https://issues.apache.org/jira/browse/SPARK-7713
I tested the performance with the following code:
```scala
import sqlContext._
import sqlContext.implicits._
(1 to 5000).foreach { i =>
val df = (1 to 1000).map(j => (j, s"str$j")).toDF("a", "b").save(s"/tmp/partitioned/i=$i")
}
sqlContext.sql("""
CREATE TEMPORARY TABLE partitionedParquet
USING org.apache.spark.sql.parquet
OPTIONS (
path '/tmp/partitioned'
)""")
table("partitionedParquet").explain(true)
```
In our master `explain` takes 40s in my laptop. With this PR, `explain` takes 14s.
Author: Yin Huai <yhuai@databricks.com>
Closes#6252 from yhuai/broadcastHadoopConf and squashes the following commits:
6fa73df [Yin Huai] Address comments of Josh and Andrew.
807fbf9 [Yin Huai] Make the new buildScan and SqlNewHadoopRDD private sql.
e393555 [Yin Huai] Cheng's comments.
2eb53bb [Yin Huai] Use a shared broadcast Hadoop Configuration for partitioned HadoopFsRelations.
follow up for #5806
Author: scwf <wangfei1@huawei.com>
Closes#6164 from scwf/FunctionRegistry and squashes the following commits:
15e6697 [scwf] use catalogconf in FunctionRegistry
```
select explode(map(value, key)) from src;
```
Throws exception
```
org.apache.spark.sql.AnalysisException: The number of aliases supplied in the AS clause does not match the number of columns output by the UDTF expected 2 aliases but got _c0 ;
at org.apache.spark.sql.catalyst.analysis.CheckAnalysis$class.failAnalysis(CheckAnalysis.scala:38)
at org.apache.spark.sql.catalyst.analysis.Analyzer.failAnalysis(Analyzer.scala:43)
at org.apache.spark.sql.catalyst.analysis.Analyzer$ResolveGenerate$.org$apache$spark$sql$catalyst$analysis$Analyzer$ResolveGenerate$$makeGeneratorOutput(Analyzer.scala:605)
at org.apache.spark.sql.catalyst.analysis.Analyzer$ResolveGenerate$$anonfun$apply$16$$anonfun$22.apply(Analyzer.scala:562)
at org.apache.spark.sql.catalyst.analysis.Analyzer$ResolveGenerate$$anonfun$apply$16$$anonfun$22.apply(Analyzer.scala:548)
at scala.collection.TraversableLike$$anonfun$flatMap$1.apply(TraversableLike.scala:251)
at scala.collection.TraversableLike$$anonfun$flatMap$1.apply(TraversableLike.scala:251)
at scala.collection.mutable.ResizableArray$class.foreach(ResizableArray.scala:59)
at scala.collection.mutable.ArrayBuffer.foreach(ArrayBuffer.scala:47)
at scala.collection.TraversableLike$class.flatMap(TraversableLike.scala:251)
at scala.collection.AbstractTraversable.flatMap(Traversable.scala:105)
at org.apache.spark.sql.catalyst.analysis.Analyzer$ResolveGenerate$$anonfun$apply$16.applyOrElse(Analyzer.scala:548)
at org.apache.spark.sql.catalyst.analysis.Analyzer$ResolveGenerate$$anonfun$apply$16.applyOrElse(Analyzer.scala:538)
at org.apache.spark.sql.catalyst.trees.TreeNode$$anonfun$3.apply(TreeNode.scala:222)
```
Author: Cheng Hao <hao.cheng@intel.com>
Closes#6178 from chenghao-intel/explode and squashes the following commits:
916fbe9 [Cheng Hao] add more strict rules for TGF alias
5c3f2c5 [Cheng Hao] fix bug in unit test
e1d93ab [Cheng Hao] Add more unit test
19db09e [Cheng Hao] resolve names for generator in projection
In `DataFrame.describe()`, the `count` aggregate produces an integer, the `avg` and `stdev` aggregates produce doubles, and `min` and `max` aggregates can produce varying types depending on what type of column they're applied to. As a result, we should cast all aggregate results to String so that `describe()`'s output types match its declared output schema.
Author: Josh Rosen <joshrosen@databricks.com>
Closes#6218 from JoshRosen/SPARK-7687 and squashes the following commits:
146b615 [Josh Rosen] Fix R test.
2974bd5 [Josh Rosen] Cast to string type instead
f206580 [Josh Rosen] Cast to double to fix SPARK-7687
307ecbf [Josh Rosen] Add failing regression test for SPARK-7687
This PR is based on #6081, thanks adrian-wang.
Closes#6081
Author: Daoyuan Wang <daoyuan.wang@intel.com>
Author: Davies Liu <davies@databricks.com>
Closes#6230 from davies/range and squashes the following commits:
d3ce5fe [Davies Liu] add tests
789eda5 [Davies Liu] add range() in Python
4590208 [Davies Liu] Merge commit 'refs/pull/6081/head' of github.com:apache/spark into range
cbf5200 [Daoyuan Wang] let's add python support in a separate PR
f45e3b2 [Daoyuan Wang] remove redundant toLong
617da76 [Daoyuan Wang] fix safe marge for corner cases
867c417 [Daoyuan Wang] fix
13dbe84 [Daoyuan Wang] update
bd998ba [Daoyuan Wang] update comments
d3a0c1b [Daoyuan Wang] add range api()
A follow-up to #6244.
Author: Michael Armbrust <michael@databricks.com>
Closes#6247 from marmbrus/fixOrcTests and squashes the following commits:
e39ee1b [Michael Armbrust] [SQL] Fix serializability of ORC table scan
Fix break caused by merging #6225 and #6194.
Author: Michael Armbrust <michael@databricks.com>
Closes#6244 from marmbrus/fixOrcBuildBreak and squashes the following commits:
b10e47b [Michael Armbrust] [HOTFIX] Fix ORC Build break
This PR revert #5404, change to pass the version of python in driver into JVM, check it in worker before deserializing closure, then it can works with different major version of Python.
Author: Davies Liu <davies@databricks.com>
Closes#6203 from davies/py_version and squashes the following commits:
b8fb76e [Davies Liu] fix test
6ce5096 [Davies Liu] use string for version
47c6278 [Davies Liu] check python version of worker with driver
This PR introduces several performance optimizations to `HadoopFsRelation` and `ParquetRelation2`:
1. Moving `FileStatus` listing from `DataSourceStrategy` into a cache within `HadoopFsRelation`.
This new cache generalizes and replaces the one used in `ParquetRelation2`.
This also introduces an interface change: to reuse cached `FileStatus` objects, `HadoopFsRelation.buildScan` methods now receive `Array[FileStatus]` instead of `Array[String]`.
1. When Parquet task side metadata reading is enabled, skip reading row group information when reading Parquet footers.
This is basically what PR #5334 does. Also, now we uses `ParquetFileReader.readAllFootersInParallel` to read footers in parallel.
Another optimization in question is, instead of asking `HadoopFsRelation.buildScan` to return an `RDD[Row]` for a single selected partition and then union them all, we ask it to return an `RDD[Row]` for all selected partitions. This optimization is based on the fact that Hadoop configuration broadcasting used in `NewHadoopRDD` takes 34% time in the following microbenchmark. However, this complicates data source user code because user code must merge partition values manually.
To check the cost of broadcasting in `NewHadoopRDD`, I also did microbenchmark after removing the `broadcast` call in `NewHadoopRDD`. All results are shown below.
### Microbenchmark
#### Preparation code
Generating a partitioned table with 50k partitions, 1k rows per partition:
```scala
import sqlContext._
import sqlContext.implicits._
for (n <- 0 until 500) {
val data = for {
p <- (n * 10) until ((n + 1) * 10)
i <- 0 until 1000
} yield (i, f"val_$i%04d", f"$p%04d")
data.
toDF("a", "b", "p").
write.
partitionBy("p").
mode("append").
parquet(path)
}
```
#### Benchmarking code
```scala
import sqlContext._
import sqlContext.implicits._
import org.apache.spark.sql.types._
import com.google.common.base.Stopwatch
val path = "hdfs://localhost:9000/user/lian/5k"
def benchmark(n: Int)(f: => Unit) {
val stopwatch = new Stopwatch()
def run() = {
stopwatch.reset()
stopwatch.start()
f
stopwatch.stop()
stopwatch.elapsedMillis()
}
val records = (0 until n).map(_ => run())
(0 until n).foreach(i => println(s"Round $i: ${records(i)} ms"))
println(s"Average: ${records.sum / n.toDouble} ms")
}
benchmark(3) { read.parquet(path).explain(extended = true) }
```
#### Results
Before:
```
Round 0: 72528 ms
Round 1: 68938 ms
Round 2: 65372 ms
Average: 68946.0 ms
```
After:
```
Round 0: 59499 ms
Round 1: 53645 ms
Round 2: 53844 ms
Round 3: 49093 ms
Round 4: 50555 ms
Average: 53327.2 ms
```
Also removing Hadoop configuration broadcasting:
(Note that I was testing on a local laptop, thus network cost is pretty low.)
```
Round 0: 15806 ms
Round 1: 14394 ms
Round 2: 14699 ms
Round 3: 15334 ms
Round 4: 14123 ms
Average: 14871.2 ms
```
Author: Cheng Lian <lian@databricks.com>
Closes#6225 from liancheng/spark-7673 and squashes the following commits:
2d58a2b [Cheng Lian] Skips reading row group information when using task side metadata reading
7aa3748 [Cheng Lian] Optimizes FileStatusCache by introducing a map from parent directories to child files
ba41250 [Cheng Lian] Reuses HadoopFsRelation FileStatusCache in ParquetRelation2
3d278f7 [Cheng Lian] Fixes a bug when reading a single Parquet data file
b84612a [Cheng Lian] Fixes Scala style issue
6a08b02 [Cheng Lian] WIP: Moves file status cache into HadoopFSRelation
cc liancheng marmbrus
Author: Yin Huai <yhuai@databricks.com>
Closes#6130 from yhuai/directOutput and squashes the following commits:
312b07d [Yin Huai] A data source can use spark.sql.sources.outputCommitterClass to override the output committer.
A modified version of https://github.com/apache/spark/pull/6110, use `semanticEquals` to make it more efficient.
Author: Wenchen Fan <cloud0fan@outlook.com>
Closes#6173 from cloud-fan/7269 and squashes the following commits:
e4a3cc7 [Wenchen Fan] address comments
cc02045 [Wenchen Fan] consider elements length equal
d7ff8f4 [Wenchen Fan] fix 7269
spark-sql>
> explain extended
> select * from (
> select key from src union all
> select key from src) t;
now the spark plan will print children in argString
```
== Physical Plan ==
Union[ HiveTableScan key#1, (MetastoreRelation default, src, None), None,
HiveTableScan key#3, (MetastoreRelation default, src, None), None]
HiveTableScan key#1, (MetastoreRelation default, src, None), None
HiveTableScan key#3, (MetastoreRelation default, src, None), None
```
after this patch:
```
== Physical Plan ==
Union
HiveTableScan [key#1], (MetastoreRelation default, src, None), None
HiveTableScan [key#3], (MetastoreRelation default, src, None), None
```
I have tested this locally
Author: scwf <wangfei1@huawei.com>
Closes#6144 from scwf/fix-argString and squashes the following commits:
1a642e0 [scwf] fix treenode argString
This PR updates PR #6135 authored by zhzhan from Hortonworks.
----
This PR implements a Spark SQL data source for accessing ORC files.
> **NOTE**
>
> Although ORC is now an Apache TLP, the codebase is still tightly coupled with Hive. That's why the new ORC data source is under `org.apache.spark.sql.hive` package, and must be used with `HiveContext`. However, it doesn't require existing Hive installation to access ORC files.
1. Saving/loading ORC files without contacting Hive metastore
1. Support for complex data types (i.e. array, map, and struct)
1. Aware of common optimizations provided by Spark SQL:
- Column pruning
- Partitioning pruning
- Filter push-down
1. Schema evolution support
1. Hive metastore table conversion
This PR also include initial work done by scwf from Huawei (PR #3753).
Author: Zhan Zhang <zhazhan@gmail.com>
Author: Cheng Lian <lian@databricks.com>
Closes#6194 from liancheng/polishing-orc and squashes the following commits:
55ecd96 [Cheng Lian] Reorganizes ORC test suites
d4afeed [Cheng Lian] Addresses comments
21ada22 [Cheng Lian] Adds @since and @Experimental annotations
128bd3b [Cheng Lian] ORC filter bug fix
d734496 [Cheng Lian] Polishes the ORC data source
2650a42 [Zhan Zhang] resolve review comments
3c9038e [Zhan Zhang] resolve review comments
7b3c7c5 [Zhan Zhang] save mode fix
f95abfd [Zhan Zhang] reuse test suite
7cc2c64 [Zhan Zhang] predicate fix
4e61c16 [Zhan Zhang] minor change
305418c [Zhan Zhang] orc data source support
<!-- Reviewable:start -->
[<img src="https://reviewable.io/review_button.png" height=40 alt="Review on Reviewable"/>](https://reviewable.io/reviews/apache/spark/6091)
<!-- Reviewable:end -->
Author: Cheng Lian <lian@databricks.com>
Closes#6091 from liancheng/spark-7570 and squashes the following commits:
8ff07e8 [Cheng Lian] Ignores _temporary during partition discovery
Replace the DriverQuirks with JdbcDialect(s) (and MySQLDialect/PostgresDialect)
and allow developers to change the dialects on the fly (for new JDBCRRDs only).
Some types (like an unsigned 64bit number) can be trivially mapped to java.
The status quo is that the RRD will fail to load.
This patch makes it possible to overwrite the type mapping to read e.g.
64Bit numbers as strings and handle them afterwards in software.
JDBCSuite has an example that maps all types to String, which should always
work (at the cost of extra code afterwards).
As a side effect it should now be possible to develop simple dialects
out-of-tree and even with spark-shell.
Author: Rene Treffer <treffer@measite.de>
Closes#5555 from rtreffer/jdbc-dialects and squashes the following commits:
3cbafd7 [Rene Treffer] [SPARK-6888] ignore classes belonging to changed API in MIMA report
fe7e2e8 [Rene Treffer] [SPARK-6888] Make the jdbc driver handling user-definable
JIRA: https://issues.apache.org/jira/browse/SPARK-7299
When connecting with oracle db through jdbc, the precision and scale of `BigDecimal` object returned by `ResultSet.getBigDecimal` is not correctly matched to the table schema reported by `ResultSetMetaData.getPrecision` and `ResultSetMetaData.getScale`.
So in case you insert a value like `19999` into a column with `NUMBER(12, 2)` type, you get through a `BigDecimal` object with scale as 0. But the dataframe schema has correct type as `DecimalType(12, 2)`. Thus, after you save the dataframe into parquet file and then retrieve it, you will get wrong result `199.99`.
Because it is reported to be problematic on jdbc connection with oracle db. It might be difficult to add test case for it. But according to the user's test on JIRA, it solves this problem.
Author: Liang-Chi Hsieh <viirya@gmail.com>
Closes#5833 from viirya/jdbc_decimal_precision and squashes the following commits:
69bc2b5 [Liang-Chi Hsieh] Merge remote-tracking branch 'upstream/master' into jdbc_decimal_precision
928f864 [Liang-Chi Hsieh] Add comments.
5f9da94 [Liang-Chi Hsieh] Set up Decimal's precision and scale according to table schema instead of returned BigDecimal.
Learnt a lesson from SPARK-7655: Spark should avoid to use `scala.concurrent.ExecutionContext.Implicits.global` because the user may submit blocking actions to `scala.concurrent.ExecutionContext.Implicits.global` and exhaust all threads in it. This could crash Spark. So Spark should always use its own thread pools for safety.
This PR removes all usages of `scala.concurrent.ExecutionContext.Implicits.global` and uses proper thread pools to replace them.
Author: zsxwing <zsxwing@gmail.com>
Closes#6223 from zsxwing/SPARK-7693 and squashes the following commits:
a33ff06 [zsxwing] Decrease the max thread number from 1024 to 128
cf4b3fc [zsxwing] Remove "import scala.concurrent.ExecutionContext.Implicits.global"
It's a follow-up of https://github.com/apache/spark/pull/5154, we can speed up scala udf evaluation by create type converter in advance.
Author: Wenchen Fan <cloud0fan@outlook.com>
Closes#6182 from cloud-fan/tmp and squashes the following commits:
241cfe9 [Wenchen Fan] use converter in ScalaUdf
Author: Michael Armbrust <michael@databricks.com>
Closes#6167 from marmbrus/configureIsolation and squashes the following commits:
6147cbe [Michael Armbrust] filter other conf
22cc3bc7 [Michael Armbrust] Merge remote-tracking branch 'origin/master' into configureIsolation
07476ee [Michael Armbrust] filter empty prefixes
dfdf19c [Michael Armbrust] [SPARK-6906][SQL] Allow configuration of classloader isolation for hive
In `SparkStrategies`, `RunnableDescribeCommand` is called with the output attributes of the table being described rather than the attributes for the `describe` command's output. I discovered this issue because it caused type conversion errors in some UnsafeRow conversion code that I'm writing.
Author: Josh Rosen <joshrosen@databricks.com>
Closes#6217 from JoshRosen/SPARK-7686 and squashes the following commits:
953a344 [Josh Rosen] Fix SPARK-7686 with a simple change in SparkStrategies.
a4eec9f [Josh Rosen] Add failing regression test for SPARK-7686
JIRA: https://issues.apache.org/jira/browse/SPARK-7447
`MetadataCache` in `ParquetRelation2` is annotated as `transient`. When `ParquetRelation2` is deserialized, we ask `MetadataCache` to refresh and perform schema merging again. It is time-consuming especially for very many parquet files.
With the new `FSBasedParquetRelation`, although `MetadataCache` is not `transient` now, `MetadataCache.refresh()` still performs schema merging again when the relation is deserialized.
Author: Liang-Chi Hsieh <viirya@gmail.com>
Closes#6012 from viirya/without_remerge_schema and squashes the following commits:
2663957 [Liang-Chi Hsieh] Merge remote-tracking branch 'upstream/master' into without_remerge_schema
6ac7d93 [Liang-Chi Hsieh] Merge remote-tracking branch 'upstream/master' into without_remerge_schema
b0fc09b [Liang-Chi Hsieh] Don't generate and merge parquetSchema multiple times.
This case clause is already covered by the one above, and generates a compilation warning.
Author: Cheng Lian <lian@databricks.com>
Closes#6214 from liancheng/remove-unreachable-code and squashes the following commits:
c38ca7c [Cheng Lian] Removes an unreachable case clause
Also moved all the deprecated functions into one place for SQLContext and DataFrame, and updated tests to use the new API.
Author: Reynold Xin <rxin@databricks.com>
Closes#6210 from rxin/df-writer-reader-jdbc and squashes the following commits:
7465c2c [Reynold Xin] Fixed unit test.
118e609 [Reynold Xin] Updated tests.
3441b57 [Reynold Xin] Updated javadoc.
13cdd1c [Reynold Xin] [SPARK-7654][SQL] Move JDBC into DataFrame's reader/writer interface.
We forgot an assignment there.
/cc rxin
Author: Cheng Lian <lian@databricks.com>
Closes#6212 from liancheng/fix-df-writer and squashes the following commits:
711fbb0 [Cheng Lian] Adds a test case
3b72d78 [Cheng Lian] Fixes DataFrameWriter.mode(String)
Because both `AkkaRpcEndpointRef.ask` and `BroadcastHashJoin` uses `scala.concurrent.ExecutionContext.Implicits.global`. However, because the tasks in `BroadcastHashJoin` are usually long-running tasks, which will occupy all threads in `global`. Then `ask` cannot get a chance to process the replies.
For `ask`, actually the tasks are very simple, so we can use `MoreExecutors.sameThreadExecutor()`. For `BroadcastHashJoin`, it's better to use `ThreadUtils.newDaemonCachedThreadPool`.
Author: zsxwing <zsxwing@gmail.com>
Closes#6200 from zsxwing/SPARK-7655-2 and squashes the following commits:
cfdc605 [zsxwing] Remove redundant imort and minor doc fix
cf83153 [zsxwing] Add "sameThread" and "newDaemonCachedThreadPool with maxThreadNumber" to ThreadUtils
08ad0ee [zsxwing] Remove 'scala.concurrent.ExecutionContext.Implicits.global' in 'ask' and 'BroadcastHashJoin'
This patch introduces DataFrameWriter and DataFrameReader.
DataFrameReader interface, accessible through SQLContext.read, contains methods that create DataFrames. These methods used to reside in SQLContext. Example usage:
```scala
sqlContext.read.json("...")
sqlContext.read.parquet("...")
```
DataFrameWriter interface, accessible through DataFrame.write, implements a builder pattern to avoid the proliferation of options in writing DataFrame out. It currently implements:
- mode
- format (e.g. "parquet", "json")
- options (generic options passed down into data sources)
- partitionBy (partitioning columns)
Example usage:
```scala
df.write.mode("append").format("json").partitionBy("date").saveAsTable("myJsonTable")
```
TODO:
- [ ] Documentation update
- [ ] Move JDBC into reader / writer?
- [ ] Deprecate the old interfaces
- [ ] Move the generic load interface into reader.
- [ ] Update example code and documentation
Author: Reynold Xin <rxin@databricks.com>
Closes#6175 from rxin/reader-writer and squashes the following commits:
b146c95 [Reynold Xin] Deprecation of old APIs.
bd8abdf [Reynold Xin] Fixed merge conflict.
26abea2 [Reynold Xin] Added general load methods.
244fbec [Reynold Xin] Added equivalent to example.
4f15d92 [Reynold Xin] Added documentation for partitionBy.
7e91611 [Reynold Xin] [SPARK-7654][SQL] DataFrameReader and DataFrameWriter for input/output API.
Otherwise, it will cause stack overflow when there are many partitions.
Author: Yin Huai <yhuai@databricks.com>
Closes#6162 from yhuai/partitionUnionedRDD and squashes the following commits:
fa016d8 [Yin Huai] Explicitly create UnionRDD.
JIRA: https://issues.apache.org/jira/browse/SPARK-7098
The WHERE clause with timstamp shows inconsistent results. This pr fixes it.
Author: Liang-Chi Hsieh <viirya@gmail.com>
Closes#5682 from viirya/consistent_timestamp and squashes the following commits:
171445a [Liang-Chi Hsieh] Merge remote-tracking branch 'upstream/master' into consistent_timestamp
4e98520 [Liang-Chi Hsieh] Make the WHERE clause with timestamp show consistent result.
Add an `explode` function for dataframes and modify the analyzer so that single table generating functions can be present in a select clause along with other expressions. There are currently the following restrictions:
- only top level TGFs are allowed (i.e. no `select(explode('list) + 1)`)
- only one may be present in a single select to avoid potentially confusing implicit Cartesian products.
TODO:
- [ ] Python
Author: Michael Armbrust <michael@databricks.com>
Closes#6107 from marmbrus/explodeFunction and squashes the following commits:
7ee2c87 [Michael Armbrust] whitespace
6f80ba3 [Michael Armbrust] Update dataframe.py
c176c89 [Michael Armbrust] Merge remote-tracking branch 'origin/master' into explodeFunction
81b5da3 [Michael Armbrust] style
d3faa05 [Michael Armbrust] fix self join case
f9e1e3e [Michael Armbrust] fix python, add since
4f0d0a9 [Michael Armbrust] Merge remote-tracking branch 'origin/master' into explodeFunction
e710fe4 [Michael Armbrust] add java and python
52ca0dc [Michael Armbrust] [SPARK-7548][SQL] Add explode function for dataframes.
A follow-up of https://github.com/apache/spark/pull/5624
Author: Wenchen Fan <cloud0fan@outlook.com>
Closes#6142 from cloud-fan/tmp and squashes the following commits:
971a92b [Wenchen Fan] use plan instead of execute
24c5ffe [Wenchen Fan] rename apply
for example:
table: src(key string, value string)
sql: with v1 as(select key, count(value) over (partition by key) cnt_val from src), v2 as(select v1.key, v1_lag.cnt_val from v1, v1 v1_lag where v1.key = v1_lag.key) select * from v2 limit 5;
then will analyze fail when resolving conflicting references in Join:
'Limit 5
'Project [*]
'Subquery v2
'Project ['v1.key,'v1_lag.cnt_val]
'Filter ('v1.key = 'v1_lag.key)
'Join Inner, None
Subquery v1
Project [key#95,cnt_val#94L]
Window [key#95,value#96], [HiveWindowFunction#org.apache.hadoop.hive.ql.udf.generic.GenericUDAFCount(value#96) WindowSpecDefinition [key#95], [], ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING AS cnt_val#94L], WindowSpecDefinition [key#95], [], ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING
Project [key#95,value#96]
MetastoreRelation default, src, None
Subquery v1_lag
Subquery v1
Project [key#97,cnt_val#94L]
Window [key#97,value#98], [HiveWindowFunction#org.apache.hadoop.hive.ql.udf.generic.GenericUDAFCount(value#98) WindowSpecDefinition [key#97], [], ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING AS cnt_val#94L], WindowSpecDefinition [key#97], [], ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING
Project [key#97,value#98]
MetastoreRelation default, src, None
Conflicting attributes: cnt_val#94L
Author: linweizhong <linweizhong@huawei.com>
Closes#6114 from Sephiroth-Lin/spark-7595 and squashes the following commits:
f8f2637 [linweizhong] Add unit test
dfe9169 [linweizhong] Handle windowExpression with self join
Same issue as #6095
cc liancheng
Author: zsxwing <zsxwing@gmail.com>
Closes#6136 from zsxwing/hotfix and squashes the following commits:
4beea54 [zsxwing] Use 'new Job' in fsBasedParquet.scala
This patch introduces a new shuffle manager that enhances the existing sort-based shuffle with a new cache-friendly sort algorithm that operates directly on binary data. The goals of this patch are to lower memory usage and Java object overheads during shuffle and to speed up sorting. It also lays groundwork for follow-up patches that will enable end-to-end processing of serialized records.
The new shuffle manager, `UnsafeShuffleManager`, can be enabled by setting `spark.shuffle.manager=tungsten-sort` in SparkConf.
The new shuffle manager uses directly-managed memory to implement several performance optimizations for certain types of shuffles. In cases where the new performance optimizations cannot be applied, the new shuffle manager delegates to SortShuffleManager to handle those shuffles.
UnsafeShuffleManager's optimizations will apply when _all_ of the following conditions hold:
- The shuffle dependency specifies no aggregation or output ordering.
- The shuffle serializer supports relocation of serialized values (this is currently supported
by KryoSerializer and Spark SQL's custom serializers).
- The shuffle produces fewer than 16777216 output partitions.
- No individual record is larger than 128 MB when serialized.
In addition, extra spill-merging optimizations are automatically applied when the shuffle compression codec supports concatenation of serialized streams. This is currently supported by Spark's LZF serializer.
At a high-level, UnsafeShuffleManager's design is similar to Spark's existing SortShuffleManager. In sort-based shuffle, incoming records are sorted according to their target partition ids, then written to a single map output file. Reducers fetch contiguous regions of this file in order to read their portion of the map output. In cases where the map output data is too large to fit in memory, sorted subsets of the output can are spilled to disk and those on-disk files are merged to produce the final output file.
UnsafeShuffleManager optimizes this process in several ways:
- Its sort operates on serialized binary data rather than Java objects, which reduces memory consumption and GC overheads. This optimization requires the record serializer to have certain properties to allow serialized records to be re-ordered without requiring deserialization. See SPARK-4550, where this optimization was first proposed and implemented, for more details.
- It uses a specialized cache-efficient sorter (UnsafeShuffleExternalSorter) that sorts arrays of compressed record pointers and partition ids. By using only 8 bytes of space per record in the sorting array, this fits more of the array into cache.
- The spill merging procedure operates on blocks of serialized records that belong to the same partition and does not need to deserialize records during the merge.
- When the spill compression codec supports concatenation of compressed data, the spill merge simply concatenates the serialized and compressed spill partitions to produce the final output partition. This allows efficient data copying methods, like NIO's `transferTo`, to be used and avoids the need to allocate decompression or copying buffers during the merge.
The shuffle read path is unchanged.
This patch is similar to [SPARK-4550](http://issues.apache.org/jira/browse/SPARK-4550) / #4450 but uses a slightly different implementation. The `unsafe`-based implementation featured in this patch lays the groundwork for followup patches that will enable sorting to operate on serialized data pages that will be prepared by Spark SQL's new `unsafe` operators (such as the new aggregation operator introduced in #5725).
### Future work
There are several tasks that build upon this patch, which will be left to future work:
- [SPARK-7271](https://issues.apache.org/jira/browse/SPARK-7271) Redesign / extend the shuffle interfaces to accept binary data as input. The goal here is to let us bypass serialization steps in cases where the sort input is produced by an operator that operates directly on binary data.
- Extension / redesign of the `Serializer` API. We can add new methods which allow serializers to determine the size requirements for serializing objects and for serializing objects directly to a specified memory address (similar to how `UnsafeRowConverter` works in Spark SQL).
<!-- Reviewable:start -->
[<img src="https://reviewable.io/review_button.png" height=40 alt="Review on Reviewable"/>](https://reviewable.io/reviews/apache/spark/5868)
<!-- Reviewable:end -->
Author: Josh Rosen <joshrosen@databricks.com>
Closes#5868 from JoshRosen/unsafe-sort and squashes the following commits:
ef0a86e [Josh Rosen] Fix scalastyle errors
7610f2f [Josh Rosen] Add tests for proper cleanup of shuffle data.
d494ffe [Josh Rosen] Fix deserialization of JavaSerializer instances.
52a9981 [Josh Rosen] Fix some bugs in the address packing code.
51812a7 [Josh Rosen] Change shuffle manager sort name to tungsten-sort
4023fa4 [Josh Rosen] Add @Private annotation to some Java classes.
de40b9d [Josh Rosen] More comments to try to explain metrics code
df07699 [Josh Rosen] Attempt to clarify confusing metrics update code
5e189c6 [Josh Rosen] Track time spend closing / flushing files; split TimeTrackingOutputStream into separate file.
d5779c6 [Josh Rosen] Merge remote-tracking branch 'origin/master' into unsafe-sort
c2ce78e [Josh Rosen] Fix a missed usage of MAX_PARTITION_ID
e3b8855 [Josh Rosen] Cleanup in UnsafeShuffleWriter
4a2c785 [Josh Rosen] rename 'sort buffer' to 'pointer array'
6276168 [Josh Rosen] Remove ability to disable spilling in UnsafeShuffleExternalSorter.
57312c9 [Josh Rosen] Clarify fileBufferSize units
2d4e4f4 [Josh Rosen] Address some minor comments in UnsafeShuffleExternalSorter.
fdcac08 [Josh Rosen] Guard against overflow when expanding sort buffer.
85da63f [Josh Rosen] Cleanup in UnsafeShuffleSorterIterator.
0ad34da [Josh Rosen] Fix off-by-one in nextInt() call
56781a1 [Josh Rosen] Rename UnsafeShuffleSorter to UnsafeShuffleInMemorySorter
e995d1a [Josh Rosen] Introduce MAX_SHUFFLE_OUTPUT_PARTITIONS.
e58a6b4 [Josh Rosen] Add more tests for PackedRecordPointer encoding.
4f0b770 [Josh Rosen] Attempt to implement proper shuffle write metrics.
d4e6d89 [Josh Rosen] Update to bit shifting constants
69d5899 [Josh Rosen] Remove some unnecessary override vals
8531286 [Josh Rosen] Add tests that automatically trigger spills.
7c953f9 [Josh Rosen] Add test that covers UnsafeShuffleSortDataFormat.swap().
e1855e5 [Josh Rosen] Fix a handful of misc. IntelliJ inspections
39434f9 [Josh Rosen] Avoid integer multiplication overflow in getMemoryUsage (thanks FindBugs!)
1e3ad52 [Josh Rosen] Delete unused ByteBufferOutputStream class.
ea4f85f [Josh Rosen] Roll back an unnecessary change in Spillable.
ae538dc [Josh Rosen] Document UnsafeShuffleManager.
ec6d626 [Josh Rosen] Add notes on maximum # of supported shuffle partitions.
0d4d199 [Josh Rosen] Bump up shuffle.memoryFraction to make tests pass.
b3b1924 [Josh Rosen] Properly implement close() and flush() in DummySerializerInstance.
1ef56c7 [Josh Rosen] Revise compression codec support in merger; test cross product of configurations.
b57c17f [Josh Rosen] Disable some overly-verbose logs that rendered DEBUG useless.
f780fb1 [Josh Rosen] Add test demonstrating which compression codecs support concatenation.
4a01c45 [Josh Rosen] Remove unnecessary log message
27b18b0 [Josh Rosen] That for inserting records AT the max record size.
fcd9a3c [Josh Rosen] Add notes + tests for maximum record / page sizes.
9d1ee7c [Josh Rosen] Fix MiMa excludes for ShuffleWriter change
fd4bb9e [Josh Rosen] Use own ByteBufferOutputStream rather than Kryo's
67d25ba [Josh Rosen] Update Exchange operator's copying logic to account for new shuffle manager
8f5061a [Josh Rosen] Strengthen assertion to check partitioning
01afc74 [Josh Rosen] Actually read data in UnsafeShuffleWriterSuite
1929a74 [Josh Rosen] Update to reflect upstream ShuffleBlockManager -> ShuffleBlockResolver rename.
e8718dd [Josh Rosen] Merge remote-tracking branch 'origin/master' into unsafe-sort
9b7ebed [Josh Rosen] More defensive programming RE: cleaning up spill files and memory after errors
7cd013b [Josh Rosen] Begin refactoring to enable proper tests for spilling.
722849b [Josh Rosen] Add workaround for transferTo() bug in merging code; refactor tests.
9883e30 [Josh Rosen] Merge remote-tracking branch 'origin/master' into unsafe-sort
b95e642 [Josh Rosen] Refactor and document logic that decides when to spill.
1ce1300 [Josh Rosen] More minor cleanup
5e8cf75 [Josh Rosen] More minor cleanup
e67f1ea [Josh Rosen] Remove upper type bound in ShuffleWriter interface.
cfe0ec4 [Josh Rosen] Address a number of minor review comments:
8a6fe52 [Josh Rosen] Rename UnsafeShuffleSpillWriter to UnsafeShuffleExternalSorter
11feeb6 [Josh Rosen] Update TODOs related to shuffle write metrics.
b674412 [Josh Rosen] Merge remote-tracking branch 'origin/master' into unsafe-sort
aaea17b [Josh Rosen] Add comments to UnsafeShuffleSpillWriter.
4f70141 [Josh Rosen] Fix merging; now passes UnsafeShuffleSuite tests.
133c8c9 [Josh Rosen] WIP towards testing UnsafeShuffleWriter.
f480fb2 [Josh Rosen] WIP in mega-refactoring towards shuffle-specific sort.
57f1ec0 [Josh Rosen] WIP towards packed record pointers for use in optimized shuffle sort.
69232fd [Josh Rosen] Enable compressible address encoding for off-heap mode.
7ee918e [Josh Rosen] Re-order imports in tests
3aeaff7 [Josh Rosen] More refactoring and cleanup; begin cleaning iterator interfaces
3490512 [Josh Rosen] Misc. cleanup
f156a8f [Josh Rosen] Hacky metrics integration; refactor some interfaces.
2776aca [Josh Rosen] First passing test for ExternalSorter.
5e100b2 [Josh Rosen] Super-messy WIP on external sort
595923a [Josh Rosen] Remove some unused variables.
8958584 [Josh Rosen] Fix bug in calculating free space in current page.
f17fa8f [Josh Rosen] Add missing newline
c2fca17 [Josh Rosen] Small refactoring of SerializerPropertiesSuite to enable test re-use:
b8a09fe [Josh Rosen] Back out accidental log4j.properties change
bfc12d3 [Josh Rosen] Add tests for serializer relocation property.
240864c [Josh Rosen] Remove PrefixComputer and require prefix to be specified as part of insert()
1433b42 [Josh Rosen] Store record length as int instead of long.
026b497 [Josh Rosen] Re-use a buffer in UnsafeShuffleWriter
0748458 [Josh Rosen] Port UnsafeShuffleWriter to Java.
87e721b [Josh Rosen] Renaming and comments
d3cc310 [Josh Rosen] Flag that SparkSqlSerializer2 supports relocation
e2d96ca [Josh Rosen] Expand serializer API and use new function to help control when new UnsafeShuffle path is used.
e267cee [Josh Rosen] Fix compilation of UnsafeSorterSuite
9c6cf58 [Josh Rosen] Refactor to use DiskBlockObjectWriter.
253f13e [Josh Rosen] More cleanup
8e3ec20 [Josh Rosen] Begin code cleanup.
4d2f5e1 [Josh Rosen] WIP
3db12de [Josh Rosen] Minor simplification and sanity checks in UnsafeSorter
767d3ca [Josh Rosen] Fix invalid range in UnsafeSorter.
e900152 [Josh Rosen] Add test for empty iterator in UnsafeSorter
57a4ea0 [Josh Rosen] Make initialSize configurable in UnsafeSorter
abf7bfe [Josh Rosen] Add basic test case.
81d52c5 [Josh Rosen] WIP on UnsafeSorter
JavaTypeInference into catalyst
types.DateUtils into catalyst
CacheManager into execution
DefaultParserDialect into catalyst
Author: Reynold Xin <rxin@databricks.com>
Closes#6108 from rxin/sql-rename and squashes the following commits:
3fc9613 [Reynold Xin] Fixed import ordering.
83d9ff4 [Reynold Xin] Fixed codegen tests.
e271e86 [Reynold Xin] mima
f4e24a6 [Reynold Xin] [SQL] Move some classes into packages that are more appropriate.
Optimize the case of `project(_, sort)` , a example is:
`select key from (select * from testData order by key) t`
before this PR:
```
== Parsed Logical Plan ==
'Project ['key]
'Subquery t
'Sort ['key ASC], true
'Project [*]
'UnresolvedRelation [testData], None
== Analyzed Logical Plan ==
Project [key#0]
Subquery t
Sort [key#0 ASC], true
Project [key#0,value#1]
Subquery testData
LogicalRDD [key#0,value#1], MapPartitionsRDD[1]
== Optimized Logical Plan ==
Project [key#0]
Sort [key#0 ASC], true
LogicalRDD [key#0,value#1], MapPartitionsRDD[1]
== Physical Plan ==
Project [key#0]
Sort [key#0 ASC], true
Exchange (RangePartitioning [key#0 ASC], 5), []
PhysicalRDD [key#0,value#1], MapPartitionsRDD[1]
```
after this PR
```
== Parsed Logical Plan ==
'Project ['key]
'Subquery t
'Sort ['key ASC], true
'Project [*]
'UnresolvedRelation [testData], None
== Analyzed Logical Plan ==
Project [key#0]
Subquery t
Sort [key#0 ASC], true
Project [key#0,value#1]
Subquery testData
LogicalRDD [key#0,value#1], MapPartitionsRDD[1]
== Optimized Logical Plan ==
Sort [key#0 ASC], true
Project [key#0]
LogicalRDD [key#0,value#1], MapPartitionsRDD[1]
== Physical Plan ==
Sort [key#0 ASC], true
Exchange (RangePartitioning [key#0 ASC], 5), []
Project [key#0]
PhysicalRDD [key#0,value#1], MapPartitionsRDD[1]
```
with this rule we will first do column pruning on the table and then do sorting.
Author: scwf <wangfei1@huawei.com>
This patch had conflicts when merged, resolved by
Committer: Michael Armbrust <michael@databricks.com>
Closes#5838 from scwf/pruning and squashes the following commits:
b00d833 [scwf] address michael's comment
e230155 [scwf] fix tests failure
b09b895 [scwf] improve column pruning
This PR migrates Parquet data source to the newly introduced `FSBasedRelation`. `FSBasedParquetRelation` is created to replace `ParquetRelation2`. Major differences are:
1. Partition discovery code has been factored out to `FSBasedRelation`
1. `AppendingParquetOutputFormat` is not used now. Instead, an anonymous subclass of `ParquetOutputFormat` is used to handle appending and writing dynamic partitions
1. When scanning partitioned tables, `FSBasedParquetRelation.buildScan` only builds an `RDD[Row]` for a single selected partition
1. `FSBasedParquetRelation` doesn't rely on Catalyst expressions for filter push down, thus it doesn't extend `CatalystScan` anymore
After migrating `JSONRelation` (which extends `CatalystScan`), we can remove `CatalystScan`.
<!-- Reviewable:start -->
[<img src="https://reviewable.io/review_button.png" height=40 alt="Review on Reviewable"/>](https://reviewable.io/reviews/apache/spark/6090)
<!-- Reviewable:end -->
Author: Cheng Lian <lian@databricks.com>
Closes#6090 from liancheng/parquet-migration and squashes the following commits:
6063f87 [Cheng Lian] Casts to OutputCommitter rather than FileOutputCommtter
bfd1cf0 [Cheng Lian] Fixes compilation error introduced while rebasing
f9ea56e [Cheng Lian] Adds ParquetRelation2 related classes to MiMa check whitelist
261d8c1 [Cheng Lian] Minor bug fix and more tests
db65660 [Cheng Lian] Migrates Parquet data source to FSBasedRelation
Some third-party UDTF extensions generate additional rows in the "GenericUDTF.close()" method, which is supported / documented by Hive.
https://cwiki.apache.org/confluence/display/Hive/DeveloperGuide+UDTF
However, Spark SQL ignores the "GenericUDTF.close()", and it causes bug while porting job from Hive to Spark SQL.
Author: Cheng Hao <hao.cheng@intel.com>
Closes#5383 from chenghao-intel/udtf_close and squashes the following commits:
98b4e4b [Cheng Hao] Support UDTF.close
Author: Cheng Lian <lian@databricks.com>
Closes#6123 from liancheng/remove-println and squashes the following commits:
03356b6 [Cheng Lian] Removes debugging println
We need to add a log entry before calling `abortTask`/`abortJob`. Otherwise, an exception from `abortTask`/`abortJob` will shadow the real cause.
cc liancheng
Author: Yin Huai <yhuai@databricks.com>
Closes#6105 from yhuai/logCause and squashes the following commits:
8dfe0d8 [Yin Huai] Log cause.
Author: Cheng Lian <lian@databricks.com>
Closes#6118 from liancheng/spark-7599 and squashes the following commits:
31e1bd6 [Cheng Lian] Don't restrict customized output committers to be subclasses of FileOutputCommitter
This makes HiveContext.analyzer overrideable.
Author: Santiago M. Mola <santi@mola.io>
Closes#6086 from smola/patch-3 and squashes the following commits:
8ece136 [Santiago M. Mola] [SPARK-7566][SQL] Add type to HiveContext.analyzer
This builds on https://github.com/apache/spark/pull/5932 and should close https://github.com/apache/spark/pull/5932 as well.
As an example:
```python
df.select(when(df['age'] == 2, 3).otherwise(4).alias("age")).collect()
```
Author: Reynold Xin <rxin@databricks.com>
Author: kaka1992 <kaka_1992@163.com>
Closes#6072 from rxin/when-expr and squashes the following commits:
8f49201 [Reynold Xin] Throw exception if otherwise is applied twice.
0455eda [Reynold Xin] Reset run-tests.
bfb9d9f [Reynold Xin] Updated documentation and test cases.
762f6a5 [Reynold Xin] Merge pull request #5932 from kaka1992/IFCASE
95724c6 [kaka1992] Update
8218d0a [kaka1992] Update
801009e [kaka1992] Update
76d6346 [kaka1992] [SPARK-7321][SQL] Add Column expression for conditional statements (if, case)
