Author: Reynold Xin <rxin@databricks.com>
Closes#3862 from rxin/stringcontext-attr and squashes the following commits:
9b10f57 [Reynold Xin] Rename StrongToAttributeConversionHelper
72121af [Reynold Xin] [SPARK-5040][SQL] Support expressing unresolved attributes using $"attribute name" notation in SQL DSL.
As we learned in https://github.com/apache/spark/pull/3580, not explicitly typing implicit functions can lead to compiler bugs and potentially unexpected runtime behavior.
Author: Reynold Xin <rxin@databricks.com>
Closes#3859 from rxin/sql-implicits and squashes the following commits:
30c2c24 [Reynold Xin] [SPARK-5038] Add explicit return type for implicit functions in Spark SQL.
JIRA issue: [SPARK-4570](https://issues.apache.org/jira/browse/SPARK-4570)
We are planning to create a `BroadcastLeftSemiJoinHash` to implement the broadcast join for `left semijoin`
In left semijoin :
If the size of data from right side is smaller than the user-settable threshold `AUTO_BROADCASTJOIN_THRESHOLD`,
the planner would mark it as the `broadcast` relation and mark the other relation as the stream side. The broadcast table will be broadcasted to all of the executors involved in the join, as a `org.apache.spark.broadcast.Broadcast` object. It will use `joins.BroadcastLeftSemiJoinHash`.,else it will use `joins.LeftSemiJoinHash`.
The benchmark suggests these made the optimized version 4x faster when `left semijoin`
<pre><code>
Original:
left semi join : 9288 ms
Optimized:
left semi join : 1963 ms
</code></pre>
The micro benchmark load `data1/kv3.txt` into a normal Hive table.
Benchmark code:
<pre><code>
def benchmark(f: => Unit) = {
val begin = System.currentTimeMillis()
f
val end = System.currentTimeMillis()
end - begin
}
val sc = new SparkContext(
new SparkConf()
.setMaster("local")
.setAppName(getClass.getSimpleName.stripSuffix("$")))
val hiveContext = new HiveContext(sc)
import hiveContext._
sql("drop table if exists left_table")
sql("drop table if exists right_table")
sql( """create table left_table (key int, value string)
""".stripMargin)
sql( s"""load data local inpath "/data1/kv3.txt" into table left_table""")
sql( """create table right_table (key int, value string)
""".stripMargin)
sql(
"""
|from left_table
|insert overwrite table right_table
|select left_table.key, left_table.value
""".stripMargin)
val leftSimeJoin = sql(
"""select a.key from left_table a
|left semi join right_table b on a.key = b.key""".stripMargin)
val leftSemiJoinDuration = benchmark(leftSimeJoin.count())
println(s"left semi join : $leftSemiJoinDuration ms ")
</code></pre>
Author: wangxiaojing <u9jing@gmail.com>
Closes#3442 from wangxiaojing/SPARK-4570 and squashes the following commits:
a4a43c9 [wangxiaojing] rebase
f103983 [wangxiaojing] change style
fbe4887 [wangxiaojing] change style
ff2e618 [wangxiaojing] add testsuite
1a8da2a [wangxiaojing] add BroadcastLeftSemiJoinHash
Convert type of RowWriteSupport.attributes to Array.
Analysis of performance for writing very wide tables shows that time is spent predominantly in apply method on attributes var. Type of attributes previously was LinearSeqOptimized and apply is O(N) which made write O(N squared).
Measurements on 575 column table showed this change made a 6x improvement in write times.
Author: Michael Davies <Michael.BellDavies@gmail.com>
Closes#3843 from MickDavies/SPARK-4386 and squashes the following commits:
892519d [Michael Davies] [SPARK-4386] Improve performance when writing Parquet files
This PR is a simplified version of several filter optimization rules introduced in #3778 authored by scwf. Newly introduced optimizations include:
1. `a && a` => `a`
2. `a || a` => `a`
3. `(a || b || c || ...) && (a || b || d || ...)` => `a && b && (c || d || ...)`
The 3rd rule is particularly useful for optimizing the following query, which is planned into a cartesian product
```sql
SELECT *
FROM t1, t2
WHERE (t1.key = t2.key AND t1.value > 10)
OR (t1.key = t2.key AND t2.value < 20)
```
to the following one, which is planned into an equi-join:
```sql
SELECT *
FROM t1, t2
WHERE t1.key = t2.key
AND (t1.value > 10 OR t2.value < 20)
```
The example above is quite artificial, but common predicates are likely to appear in real life complex queries (like the one mentioned in #3778).
A difference between this PR and #3778 is that these optimizations are not limited to `Filter`, but are generalized to all logical plan nodes. Thanks to scwf for bringing up these optimizations, and chenghao-intel for the generalization suggestion.
<!-- Reviewable:start -->
[<img src="https://reviewable.io/review_button.png" height=40 alt="Review on Reviewable"/>](https://reviewable.io/reviews/apache/spark/3784)
<!-- Reviewable:end -->
Author: Cheng Lian <lian@databricks.com>
Closes#3784 from liancheng/normalize-filters and squashes the following commits:
caca560 [Cheng Lian] Moves filter normalization into BooleanSimplification rule
4ab3a58 [Cheng Lian] Fixes test failure, adds more tests
5d54349 [Cheng Lian] Fixes typo in comment
2abbf8e [Cheng Lian] Forgot our sacred Apache licence header...
cf95639 [Cheng Lian] Adds an optimization rule for filter normalization
This is a follow-up of #3367 and #3644.
At the time #3644 was written, #3367 hadn't been merged yet, thus `IsNull` and `IsNotNull` filters are not covered in the first version of `ParquetFilterSuite`. This PR adds corresponding test cases.
<!-- Reviewable:start -->
[<img src="https://reviewable.io/review_button.png" height=40 alt="Review on Reviewable"/>](https://reviewable.io/reviews/apache/spark/3748)
<!-- Reviewable:end -->
Author: Cheng Lian <lian@databricks.com>
Closes#3748 from liancheng/test-null-filters and squashes the following commits:
1ab943f [Cheng Lian] IsNull and IsNotNull Parquet filter test case for boolean type
bcd616b [Cheng Lian] Adds Parquet filter pushedown tests for IsNull and IsNotNull
It will cause exception while do query like:
SELECT key+key FROM src sort by value;
Author: Cheng Hao <hao.cheng@intel.com>
Closes#3386 from chenghao-intel/sort and squashes the following commits:
38c78cc [Cheng Hao] revert the SortPartition in SparkStrategies
7e9dd15 [Cheng Hao] update the typo
fcd1d64 [Cheng Hao] rebase the latest master and update the SortBy unit test
spark sql does not support ```SELECT a, b FROM testData2 ORDER BY a desc, b```.
Author: wangfei <wangfei1@huawei.com>
Closes#3838 from scwf/orderby and squashes the following commits:
114b64a [wangfei] remove nouse methods
48145d3 [wangfei] fix order, using asc by default
There are a number of warnings generated in a normal, successful build right now. They're mostly Java unchecked cast warnings, which can be suppressed. But there's a grab bag of other Scala language warnings and so on that can all be easily fixed. The forthcoming PR fixes about 90% of the build warnings I see now.
Author: Sean Owen <sowen@cloudera.com>
Closes#3157 from srowen/SPARK-4297 and squashes the following commits:
8c9e469 [Sean Owen] Suppress unchecked cast warnings, and several other build warning fixes
This PR modifies the python `SchemaRDD` to use `sample()` and `takeSample()` from Scala instead of the slower python implementations from `rdd.py`. This is worthwhile because the `Row`'s are already serialized as Java objects.
In order to use the faster `takeSample()`, a `takeSampleToPython()` method was implemented in `SchemaRDD.scala` following the pattern of `collectToPython()`.
Author: jbencook <jbenjamincook@gmail.com>
Author: J. Benjamin Cook <jbenjamincook@gmail.com>
Closes#3764 from jbencook/master and squashes the following commits:
6fbc769 [J. Benjamin Cook] [SPARK-4860][pyspark][sql] fixing sloppy indentation for takeSampleToPython() arguments
5170da2 [J. Benjamin Cook] [SPARK-4860][pyspark][sql] fixing typo: from RDD to SchemaRDD
de22f70 [jbencook] [SPARK-4860][pyspark][sql] using sample() method from JavaSchemaRDD
b916442 [jbencook] [SPARK-4860][pyspark][sql] adding sample() to JavaSchemaRDD
020cbdf [jbencook] [SPARK-4860][pyspark][sql] using Scala implementations of `sample()` and `takeSample()`
...arquetFile accept hadoop glob pattern in path.
Author: Thu Kyaw <trk007@gmail.com>
Closes#3407 from tkyaw/master and squashes the following commits:
19115ad [Thu Kyaw] Merge https://github.com/apache/spark
ceded32 [Thu Kyaw] [SPARK-3928][SQL] Support wildcard matches on Parquet files.
d322c28 [Thu Kyaw] [SPARK-3928][SQL] Support wildcard matches on Parquet files.
ce677c6 [Thu Kyaw] [SPARK-3928][SQL] Support wildcard matches on Parquet files.
Add support for `GROUPING SETS`, `ROLLUP`, `CUBE` and the the virtual column `GROUPING__ID`.
More details on how to use the `GROUPING SETS" can be found at: https://cwiki.apache.org/confluence/display/Hive/Enhanced+Aggregation,+Cube,+Grouping+and+Rolluphttps://issues.apache.org/jira/secure/attachment/12676811/grouping_set.pdf
The generic idea of the implementations are :
1 Replace the `ROLLUP`, `CUBE` with `GROUPING SETS`
2 Explode each of the input row, and then feed them to `Aggregate`
* Each grouping set are represented as the bit mask for the `GroupBy Expression List`, for each bit, `1` means the expression is selected, otherwise `0` (left is the lower bit, and right is the higher bit in the `GroupBy Expression List`)
* Several of projections are constructed according to the grouping sets, and within each projection(Seq[Expression), we replace those expressions with `Literal(null)` if it's not selected in the grouping set (based on the bit mask)
* Output Schema of `Explode` is `child.output :+ grouping__id`
* GroupBy Expressions of `Aggregate` is `GroupBy Expression List :+ grouping__id`
* Keep the `Aggregation expressions` the same for the `Aggregate`
The expressions substitutions happen in Logic Plan analyzing, so we will benefit from the Logical Plan optimization (e.g. expression constant folding, and map side aggregation etc.), Only an `Explosive` operator added for Physical Plan, which will explode the rows according the pre-set projections.
A known issue will be done in the follow up PR:
* Optimization `ColumnPruning` is not supported yet for `Explosive` node.
Author: Cheng Hao <hao.cheng@intel.com>
Closes#1567 from chenghao-intel/grouping_sets and squashes the following commits:
fe65fcc [Cheng Hao] Remove the extra space
3547056 [Cheng Hao] Add more doc and Simplify the Expand
a7c869d [Cheng Hao] update code as feedbacks
d23c672 [Cheng Hao] Add GroupingExpression to replace the Seq[Expression]
414b165 [Cheng Hao] revert the unnecessary changes
ec276c6 [Cheng Hao] Support Rollup/Cube/GroupingSets
```
TestSQLContext.sparkContext.parallelize(
"""{"ip":"27.31.100.29","headers":{"Host":"1.abc.com","Charset":"UTF-8"}}""" ::
"""{"ip":"27.31.100.29","headers":{}}""" ::
"""{"ip":"27.31.100.29","headers":""}""" :: Nil)
```
As empty string (the "headers") will be considered as String in the beginning (in line 2 and 3), it ignores the real nested data type (struct type "headers" in line 1), and also take the line 1 (the "headers") as String Type, which is not our expected.
Author: Cheng Hao <hao.cheng@intel.com>
Closes#3708 from chenghao-intel/json and squashes the following commits:
e7a72e9 [Cheng Hao] add more concise unit test
853de51 [Cheng Hao] NullType instead of StringType when sampling against empty string or null value
Author: Michael Armbrust <michael@databricks.com>
Closes#3727 from marmbrus/parquetNotEq and squashes the following commits:
2157bfc [Michael Armbrust] Fix parquet filter suite
Predicates like `a = NULL` and `a < NULL` can't be pushed down since Parquet `Lt`, `LtEq`, `Gt`, `GtEq` doesn't accept null value. Note that `Eq` and `NotEq` can only be used with `null` to represent predicates like `a IS NULL` and `a IS NOT NULL`.
However, normally this issue doesn't cause NPE because any value compared to `NULL` results `NULL`, and Spark SQL automatically optimizes out `NULL` predicate in the `SimplifyFilters` rule. Only testing code that intentionally disables the optimizer may trigger this issue. (That's why this issue is not marked as blocker and I do **NOT** think we need to backport this to branch-1.1
This PR restricts `Lt`, `LtEq`, `Gt` and `GtEq` to non-null values only, and only uses `Eq` with null value to pushdown `IsNull` and `IsNotNull`. Also, added support for Parquet `NotEq` filter for completeness and (tiny) performance gain, it's also used to pushdown `IsNotNull`.
<!-- Reviewable:start -->
[<img src="https://reviewable.io/review_button.png" height=40 alt="Review on Reviewable"/>](https://reviewable.io/reviews/apache/spark/3367)
<!-- Reviewable:end -->
Author: Cheng Lian <lian@databricks.com>
Closes#3367 from liancheng/filters-with-null and squashes the following commits:
cc41281 [Cheng Lian] Fixes several styling issues
de7de28 [Cheng Lian] Adds stricter rules for Parquet filters with null
Add `sort by` support for both DSL & SqlParser.
This PR is relevant with #3386, either one merged, will cause the other rebased.
Author: Cheng Hao <hao.cheng@intel.com>
Closes#3481 from chenghao-intel/sortby and squashes the following commits:
041004f [Cheng Hao] Add sort by for DSL & SimpleSqlParser
Using lowercase for ```options``` key to make it case-insensitive, then we should use lower case to get value from parameters.
So flowing cmd work
```
create temporary table normal_parquet
USING org.apache.spark.sql.parquet
OPTIONS (
PATH '/xxx/data'
)
```
Author: scwf <wangfei1@huawei.com>
Author: wangfei <wangfei1@huawei.com>
Closes#3470 from scwf/ddl-ulcase and squashes the following commits:
ae78509 [scwf] address comments
8f4f585 [wangfei] address comments
3c132ef [scwf] minor fix
a0fc20b [scwf] Merge branch 'master' of https://github.com/apache/spark into ddl-ulcase
4f86401 [scwf] adding CaseInsensitiveMap
e244e8d [wangfei] using lower case in json
e0cb017 [wangfei] make options in-casesensitive
This PR brings support of using StructType(and other hashable types) as key in MapType.
Author: Davies Liu <davies@databricks.com>
Closes#3714 from davies/fix_struct_in_map and squashes the following commits:
68585d7 [Davies Liu] fix primitive types in MapType
9601534 [Davies Liu] support StructType as key in MapType
Author: Cheng Hao <hao.cheng@intel.com>
Closes#3595 from chenghao-intel/udf0 and squashes the following commits:
a858973 [Cheng Hao] Add 0 arguments support for udf
This PR provides a set Parquet testing API (see trait `ParquetTest`) that enables developers to write more concise test cases. A new set of Parquet test suites built upon this API are added and aim to replace the old `ParquetQuerySuite`. To avoid potential merge conflicts, old testing code are not removed yet. The following classes can be safely removed after most Parquet related PRs are handled:
- `ParquetQuerySuite`
- `ParquetTestData`
<!-- Reviewable:start -->
[<img src="https://reviewable.io/review_button.png" height=40 alt="Review on Reviewable"/>](https://reviewable.io/reviews/apache/spark/3644)
<!-- Reviewable:end -->
Author: Cheng Lian <lian@databricks.com>
Closes#3644 from liancheng/parquet-tests and squashes the following commits:
800e745 [Cheng Lian] Enforces ordering of test output
3bb8731 [Cheng Lian] Refactors HiveParquetSuite
aa2cb2e [Cheng Lian] Decouples ParquetTest and TestSQLContext
7b43a68 [Cheng Lian] Updates ParquetTest Scaladoc
7f07af0 [Cheng Lian] Adds a new set of Parquet test suites
In BroadcastHashJoin, currently it is using a hard coded value (5 minutes) to wait for the execution and broadcast of the small table.
In my opinion, it should be a configurable value since broadcast may exceed 5 minutes in some case, like in a busy/congested network environment.
Author: Jacky Li <jacky.likun@huawei.com>
Closes#3133 from jackylk/timeout-config and squashes the following commits:
733ac08 [Jacky Li] add spark.sql.broadcastTimeout in SQLConf.scala
557acd4 [Jacky Li] switch to sqlContext.getConf
81a5e20 [Jacky Li] make wait time configurable in BroadcastHashJoin
In `HashOuterJoin.scala`, spark read data from both side of join operation before zip them together. It is a waste for memory. We are trying to read data from only one side, put them into a hashmap, and then generate the `JoinedRow` with data from other side one by one.
Currently, we could only do this optimization for `left outer join` and `right outer join`. For `full outer join`, we will do something in another issue.
for
table test_csv contains 1 million records
table dim_csv contains 10 thousand records
SQL:
`select * from test_csv a left outer join dim_csv b on a.key = b.key`
the result is:
master:
```
CSV: 12671 ms
CSV: 9021 ms
CSV: 9200 ms
Current Mem Usage:787788984
```
after patch：
```
CSV: 10382 ms
CSV: 7543 ms
CSV: 7469 ms
Current Mem Usage:208145728
```
Author: tianyi <tianyi@asiainfo-linkage.com>
Author: tianyi <tianyi.asiainfo@gmail.com>
Closes#3375 from tianyi/SPARK-4483 and squashes the following commits:
72a8aec [tianyi] avoid having mutable state stored inside of the task
99c5c97 [tianyi] performance optimization
d2f94d7 [tianyi] fix bug: missing output when the join-key is null.
2be45d1 [tianyi] fix spell bug
1f2c6f1 [tianyi] remove commented codes
a676de6 [tianyi] optimize some codes
9e7d5b5 [tianyi] remove commented old codes
838707d [tianyi] Optimization about reduce memory costs during the HashOuterJoin
The problem is `codegenEnabled` is `val`, but it uses a `val` `sqlContext`, which can be override by subclasses. Here is a simple example to show this issue.
```Scala
scala> :paste
// Entering paste mode (ctrl-D to finish)
abstract class Foo {
protected val sqlContext = "Foo"
val codegenEnabled: Boolean = {
println(sqlContext) // it will call subclass's `sqlContext` which has not yet been initialized.
if (sqlContext != null) {
true
} else {
false
}
}
}
class Bar extends Foo {
override val sqlContext = "Bar"
}
println(new Bar().codegenEnabled)
// Exiting paste mode, now interpreting.
null
false
defined class Foo
defined class Bar
```
We should make `sqlContext` `final` to prevent subclasses from overriding it incorrectly.
Author: zsxwing <zsxwing@gmail.com>
Closes#3660 from zsxwing/SPARK-4812 and squashes the following commits:
1cbb623 [zsxwing] Make `sqlContext` final to prevent subclasses from overriding it incorrectly
Author: jerryshao <saisai.shao@intel.com>
Closes#3698 from jerryshao/SPARK-4847 and squashes the following commits:
4741130 [jerryshao] Make later added extraStrategies effect when calling strategies
When I use Parquet File as a output file using ParquetOutputFormat#getDefaultWorkFile, the file name is not zero padded while RDD#saveAsText does zero padding.
Author: Sasaki Toru <sasakitoa@nttdata.co.jp>
Closes#3602 from sasakitoa/parquet-zeroPadding and squashes the following commits:
6b0e58f [Sasaki Toru] Merge branch 'master' of git://github.com/apache/spark into parquet-zeroPadding
20dc79d [Sasaki Toru] Fixed the name of Parquet File generated by AppendingParquetOutputFormat
Unpersist a uncached RDD, will not raise exception, for example:
```
val data = Array(1, 2, 3, 4, 5)
val distData = sc.parallelize(data)
distData.unpersist(true)
```
But the `SchemaRDD` will raise exception if the `SchemaRDD` is not cached. Since `SchemaRDD` is the subclasses of the `RDD`, we should follow the same behavior.
Author: Cheng Hao <hao.cheng@intel.com>
Closes#3572 from chenghao-intel/try_uncache and squashes the following commits:
50a7a89 [Cheng Hao] SchemaRDD.unpersist() should not raise exception if it is not persisted
Author: Jacky Li <jacky.likun@huawei.com>
Closes#3630 from jackylk/remove and squashes the following commits:
150e7e0 [Jacky Li] remove unnecessary import
Author: Michael Armbrust <michael@databricks.com>
Closes#3613 from marmbrus/parquetPartitionPruning and squashes the following commits:
4f138f8 [Michael Armbrust] Use catalyst for partition pruning in newParquet.
Author: Jacky Li <jacky.likun@huawei.com>
Closes#3585 from jackylk/remove and squashes the following commits:
045423d [Jacky Li] remove unnecessary import
This is a very small fix that catches one specific exception and returns an empty table. #3441 will address this in a more principled way.
Author: Michael Armbrust <michael@databricks.com>
Closes#3586 from marmbrus/fixEmptyParquet and squashes the following commits:
2781d9f [Michael Armbrust] Handle empty lists for newParquet
04dd376 [Michael Armbrust] Avoid exception when reading empty parquet data through Hive
val jsc = new org.apache.spark.api.java.JavaSparkContext(sc)
val jhc = new org.apache.spark.sql.hive.api.java.JavaHiveContext(jsc)
val nrdd = jhc.hql("select null from spark_test.for_test")
println(nrdd.schema)
Then the error is thrown as follows:
scala.MatchError: NullType (of class org.apache.spark.sql.catalyst.types.NullType$)
at org.apache.spark.sql.types.util.DataTypeConversions$.asJavaDataType(DataTypeConversions.scala:43)
Author: YanTangZhai <hakeemzhai@tencent.com>
Author: yantangzhai <tyz0303@163.com>
Author: Michael Armbrust <michael@databricks.com>
Closes#3538 from YanTangZhai/MatchNullType and squashes the following commits:
e052dff [yantangzhai] [SPARK-4676] [SQL] JavaSchemaRDD.schema may throw NullType MatchError if sql has null
4b4bb34 [yantangzhai] [SPARK-4676] [SQL] JavaSchemaRDD.schema may throw NullType MatchError if sql has null
896c7b7 [yantangzhai] fix NullType MatchError in JavaSchemaRDD when sql has null
6e643f8 [YanTangZhai] Merge pull request #11 from apache/master
e249846 [YanTangZhai] Merge pull request #10 from apache/master
d26d982 [YanTangZhai] Merge pull request #9 from apache/master
76d4027 [YanTangZhai] Merge pull request #8 from apache/master
03b62b0 [YanTangZhai] Merge pull request #7 from apache/master
8a00106 [YanTangZhai] Merge pull request #6 from apache/master
cbcba66 [YanTangZhai] Merge pull request #3 from apache/master
cdef539 [YanTangZhai] Merge pull request #1 from apache/master
Author: baishuo <vc_java@hotmail.com>
Closes#3526 from baishuo/master-trycatch and squashes the following commits:
d446e14 [baishuo] correct the code style
b36bf96 [baishuo] correct the code style
ae0e447 [baishuo] add finally to avoid resource leak
Spark SQL has embeded sqrt and abs but DSL doesn't support those functions.
Author: Kousuke Saruta <sarutak@oss.nttdata.co.jp>
Closes#3401 from sarutak/dsl-missing-operator and squashes the following commits:
07700cf [Kousuke Saruta] Modified Literal(null, NullType) to Literal(null) in DslQuerySuite
8f366f8 [Kousuke Saruta] Merge branch 'master' of git://git.apache.org/spark into dsl-missing-operator
1b88e2e [Kousuke Saruta] Merge branch 'master' of git://git.apache.org/spark into dsl-missing-operator
0396f89 [Kousuke Saruta] Added sqrt and abs to Spark SQL DSL
A very small nit update.
Author: Reynold Xin <rxin@databricks.com>
Closes#3552 from rxin/license-header and squashes the following commits:
df8d1a4 [Reynold Xin] Indent license header properly for interfaces.scala.
Author: ravipesala <ravindra.pesala@huawei.com>
Closes#3516 from ravipesala/ddl_doc and squashes the following commits:
d101fdf [ravipesala] Style issues fixed
d2238cd [ravipesala] Corrected documentation
Supporting multi column support in countDistinct function like count(distinct c1,c2..) in Spark SQL
Author: ravipesala <ravindra.pesala@huawei.com>
Author: Michael Armbrust <michael@databricks.com>
Closes#3511 from ravipesala/countdistinct and squashes the following commits:
cc4dbb1 [ravipesala] style
070e12a [ravipesala] Supporting multi column support in count(distinct c1,c2..) in Spark SQL
group tab is missing for scaladoc
Author: Jacky Li <jacky.likun@gmail.com>
Closes#3458 from jackylk/patch-7 and squashes the following commits:
0121a70 [Jacky Li] add @group tab in limit() and count()
Re-implement the Python broadcast using file:
1) serialize the python object using cPickle, write into disks.
2) Create a wrapper in JVM (for the dumped file), it read data from during serialization
3) Using TorrentBroadcast or HttpBroadcast to transfer the data (compressed) into executors
4) During deserialization, writing the data into disk.
5) Passing the path into Python worker, read data from disk and unpickle it into python object, until the first access.
It fixes the performance regression introduced in #2659, has similar performance as 1.1, but support object larger than 2G, also improve the memory efficiency (only one compressed copy in driver and executor).
Testing with a 500M broadcast and 4 tasks (excluding the benefit from reused worker in 1.2):
name | 1.1 | 1.2 with this patch | improvement
---------|--------|---------|--------
python-broadcast-w-bytes | 25.20 | 9.33 | 170.13% |
python-broadcast-w-set | 4.13 | 4.50 | -8.35% |
Testing with 100 tasks (16 CPUs):
name | 1.1 | 1.2 with this patch | improvement
---------|--------|---------|--------
python-broadcast-w-bytes | 38.16 | 8.40 | 353.98%
python-broadcast-w-set | 23.29 | 9.59 | 142.80%
Author: Davies Liu <davies@databricks.com>
Closes#3417 from davies/pybroadcast and squashes the following commits:
50a58e0 [Davies Liu] address comments
b98de1d [Davies Liu] disable gc while unpickle
e5ee6b9 [Davies Liu] support large string
09303b8 [Davies Liu] read all data into memory
dde02dd [Davies Liu] improve performance of python broadcast
When we use ORDER BY clause, at first, attributes referenced by projection are resolved (1).
And then, attributes referenced at ORDER BY clause are resolved (2).
But when resolving attributes referenced at ORDER BY clause, the resolution result generated in (1) is discarded so for example, following query fails.
SELECT c1 + c2 FROM mytable ORDER BY c1;
The query above fails because when resolving the attribute reference 'c1', the resolution result of 'c2' is discarded.
Author: Kousuke Saruta <sarutak@oss.nttdata.co.jp>
Closes#3363 from sarutak/SPARK-4487 and squashes the following commits:
fd314f3 [Kousuke Saruta] Fixed attribute resolution logic in Analyzer
6e60c20 [Kousuke Saruta] Fixed conflicts
cb5b7e9 [Kousuke Saruta] Added test case for SPARK-4487
282d529 [Kousuke Saruta] Fixed attributes reference resolution error
b6123e6 [Kousuke Saruta] Merge branch 'master' of git://git.apache.org/spark into concat-feature
317b7fb [Kousuke Saruta] WIP
This PR is a workaround for SPARK-4479. Two changes are introduced: when merge sort is bypassed in `ExternalSorter`,
1. also bypass RDD elements buffering as buffering is the reason that `MutableRow` backed row objects must be copied, and
2. avoids defensive copies in `Exchange` operator
<!-- Reviewable:start -->
[<img src="https://reviewable.io/review_button.png" height=40 alt="Review on Reviewable"/>](https://reviewable.io/reviews/apache/spark/3422)
<!-- Reviewable:end -->
Author: Cheng Lian <lian@databricks.com>
Closes#3422 from liancheng/avoids-defensive-copies and squashes the following commits:
591f2e9 [Cheng Lian] Passes all shuffle suites
0c3c91e [Cheng Lian] Fixes shuffle write metrics when merge sort is bypassed
ed5df3c [Cheng Lian] Fixes styling changes
f75089b [Cheng Lian] Avoids unnecessary defensive copies when sort based shuffle is on
Goals:
- Support for accessing parquet using SQL but not requiring Hive (thus allowing support of parquet tables with decimal columns)
- Support for folder based partitioning with automatic discovery of available partitions
- Caching of file metadata
See scaladoc of `ParquetRelation2` for more details.
Author: Michael Armbrust <michael@databricks.com>
Closes#3269 from marmbrus/newParquet and squashes the following commits:
1dd75f1 [Michael Armbrust] Pass all paths for FileInputFormat at once.
645768b [Michael Armbrust] Review comments.
abd8e2f [Michael Armbrust] Alternative implementation of parquet based on the datasources API.
938019e [Michael Armbrust] Add an experimental interface to data sources that exposes catalyst expressions.
e9d2641 [Michael Armbrust] logging / formatting improvements.
Sample code in the description of SchemaRDD.where is not correct
Author: Jacky Li <jacky.likun@gmail.com>
Closes#3344 from jackylk/patch-6 and squashes the following commits:
62cd126 [Jacky Li] [SQL] fix function description mistake
Executing sum distinct for empty table throws `java.lang.UnsupportedOperationException: empty.reduceLeft`.
Author: Takuya UESHIN <ueshin@happy-camper.st>
Closes#3184 from ueshin/issues/SPARK-4318 and squashes the following commits:
8168c42 [Takuya UESHIN] Merge branch 'master' into issues/SPARK-4318
66fdb0a [Takuya UESHIN] Re-refine aggregate functions.
6186eb4 [Takuya UESHIN] Fix Sum of GeneratedAggregate.
d2975f6 [Takuya UESHIN] Refine Sum and Average of GeneratedAggregate.
1bba675 [Takuya UESHIN] Refine Sum, SumDistinct and Average functions.
917e533 [Takuya UESHIN] Use aggregate instead of groupBy().
1a5f874 [Takuya UESHIN] Add tests to be executed as non-partial aggregation.
a5a57d2 [Takuya UESHIN] Fix empty Average.
22799dc [Takuya UESHIN] Fix empty Sum and SumDistinct.
65b7dd2 [Takuya UESHIN] Fix empty sum distinct.
The relational operator '<=>' is not working in Spark SQL. Same works in Spark HiveQL
Author: ravipesala <ravindra.pesala@huawei.com>
Closes#3387 from ravipesala/<=> and squashes the following commits:
7198e90 [ravipesala] Supporting relational operator '<=>' in Spark SQL
This PR enables the Web UI storage tab to show the in-memory table name instead of the mysterious query plan string as the name of the in-memory columnar RDD.
Note that after #2501, a single columnar RDD can be shared by multiple in-memory tables, as long as their query results are the same. In this case, only the first cached table name is shown. For example:
```sql
CACHE TABLE first AS SELECT * FROM src;
CACHE TABLE second AS SELECT * FROM src;
```
The Web UI only shows "In-memory table first".
<!-- Reviewable:start -->
[<img src="https://reviewable.io/review_button.png" height=40 alt="Review on Reviewable"/>](https://reviewable.io/reviews/apache/spark/3383)
<!-- Reviewable:end -->
Author: Cheng Lian <lian@databricks.com>
Closes#3383 from liancheng/columnar-rdd-name and squashes the following commits:
071907f [Cheng Lian] Fixes tests
12ddfa6 [Cheng Lian] Names in-memory columnar RDD with corresponding table name
Author: Marcelo Vanzin <vanzin@cloudera.com>
Closes#3277 from vanzin/version-1.3 and squashes the following commits:
7c3c396 [Marcelo Vanzin] Added temp repo to sbt build.
5f404ff [Marcelo Vanzin] Add another exclusion.
19457e7 [Marcelo Vanzin] Update old version to 1.2, add temporary 1.2 repo.
3c8d705 [Marcelo Vanzin] Workaround for MIMA checks.
e940810 [Marcelo Vanzin] Bumping version to 1.3.0-SNAPSHOT.
