArithmeticException: BigInteger out of long range
staticinvoke(class org.apache.spark.sql.types.Decimal$, DecimalType(38,0), apply, assertnotnull(assertnotnull(input[0, org.apache.spark.sql.BigIntWrapper, true])).value, true) AS value#0
at org.apache.spark.sql.catalyst.encoders.ExpressionEncoder.toRow(ExpressionEncoder.scala:290)
at org.apache.spark.sql.SparkSession$$anonfun$2.apply(SparkSession.scala:454)
at org.apache.spark.sql.SparkSession$$anonfun$2.apply(SparkSession.scala:454)
at scala.collection.TraversableLike$$anonfun$map$1.apply(TraversableLike.scala:234)
at scala.collection.TraversableLike$$anonfun$map$1.apply(TraversableLike.scala:234)
at scala.collection.immutable.List.foreach(List.scala:381)
at scala.collection.TraversableLike$class.map(TraversableLike.scala:234)
at scala.collection.immutable.List.map(List.scala:285)
at org.apache.spark.sql.SparkSession.createDataset(SparkSession.scala:454)
at org.apache.spark.sql.Agg$$anonfun$18.apply$mcV$sp(MySuite.scala:192)
at org.apache.spark.sql.Agg$$anonfun$18.apply(MySuite.scala:192)
at org.apache.spark.sql.Agg$$anonfun$18.apply(MySuite.scala:192)
at org.scalatest.Transformer$$anonfun$apply$1.apply$mcV$sp(Transformer.scala:22)
at org.scalatest.OutcomeOf$class.outcomeOf(OutcomeOf.scala:85)
at org.scalatest.OutcomeOf$.outcomeOf(OutcomeOf.scala:104)
at org.scalatest.Transformer.apply(Transformer.scala:22)
at org.scalatest.Transformer.apply(Transformer.scala:20)
at org.scalatest.FunSuiteLike$$anon$1.apply(FunSuiteLike.scala:166)
at org.apache.spark.SparkFunSuite.withFixture(SparkFunSuite.scala:68)
at org.scalatest.FunSuiteLike$class.invokeWithFixture$1(FunSuiteLike.scala:163)
at org.scalatest.FunSuiteLike$$anonfun$runTest$1.apply(FunSuiteLike.scala:175)
at org.scalatest.FunSuiteLike$$anonfun$runTest$1.apply(FunSuiteLike.scala:175)
at org.scalatest.SuperEngine.runTestImpl(Engine.scala:306)
at org.scalatest.FunSuiteLike$class.runTest(FunSuiteLike.scala:175)
...
Caused by: java.lang.ArithmeticException: BigInteger out of long range
at java.math.BigInteger.longValueExact(BigInteger.java:4531)
at org.apache.spark.sql.types.Decimal.set(Decimal.scala:140)
at org.apache.spark.sql.types.Decimal$.apply(Decimal.scala:434)
at org.apache.spark.sql.types.Decimal.apply(Decimal.scala)
at org.apache.spark.sql.catalyst.expressions.GeneratedClass$SpecificUnsafeProjection.apply(Unknown Source)
at org.apache.spark.sql.catalyst.encoders.ExpressionEncoder.toRow(ExpressionEncoder.scala:287)
... 59 more
```
## How was this patch tested?
Add new test suite into `DecimalSuite`
Author: Kazuaki Ishizaki <ishizaki@jp.ibm.com>
Closes#17684 from kiszk/SPARK-20341.
## What changes were proposed in this pull request?
It is often useful to be able to track changes to the `ExternalCatalog`. This PR makes the `ExternalCatalog` emit events when a catalog object is changed. Events are fired before and after the change.
The following events are fired per object:
- Database
- CreateDatabasePreEvent: event fired before the database is created.
- CreateDatabaseEvent: event fired after the database has been created.
- DropDatabasePreEvent: event fired before the database is dropped.
- DropDatabaseEvent: event fired after the database has been dropped.
- Table
- CreateTablePreEvent: event fired before the table is created.
- CreateTableEvent: event fired after the table has been created.
- RenameTablePreEvent: event fired before the table is renamed.
- RenameTableEvent: event fired after the table has been renamed.
- DropTablePreEvent: event fired before the table is dropped.
- DropTableEvent: event fired after the table has been dropped.
- Function
- CreateFunctionPreEvent: event fired before the function is created.
- CreateFunctionEvent: event fired after the function has been created.
- RenameFunctionPreEvent: event fired before the function is renamed.
- RenameFunctionEvent: event fired after the function has been renamed.
- DropFunctionPreEvent: event fired before the function is dropped.
- DropFunctionPreEvent: event fired after the function has been dropped.
The current events currently only contain the names of the object modified. We add more events, and more details at a later point.
A user can monitor changes to the external catalog by adding a listener to the Spark listener bus checking for `ExternalCatalogEvent`s using the `SparkListener.onOtherEvent` hook. A more direct approach is add listener directly to the `ExternalCatalog`.
## How was this patch tested?
Added the `ExternalCatalogEventSuite`.
Author: Herman van Hovell <hvanhovell@databricks.com>
Closes#17710 from hvanhovell/SPARK-20420.
## What changes were proposed in this pull request?
A cast expression with a resolved time zone is not equal to a cast expression without a resolved time zone. The `ResolveAggregateFunction` assumed that these expression were the same, and would fail to resolve `HAVING` clauses which contain a `Cast` expression.
This is in essence caused by the fact that a `TimeZoneAwareExpression` can be resolved without a set time zone. This PR fixes this, and makes a `TimeZoneAwareExpression` unresolved as long as it has no TimeZone set.
## How was this patch tested?
Added a regression test to the `SQLQueryTestSuite.having` file.
Author: Herman van Hovell <hvanhovell@databricks.com>
Closes#17641 from hvanhovell/SPARK-20329.
## What changes were proposed in this pull request?
Apply Complementation Laws during boolean expression simplification.
## How was this patch tested?
Tested using unit tests, integration tests, and manual tests.
Author: ptkool <michael.styles@shopify.com>
Author: Michael Styles <michael.styles@shopify.com>
Closes#17650 from ptkool/apply_complementation_laws.
## What changes were proposed in this pull request?
If a plan has multi-level successive joins, e.g.:
```
Join
/ \
Union t5
/ \
Join t4
/ \
Join t3
/ \
t1 t2
```
Currently we fail to reorder the inside joins, i.e. t1, t2, t3.
In join reorder, we use `OrderedJoin` to indicate a join has been ordered, such that when transforming down the plan, these joins don't need to be rerodered again.
But there's a problem in the definition of `OrderedJoin`:
The real join node is a parameter, but not a child. This breaks the transform procedure because `mapChildren` applies transform function on parameters which should be children.
In this patch, we change `OrderedJoin` to a class having the same structure as a join node.
## How was this patch tested?
Add a corresponding test case.
Author: wangzhenhua <wangzhenhua@huawei.com>
Closes#17668 from wzhfy/recursiveReorder.
## What changes were proposed in this pull request?
Replace non-existent `repartitionBy` with `distribute` in `CollapseRepartitionSuite`.
## How was this patch tested?
local build and `catalyst/testOnly *CollapseRepartitionSuite`
Author: Jacek Laskowski <jacek@japila.pl>
Closes#17657 from jaceklaskowski/CollapseRepartitionSuite.
## What changes were proposed in this pull request?
This patch fixes a bug in the way LIKE patterns are translated to Java regexes. The bug causes any character following an escaped backslash to be escaped, i.e. there is double-escaping.
A concrete example is the following pattern:`'%\\%'`. The expected Java regex that this pattern should correspond to (according to the behavior described below) is `'.*\\.*'`, however the current situation leads to `'.*\\%'` instead.
---
Update: in light of the discussion that ensued, we should explicitly define the expected behaviour of LIKE expressions, especially in certain edge cases. With the help of gatorsmile, we put together a list of different RDBMS and their variations wrt to certain standard features.
| RDBMS\Features | Wildcards | Default escape [1] | Case sensitivity |
| --- | --- | --- | --- |
| [MS SQL Server](https://msdn.microsoft.com/en-us/library/ms179859.aspx) | _, %, [], [^] | none | no |
| [Oracle](https://docs.oracle.com/cd/B12037_01/server.101/b10759/conditions016.htm) | _, % | none | yes |
| [DB2 z/OS](http://www.ibm.com/support/knowledgecenter/SSEPEK_11.0.0/sqlref/src/tpc/db2z_likepredicate.html) | _, % | none | yes |
| [MySQL](http://dev.mysql.com/doc/refman/5.7/en/string-comparison-functions.html) | _, % | none | no |
| [PostreSQL](https://www.postgresql.org/docs/9.0/static/functions-matching.html) | _, % | \ | yes |
| [Hive](https://cwiki.apache.org/confluence/display/Hive/LanguageManual+UDF) | _, % | none | yes |
| Current Spark | _, % | \ | yes |
[1] Default escape character: most systems do not have a default escape character, instead the user can specify one by calling a like expression with an escape argument [A] LIKE [B] ESCAPE [C]. This syntax is currently not supported by Spark, however I would volunteer to implement this feature in a separate ticket.
The specifications are often quite terse and certain scenarios are undocumented, so here is a list of scenarios that I am uncertain about and would appreciate any input. Specifically I am looking for feedback on whether or not Spark's current behavior should be changed.
1. [x] Ending a pattern with the escape sequence, e.g. `like 'a\'`.
PostreSQL gives an error: 'LIKE pattern must not end with escape character', which I personally find logical. Currently, Spark allows "non-terminated" escapes and simply ignores them as part of the pattern.
According to [DB2's documentation](http://www.ibm.com/support/knowledgecenter/SSEPGG_9.7.0/com.ibm.db2.luw.messages.sql.doc/doc/msql00130n.html), ending a pattern in an escape character is invalid.
_Proposed new behaviour in Spark: throw AnalysisException_
2. [x] Empty input, e.g. `'' like ''`
Postgres and DB2 will match empty input only if the pattern is empty as well, any other combination of empty input will not match. Spark currently follows this rule.
3. [x] Escape before a non-special character, e.g. `'a' like '\a'`.
Escaping a non-wildcard character is not really documented but PostgreSQL just treats it verbatim, which I also find the least surprising behavior. Spark does the same.
According to [DB2's documentation](http://www.ibm.com/support/knowledgecenter/SSEPGG_9.7.0/com.ibm.db2.luw.messages.sql.doc/doc/msql00130n.html), it is invalid to follow an escape character with anything other than an escape character, an underscore or a percent sign.
_Proposed new behaviour in Spark: throw AnalysisException_
The current specification is also described in the operator's source code in this patch.
## How was this patch tested?
Extra case in regex unit tests.
Author: Jakob Odersky <jakob@odersky.com>
This patch had conflicts when merged, resolved by
Committer: Reynold Xin <rxin@databricks.com>
Closes#15398 from jodersky/SPARK-17647.
## What changes were proposed in this pull request?
Currently when estimating predicates like col > literal or col = literal, we will update min or max in column stats based on literal value. However, literal value is of Catalyst type (internal type), while min/max is of external type. Then for the next predicate, we again need to do type conversion to compare and update column stats. This is awkward and causes many unnecessary conversions in estimation.
To solve this, we use Catalyst type for min/max in `ColumnStat`. Note that the persistent format in metastore is still of external type, so there's no inconsistency for statistics in metastore.
This pr also fixes a bug for boolean type in `IN` condition.
## How was this patch tested?
The changes for ColumnStat are covered by existing tests.
For bug fix, a new test for boolean type in IN condition is added
Author: wangzhenhua <wangzhenhua@huawei.com>
Closes#17630 from wzhfy/refactorColumnStat.
### What changes were proposed in this pull request?
Session catalog API `createTempFunction` is being used by Hive build-in functions, persistent functions, and temporary functions. Thus, the name is confusing. This PR is to rename it by `registerFunction`. Also we can move construction of `FunctionBuilder` and `ExpressionInfo` into the new `registerFunction`, instead of duplicating the logics everywhere.
In the next PRs, the remaining Function-related APIs also need cleanups.
### How was this patch tested?
Existing test cases.
Author: Xiao Li <gatorsmile@gmail.com>
Closes#17615 from gatorsmile/cleanupCreateTempFunction.
## What changes were proposed in this pull request?
This PR proposes to run Spark unidoc to test Javadoc 8 build as Javadoc 8 is easily re-breakable.
There are several problems with it:
- It introduces little extra bit of time to run the tests. In my case, it took 1.5 mins more (`Elapsed :[94.8746569157]`). How it was tested is described in "How was this patch tested?".
- > One problem that I noticed was that Unidoc appeared to be processing test sources: if we can find a way to exclude those from being processed in the first place then that might significantly speed things up.
(see joshrosen's [comment](https://issues.apache.org/jira/browse/SPARK-18692?focusedCommentId=15947627&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-15947627))
To complete this automated build, It also suggests to fix existing Javadoc breaks / ones introduced by test codes as described above.
There fixes are similar instances that previously fixed. Please refer https://github.com/apache/spark/pull/15999 and https://github.com/apache/spark/pull/16013
Note that this only fixes **errors** not **warnings**. Please see my observation https://github.com/apache/spark/pull/17389#issuecomment-288438704 for spurious errors by warnings.
## How was this patch tested?
Manually via `jekyll build` for building tests. Also, tested via running `./dev/run-tests`.
This was tested via manually adding `time.time()` as below:
```diff
profiles_and_goals = build_profiles + sbt_goals
print("[info] Building Spark unidoc (w/Hive 1.2.1) using SBT with these arguments: ",
" ".join(profiles_and_goals))
+ import time
+ st = time.time()
exec_sbt(profiles_and_goals)
+ print("Elapsed :[%s]" % str(time.time() - st))
```
produces
```
...
========================================================================
Building Unidoc API Documentation
========================================================================
...
[info] Main Java API documentation successful.
...
Elapsed :[94.8746569157]
...
Author: hyukjinkwon <gurwls223@gmail.com>
Closes#17477 from HyukjinKwon/SPARK-18692.
## What changes were proposed in this pull request?
When we perform a cast expression and the from and to types are structurally the same (having the same structure but different field names), we should be able to skip the actual cast.
## How was this patch tested?
Added unit tests for the newly introduced functions.
Author: Reynold Xin <rxin@databricks.com>
Closes#17614 from rxin/SPARK-20302.
## What changes were proposed in this pull request?
`NaNvl(float value, null)` will be converted into `NaNvl(float value, Cast(null, DoubleType))` and finally `NaNvl(Cast(float value, DoubleType), Cast(null, DoubleType))`.
This will cause mismatching in the output type when the input type is float.
By adding extra rule in TypeCoercion can resolve this issue.
## How was this patch tested?
unite tests.
Please review http://spark.apache.org/contributing.html before opening a pull request.
Author: DB Tsai <dbt@netflix.com>
Closes#17606 from dbtsai/fixNaNvl.
## What changes were proposed in this pull request?
This is a regression caused by SPARK-19716.
Before SPARK-19716, we will cast an array field to the expected array type. However, after SPARK-19716, the cast is removed, but we forgot to push the cast to the element level.
## How was this patch tested?
new regression tests
Author: Wenchen Fan <wenchen@databricks.com>
Closes#17587 from cloud-fan/array.
## What changes were proposed in this pull request?
Add Locale.ROOT to internal calls to String `toLowerCase`, `toUpperCase`, to avoid inadvertent locale-sensitive variation in behavior (aka the "Turkish locale problem").
The change looks large but it is just adding `Locale.ROOT` (the locale with no country or language specified) to every call to these methods.
## How was this patch tested?
Existing tests.
Author: Sean Owen <sowen@cloudera.com>
Closes#17527 from srowen/SPARK-20156.
## What changes were proposed in this pull request?
```
sql("SELECT t1.b, rand(0) as r FROM cachedData, cachedData t1 GROUP BY t1.b having r > 0.5").show()
```
We will get the following error:
```
Job aborted due to stage failure: Task 1 in stage 4.0 failed 1 times, most recent failure: Lost task 1.0 in stage 4.0 (TID 8, localhost, executor driver): java.lang.NullPointerException
at org.apache.spark.sql.catalyst.expressions.GeneratedClass$SpecificPredicate.eval(Unknown Source)
at org.apache.spark.sql.execution.joins.BroadcastNestedLoopJoinExec$$anonfun$org$apache$spark$sql$execution$joins$BroadcastNestedLoopJoinExec$$boundCondition$1.apply(BroadcastNestedLoopJoinExec.scala:87)
at org.apache.spark.sql.execution.joins.BroadcastNestedLoopJoinExec$$anonfun$org$apache$spark$sql$execution$joins$BroadcastNestedLoopJoinExec$$boundCondition$1.apply(BroadcastNestedLoopJoinExec.scala:87)
at scala.collection.Iterator$$anon$13.hasNext(Iterator.scala:463)
```
Filters could be pushed down to the join conditions by the optimizer rule `PushPredicateThroughJoin`. However, Analyzer [blocks users to add non-deterministics conditions](https://github.com/apache/spark/blob/master/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/CheckAnalysis.scala#L386-L395) (For details, see the PR https://github.com/apache/spark/pull/7535).
We should not push down non-deterministic conditions; otherwise, we need to explicitly initialize the non-deterministic expressions. This PR is to simply block it.
### How was this patch tested?
Added a test case
Author: Xiao Li <gatorsmile@gmail.com>
Closes#17585 from gatorsmile/joinRandCondition.
## What changes were proposed in this pull request?
This PR proposes to add `IGNORE NULLS` keyword in `first`/`last` in Spark's parser likewise http://docs.oracle.com/cd/B19306_01/server.102/b14200/functions057.htm. This simply maps the keywords to existing `ignoreNullsExpr`.
**Before**
```scala
scala> sql("select first('a' IGNORE NULLS)").show()
```
```
org.apache.spark.sql.catalyst.parser.ParseException:
extraneous input 'NULLS' expecting {')', ','}(line 1, pos 24)
== SQL ==
select first('a' IGNORE NULLS)
------------------------^^^
at org.apache.spark.sql.catalyst.parser.ParseException.withCommand(ParseDriver.scala:210)
at org.apache.spark.sql.catalyst.parser.AbstractSqlParser.parse(ParseDriver.scala:112)
at org.apache.spark.sql.execution.SparkSqlParser.parse(SparkSqlParser.scala:46)
at org.apache.spark.sql.catalyst.parser.AbstractSqlParser.parsePlan(ParseDriver.scala:66)
at org.apache.spark.sql.SparkSession.sql(SparkSession.scala:622)
... 48 elided
```
**After**
```scala
scala> sql("select first('a' IGNORE NULLS)").show()
```
```
+--------------+
|first(a, true)|
+--------------+
| a|
+--------------+
```
## How was this patch tested?
Unit tests in `ExpressionParserSuite`.
Author: hyukjinkwon <gurwls223@gmail.com>
Closes#17566 from HyukjinKwon/SPARK-19518.
## What changes were proposed in this pull request?
Similar to `Project`, when `Aggregate` has non-deterministic expressions, we should not push predicate down through it, as it will change the number of input rows and thus change the evaluation result of non-deterministic expressions in `Aggregate`.
## How was this patch tested?
new regression test
Author: Wenchen Fan <wenchen@databricks.com>
Closes#17562 from cloud-fan/filter.
## What changes were proposed in this pull request?
This commit moves star schema code from ```join.scala``` to ```StarSchemaDetection.scala```. It also applies some minor fixes in ```StarJoinReorderSuite.scala```.
## How was this patch tested?
Run existing ```StarJoinReorderSuite.scala```.
Author: Ioana Delaney <ioanamdelaney@gmail.com>
Closes#17544 from ioana-delaney/starSchemaCBOv2.
## What changes were proposed in this pull request?
Previously when we construct deserializer expression for array type, we will first cast the corresponding field to expected array type and then apply `MapObjects`.
However, by doing that, we lose the opportunity to do by-name resolution for struct type inside array type. In this PR, I introduce a `UnresolvedMapObjects` to hold the lambda function and the input array expression. Then during analysis, after the input array expression is resolved, we get the actual array element type and apply by-name resolution. Then we don't need to add `Cast` for array type when constructing the deserializer expression, as the element type is determined later at analyzer.
## How was this patch tested?
new regression test
Author: Wenchen Fan <wenchen@databricks.com>
Closes#17398 from cloud-fan/dataset.
## What changes were proposed in this pull request?
This is a follow-up of https://github.com/apache/spark/pull/17285 .
## How was this patch tested?
existing tests
Author: Wenchen Fan <wenchen@databricks.com>
Closes#17521 from cloud-fan/conf.
## What changes were proposed in this pull request?
In SQL queries, we also see predicate expressions involving two columns such as "column-1 (op) column-2" where column-1 and column-2 belong to same table. Note that, if column-1 and column-2 belong to different tables, then it is a join operator's work, NOT a filter operator's work.
This PR estimates filter selectivity on two columns of same table. For example, multiple tpc-h queries have this predicate "WHERE l_commitdate < l_receiptdate"
## How was this patch tested?
We added 6 new test cases to test various logical predicates involving two columns of same table.
Please review http://spark.apache.org/contributing.html before opening a pull request.
Author: Ron Hu <ron.hu@huawei.com>
Author: U-CHINA\r00754707 <r00754707@R00754707-SC04.china.huawei.com>
Closes#17415 from ron8hu/filterTwoColumns.
## What changes were proposed in this pull request?
This patch implements `listPartitionsByFilter()` for `InMemoryCatalog` and thus resolves an outstanding TODO causing the `PruneFileSourcePartitions` optimizer rule not to apply when "spark.sql.catalogImplementation" is set to "in-memory" (which is the default).
The change is straightforward: it extracts the code for further filtering of the list of partitions returned by the metastore's `getPartitionsByFilter()` out from `HiveExternalCatalog` into `ExternalCatalogUtils` and calls this new function from `InMemoryCatalog` on the whole list of partitions.
Now that this method is implemented we can always pass the `CatalogTable` to the `DataSource` in `FindDataSourceTable`, so that the latter is resolved to a relation with a `CatalogFileIndex`, which is what the `PruneFileSourcePartitions` rule matches for.
## How was this patch tested?
Ran existing tests and added new test for `listPartitionsByFilter` in `ExternalCatalogSuite`, which is subclassed by both `InMemoryCatalogSuite` and `HiveExternalCatalogSuite`.
Author: Adrian Ionescu <adrian@databricks.com>
Closes#17510 from adrian-ionescu/InMemoryCatalog.
## What changes were proposed in this pull request?
Currently, `DataType.fromJson` throws `scala.MatchError` or `java.util.NoSuchElementException` in some cases when the JSON input is invalid as below:
```scala
DataType.fromJson(""""abcd"""")
```
```
java.util.NoSuchElementException: key not found: abcd
at ...
```
```scala
DataType.fromJson("""{"abcd":"a"}""")
```
```
scala.MatchError: JObject(List((abcd,JString(a)))) (of class org.json4s.JsonAST$JObject)
at ...
```
```scala
DataType.fromJson("""{"fields": [{"a":123}], "type": "struct"}""")
```
```
scala.MatchError: JObject(List((a,JInt(123)))) (of class org.json4s.JsonAST$JObject)
at ...
```
After this PR,
```scala
DataType.fromJson(""""abcd"""")
```
```
java.lang.IllegalArgumentException: Failed to convert the JSON string 'abcd' to a data type.
at ...
```
```scala
DataType.fromJson("""{"abcd":"a"}""")
```
```
java.lang.IllegalArgumentException: Failed to convert the JSON string '{"abcd":"a"}' to a data type.
at ...
```
```scala
DataType.fromJson("""{"fields": [{"a":123}], "type": "struct"}""")
at ...
```
```
java.lang.IllegalArgumentException: Failed to convert the JSON string '{"a":123}' to a field.
```
## How was this patch tested?
Unit test added in `DataTypeSuite`.
Author: hyukjinkwon <gurwls223@gmail.com>
Closes#17468 from HyukjinKwon/fromjson_exception.
## What changes were proposed in this pull request?
`BroadcastHint` should use child's statistics and set `isBroadcastable` to true.
## How was this patch tested?
Added a new stats estimation test for `BroadcastHint`.
Author: wangzhenhua <wangzhenhua@huawei.com>
Closes#17504 from wzhfy/broadcastHintEstimation.
…adoc
## What changes were proposed in this pull request?
Use recommended values for row boundaries in Window's scaladoc, i.e. `Window.unboundedPreceding`, `Window.unboundedFollowing`, and `Window.currentRow` (that were introduced in 2.1.0).
## How was this patch tested?
Local build
Author: Jacek Laskowski <jacek@japila.pl>
Closes#17417 from jaceklaskowski/window-expression-scaladoc.
### What changes were proposed in this pull request?
`FalseLiteral` and `TrueLiteral` should have been eliminated by optimizer rule `BooleanSimplification`, but null literals might be added by optimizer rule `NullPropagation`. For safety, our filter estimation should handle all the eligible literal cases.
Our optimizer rule BooleanSimplification is unable to remove the null literal in many cases. For example, `a < 0 or null`. Thus, we need to handle null literal in filter estimation.
`Not` can be pushed down below `And` and `Or`. Then, we could see two consecutive `Not`, which need to be collapsed into one. Because of the limited expression support for filter estimation, we just need to handle the case `Not(null)` for avoiding incorrect error due to the boolean operation on null. For details, see below matrix.
```
not NULL = NULL
NULL or false = NULL
NULL or true = true
NULL or NULL = NULL
NULL and false = false
NULL and true = NULL
NULL and NULL = NULL
```
### How was this patch tested?
Added the test cases.
Author: Xiao Li <gatorsmile@gmail.com>
Closes#17446 from gatorsmile/constantFilterEstimation.
## What changes were proposed in this pull request?
This pr added `StructType.fromDDL` to convert a DDL format string into `StructType` for defining schemas in `functions.from_json`.
## How was this patch tested?
Added tests in `JsonFunctionsSuite`.
Author: Takeshi Yamamuro <yamamuro@apache.org>
Closes#17406 from maropu/SPARK-20009.
## What changes were proposed in this pull request?
Join reorder algorithm should keep exactly the same order of output attributes in the top project.
For example, if user want to select a, b, c, after reordering, we should output a, b, c in the same order as specified by user, instead of b, a, c or other orders.
## How was this patch tested?
A new test case is added in `JoinReorderSuite`.
Author: wangzhenhua <wangzhenhua@huawei.com>
Closes#17453 from wzhfy/keepOrderInProject.
## What changes were proposed in this pull request?
TPCDS q45 fails becuase:
`ReorderJoin` collects all predicates and try to put them into join condition when creating ordered join. If a predicate with an IN subquery (`ListQuery`) is in a join condition instead of a filter condition, `RewritePredicateSubquery.rewriteExistentialExpr` would fail to convert the subquery to an `ExistenceJoin`, and thus result in error.
We should prevent push down of IN subquery to Join operator.
## How was this patch tested?
Add a new test case in `FilterPushdownSuite`.
Author: wangzhenhua <wangzhenhua@huawei.com>
Closes#17428 from wzhfy/noSubqueryInJoinCond.
## What changes were proposed in this pull request?
The current SessionState initialization code path is quite complex. A part of the creation is done in the SessionState companion objects, a part of the creation is one inside the SessionState class, and a part is done by passing functions.
This PR refactors this code path, and consolidates SessionState initialization into a builder class. This SessionState will not do any initialization and just becomes a place holder for the various Spark SQL internals. This also lays the ground work for two future improvements:
1. This provides us with a start for removing the `HiveSessionState`. Removing the `HiveSessionState` would also require us to move resource loading into a separate class, and to (re)move metadata hive.
2. This makes it easier to customize the Spark Session. Currently you will need to create a custom version of the builder. I have added hooks to facilitate this. A future step will be to create a semi stable API on top of this.
## How was this patch tested?
Existing tests.
Author: Herman van Hovell <hvanhovell@databricks.com>
Closes#17433 from hvanhovell/SPARK-20100.
## What changes were proposed in this pull request?
In current stage, we don't have advanced statistics such as sketches or histograms. As a result, some operator can't estimate `nullCount` accurately. E.g. left outer join estimation does not accurately update `nullCount` currently. So for `IsNull` and `IsNotNull` predicates, we only estimate them when the child is a leaf node, whose `nullCount` is accurate.
## How was this patch tested?
A new test case is added in `FilterEstimationSuite`.
Author: wangzhenhua <wangzhenhua@huawei.com>
Closes#17438 from wzhfy/nullEstimation.
## What changes were proposed in this pull request?
The `CollapseWindow` is currently to aggressive when collapsing adjacent windows. It also collapses windows in the which the parent produces a column that is consumed by the child; this creates an invalid window which will fail at runtime.
This PR fixes this by adding a check for dependent adjacent windows to the `CollapseWindow` rule.
## How was this patch tested?
Added a new test case to `CollapseWindowSuite`
Author: Herman van Hovell <hvanhovell@databricks.com>
Closes#17432 from hvanhovell/SPARK-20086.
## What changes were proposed in this pull request?
Constraint propagation can be computation expensive and block the driver execution for long time. For example, the below benchmark needs 30mins.
Compared with previous PRs #16998, #16785, this is a much simpler option: add a flag to disable constraint propagation.
### Benchmark
Run the following codes locally.
import org.apache.spark.ml.{Pipeline, PipelineStage}
import org.apache.spark.ml.feature.{OneHotEncoder, StringIndexer, VectorAssembler}
import org.apache.spark.sql.internal.SQLConf
spark.conf.set(SQLConf.CONSTRAINT_PROPAGATION_ENABLED.key, false)
val df = (1 to 40).foldLeft(Seq((1, "foo"), (2, "bar"), (3, "baz")).toDF("id", "x0"))((df, i) => df.withColumn(s"x$i", $"x0"))
val indexers = df.columns.tail.map(c => new StringIndexer()
.setInputCol(c)
.setOutputCol(s"${c}_indexed")
.setHandleInvalid("skip"))
val encoders = indexers.map(indexer => new OneHotEncoder()
.setInputCol(indexer.getOutputCol)
.setOutputCol(s"${indexer.getOutputCol}_encoded")
.setDropLast(true))
val stages: Array[PipelineStage] = indexers ++ encoders
val pipeline = new Pipeline().setStages(stages)
val startTime = System.nanoTime
pipeline.fit(df).transform(df).show
val runningTime = System.nanoTime - startTime
Before this patch: 1786001 ms ~= 30 mins
After this patch: 26392 ms = less than half of a minute
Related PRs: #16998, #16785.
## 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#17186 from viirya/add-flag-disable-constraint-propagation.
## What changes were proposed in this pull request?
Since the state is tied a "group" in the "mapGroupsWithState" operations, its better to call the state "GroupState" instead of a key. This would make it more general if you extends this operation to RelationGroupedDataset and python APIs.
## How was this patch tested?
Existing unit tests.
Author: Tathagata Das <tathagata.das1565@gmail.com>
Closes#17385 from tdas/SPARK-20057.
## What changes were proposed in this pull request?
This PR proposes to make `mode` options in both CSV and JSON to use `cass object` and fix some related comments related previous fix.
Also, this PR modifies some tests related parse modes.
## How was this patch tested?
Modified unit tests in both `CSVSuite.scala` and `JsonSuite.scala`.
Author: hyukjinkwon <gurwls223@gmail.com>
Closes#17377 from HyukjinKwon/SPARK-19949.
## What changes were proposed in this pull request?
Adding event time based timeout. The user sets the timeout timestamp directly using `KeyedState.setTimeoutTimestamp`. The keys times out when the watermark crosses the timeout timestamp.
## How was this patch tested?
Unit tests
Author: Tathagata Das <tathagata.das1565@gmail.com>
Closes#17361 from tdas/SPARK-20030.
## What changes were proposed in this pull request?
Change the nullability of function `StringToMap` from `false` to `true`.
Author: zhaorongsheng <334362872@qq.com>
Closes#17350 from zhaorongsheng/bug-fix_strToMap_NPE.
## What changes were proposed in this pull request?
Support` ALTER TABLE ADD COLUMNS (...) `syntax for Hive serde and some datasource tables.
In this PR, we consider a few aspects:
1. View is not supported for `ALTER ADD COLUMNS`
2. Since tables created in SparkSQL with Hive DDL syntax will populate table properties with schema information, we need make sure the consistency of the schema before and after ALTER operation in order for future use.
3. For embedded-schema type of format, such as `parquet`, we need to make sure that the predicate on the newly-added columns can be evaluated properly, or pushed down properly. In case of the data file does not have the columns for the newly-added columns, such predicates should return as if the column values are NULLs.
4. For datasource table, this feature does not support the following:
4.1 TEXT format, since there is only one default column `value` is inferred for text format data.
4.2 ORC format, since SparkSQL native ORC reader does not support the difference between user-specified-schema and inferred schema from ORC files.
4.3 Third party datasource types that implements RelationProvider, including the built-in JDBC format, since different implementations by the vendors may have different ways to dealing with schema.
4.4 Other datasource types, such as `parquet`, `json`, `csv`, `hive` are supported.
5. Column names being added can not be duplicate of any existing data column or partition column names. Case sensitivity is taken into consideration according to the sql configuration.
6. This feature also supports In-Memory catalog, while Hive support is turned off.
## How was this patch tested?
Add new test cases
Author: Xin Wu <xinwu@us.ibm.com>
Closes#16626 from xwu0226/alter_add_columns.
### What changes were proposed in this pull request?
SessionCatalog API setCurrentDatabase does not set the current database of the underlying ExternalCatalog. Thus, weird errors could come in the test suites after we call reset. We need to fix it.
So far, have not found the direct impact in the other code paths because we expect all the SessionCatalog APIs should always use the current database value we managed, unless some of code paths skip it. Thus, we fix it in the test-only function reset().
### How was this patch tested?
Multiple test case failures are observed in mvn and add a test case in SessionCatalogSuite.
Author: Xiao Li <gatorsmile@gmail.com>
Closes#17354 from gatorsmile/useDB.
## What changes were proposed in this pull request?
update `StatFunctions.multipleApproxQuantiles` to handle NaN/null
## How was this patch tested?
existing tests and added tests
Author: Zheng RuiFeng <ruifengz@foxmail.com>
Closes#16971 from zhengruifeng/quantiles_nan.
## What changes were proposed in this pull request?
Star schema consists of one or more fact tables referencing a number of dimension tables. In general, queries against star schema are expected to run fast because of the established RI constraints among the tables. This design proposes a join reordering based on natural, generally accepted heuristics for star schema queries:
- Finds the star join with the largest fact table and places it on the driving arm of the left-deep join. This plan avoids large tables on the inner, and thus favors hash joins.
- Applies the most selective dimensions early in the plan to reduce the amount of data flow.
The design document was included in SPARK-17791.
Link to the google doc: [StarSchemaDetection](https://docs.google.com/document/d/1UAfwbm_A6wo7goHlVZfYK99pqDMEZUumi7pubJXETEA/edit?usp=sharing)
## How was this patch tested?
A new test suite StarJoinSuite.scala was implemented.
Author: Ioana Delaney <ioanamdelaney@gmail.com>
Closes#15363 from ioana-delaney/starJoinReord2.
## What changes were proposed in this pull request?
This PR proposes to support an array of struct type in `to_json` as below:
```scala
import org.apache.spark.sql.functions._
val df = Seq(Tuple1(Tuple1(1) :: Nil)).toDF("a")
df.select(to_json($"a").as("json")).show()
```
```
+----------+
| json|
+----------+
|[{"_1":1}]|
+----------+
```
Currently, it throws an exception as below (a newline manually inserted for readability):
```
org.apache.spark.sql.AnalysisException: cannot resolve 'structtojson(`array`)' due to data type
mismatch: structtojson requires that the expression is a struct expression.;;
```
This allows the roundtrip with `from_json` as below:
```scala
import org.apache.spark.sql.functions._
import org.apache.spark.sql.types._
val schema = ArrayType(StructType(StructField("a", IntegerType) :: Nil))
val df = Seq("""[{"a":1}, {"a":2}]""").toDF("json").select(from_json($"json", schema).as("array"))
df.show()
// Read back.
df.select(to_json($"array").as("json")).show()
```
```
+----------+
| array|
+----------+
|[[1], [2]]|
+----------+
+-----------------+
| json|
+-----------------+
|[{"a":1},{"a":2}]|
+-----------------+
```
Also, this PR proposes to rename from `StructToJson` to `StructsToJson ` and `JsonToStruct` to `JsonToStructs`.
## How was this patch tested?
Unit tests in `JsonFunctionsSuite` and `JsonExpressionsSuite` for Scala, doctest for Python and test in `test_sparkSQL.R` for R.
Author: hyukjinkwon <gurwls223@gmail.com>
Closes#17192 from HyukjinKwon/SPARK-19849.
## What changes were proposed in this pull request?
When a key does not get any new data in `mapGroupsWithState`, the mapping function is never called on it. So we need a timeout feature that calls the function again in such cases, so that the user can decide whether to continue waiting or clean up (remove state, save stuff externally, etc.).
Timeouts can be either based on processing time or event time. This JIRA is for processing time, but defines the high level API design for both. The usage would look like this.
```
def stateFunction(key: K, value: Iterator[V], state: KeyedState[S]): U = {
...
state.setTimeoutDuration(10000)
...
}
dataset // type is Dataset[T]
.groupByKey[K](keyingFunc) // generates KeyValueGroupedDataset[K, T]
.mapGroupsWithState[S, U](
func = stateFunction,
timeout = KeyedStateTimeout.withProcessingTime) // returns Dataset[U]
```
Note the following design aspects.
- The timeout type is provided as a param in mapGroupsWithState as a parameter global to all the keys. This is so that the planner knows this at planning time, and accordingly optimize the execution based on whether to saves extra info in state or not (e.g. timeout durations or timestamps).
- The exact timeout duration is provided inside the function call so that it can be customized on a per key basis.
- When the timeout occurs for a key, the function is called with no values, and KeyedState.isTimingOut() set to true.
- The timeout is reset for key every time the function is called on the key, that is, when the key has new data, or the key has timed out. So the user has to set the timeout duration everytime the function is called, otherwise there will not be any timeout set.
Guarantees provided on timeout of key, when timeout duration is D ms:
- Timeout will never be called before real clock time has advanced by D ms
- Timeout will be called eventually when there is a trigger with any data in it (i.e. after D ms). So there is a no strict upper bound on when the timeout would occur. For example, if there is no data in the stream (for any key) for a while, then the timeout will not be hit.
Implementation details:
- Added new param to `mapGroupsWithState` for timeout
- Added new method to `StateStore` to filter data based on timeout timestamp
- Changed the internal map type of `HDFSBackedStateStore` from Java's `HashMap` to `ConcurrentHashMap` as the latter allows weakly-consistent fail-safe iterators on the map data. See comments in code for more details.
- Refactored logic of `MapGroupsWithStateExec` to
- Save timeout info to state store for each key that has data.
- Then, filter states that should be timed out based on the current batch processing timestamp.
- Moved KeyedState for `o.a.s.sql` to `o.a.s.sql.streaming`. I remember that this was a feedback in the MapGroupsWithState PR that I had forgotten to address.
## How was this patch tested?
New unit tests in
- MapGroupsWithStateSuite for timeouts.
- StateStoreSuite for new APIs in StateStore.
Author: Tathagata Das <tathagata.das1565@gmail.com>
Closes#17179 from tdas/mapgroupwithstate-timeout.