## What changes were proposed in this pull request?
This PR proposes a new data-structure based on a vectorized hashmap that can be potentially _codegened_ in `TungstenAggregate` to speed up aggregates with group by. Micro-benchmarks show a 10x improvement over the current `BytesToBytes` aggregation map.
## How was this patch tested?
Intel(R) Core(TM) i7-4960HQ CPU 2.60GHz
BytesToBytesMap: Best/Avg Time(ms) Rate(M/s) Per Row(ns) Relative
-------------------------------------------------------------------------------------------
hash 108 / 119 96.9 10.3 1.0X
fast hash 63 / 70 166.2 6.0 1.7X
arrayEqual 70 / 73 150.8 6.6 1.6X
Java HashMap (Long) 141 / 200 74.3 13.5 0.8X
Java HashMap (two ints) 145 / 185 72.3 13.8 0.7X
Java HashMap (UnsafeRow) 499 / 524 21.0 47.6 0.2X
BytesToBytesMap (off Heap) 483 / 548 21.7 46.0 0.2X
BytesToBytesMap (on Heap) 485 / 562 21.6 46.2 0.2X
Vectorized Hashmap 54 / 60 193.7 5.2 2.0X
Author: Sameer Agarwal <sameer@databricks.com>
Closes#12055 from sameeragarwal/vectorized-hashmap.
### What changes were proposed in this pull request?
This PR removes the ANTLR3 based parser, and moves the new ANTLR4 based parser into the `org.apache.spark.sql.catalyst.parser package`.
### How was this patch tested?
Existing unit tests.
cc rxin andrewor14 yhuai
Author: Herman van Hovell <hvanhovell@questtec.nl>
Closes#12071 from hvanhovell/SPARK-14211.
## What changes were proposed in this pull request?
Major changes:
1. Implement `FileFormat.buildReader()` for the CSV data source.
1. Add an extra argument to `FileFormat.buildReader()`, `physicalSchema`, which is basically the result of `FileFormat.inferSchema` or user specified schema.
This argument is necessary because the CSV data source needs to know all the columns of the underlying files to read the file.
## How was this patch tested?
Existing tests should do the work.
Author: Cheng Lian <lian@databricks.com>
Closes#12002 from liancheng/spark-14206-csv-build-reader.
## What changes were proposed in this pull request?
This change resolves an issue where `DataFrameNaFunctions.fill` changes a `FloatType` column to a `DoubleType`. We also clarify the contract that replacement values will be cast to the column data type, which may change the replacement value when casting to a lower precision type.
## How was this patch tested?
This patch has associated unit tests.
Author: Travis Crawford <travis@medium.com>
Closes#11967 from traviscrawford/SPARK-14081-dataframena.
## What changes were proposed in this pull request?
This pr is to add a config to control the maximum number of files as even small files have a non-trivial fixed cost. The current packing can put a lot of small files together which cases straggler tasks.
## How was this patch tested?
I added tests to check if many files get split into partitions in FileSourceStrategySuite.
Author: Takeshi YAMAMURO <linguin.m.s@gmail.com>
Closes#12068 from maropu/SPARK-14259.
## What changes were proposed in this pull request?
In `ExpressionEncoder`, we use `constructorFor` to build `fromRowExpression` as the `deserializer` in `ObjectOperator`. It's kind of confusing, we should make the name consistent.
## How was this patch tested?
existing tests.
Author: Wenchen Fan <wenchen@databricks.com>
Closes#12058 from cloud-fan/rename.
## What changes were proposed in this pull request?
This PR implements buildReader for text data source and enable it in the new data source code path.
## How was this patch tested?
Existing tests.
Author: Wenchen Fan <wenchen@databricks.com>
Closes#11934 from cloud-fan/text.
#### What changes were proposed in this pull request?
This PR is to implement the following four Database-related DDL commands:
- `CREATE DATABASE|SCHEMA [IF NOT EXISTS] database_name`
- `DROP DATABASE [IF EXISTS] database_name [RESTRICT|CASCADE]`
- `DESCRIBE DATABASE [EXTENDED] db_name`
- `ALTER (DATABASE|SCHEMA) database_name SET DBPROPERTIES (property_name=property_value, ...)`
Another PR will be submitted to handle the unsupported commands. In the Database-related DDL commands, we will issue an error exception for `ALTER (DATABASE|SCHEMA) database_name SET OWNER [USER|ROLE] user_or_role`.
cc yhuai andrewor14 rxin Could you review the changes? Is it in the right direction? Thanks!
#### How was this patch tested?
Added a few test cases in `command/DDLSuite.scala` for testing DDL command execution in `SQLContext`. Since `HiveContext` also shares the same implementation, the existing test cases in `\hive` also verifies the correctness of these commands.
Author: gatorsmile <gatorsmile@gmail.com>
Author: xiaoli <lixiao1983@gmail.com>
Author: Xiao Li <xiaoli@Xiaos-MacBook-Pro.local>
Closes#12009 from gatorsmile/dbDDL.
## What changes were proposed in this pull request?
This PR brings the support for chained Python UDFs, for example
```sql
select udf1(udf2(a))
select udf1(udf2(a) + 3)
select udf1(udf2(a) + udf3(b))
```
Also directly chained unary Python UDFs are put in single batch of Python UDFs, others may require multiple batches.
For example,
```python
>>> sqlContext.sql("select double(double(1))").explain()
== Physical Plan ==
WholeStageCodegen
: +- Project [pythonUDF#10 AS double(double(1))#9]
: +- INPUT
+- !BatchPythonEvaluation double(double(1)), [pythonUDF#10]
+- Scan OneRowRelation[]
>>> sqlContext.sql("select double(double(1) + double(2))").explain()
== Physical Plan ==
WholeStageCodegen
: +- Project [pythonUDF#19 AS double((double(1) + double(2)))#16]
: +- INPUT
+- !BatchPythonEvaluation double((pythonUDF#17 + pythonUDF#18)), [pythonUDF#17,pythonUDF#18,pythonUDF#19]
+- !BatchPythonEvaluation double(2), [pythonUDF#17,pythonUDF#18]
+- !BatchPythonEvaluation double(1), [pythonUDF#17]
+- Scan OneRowRelation[]
```
TODO: will support multiple unrelated Python UDFs in one batch (another PR).
## How was this patch tested?
Added new unit tests for chained UDFs.
Author: Davies Liu <davies@databricks.com>
Closes#12014 from davies/py_udfs.
## What changes were proposed in this pull request?
This PR is a simple fix for an exception message to print `string[]` content correctly.
```java
String[] colPath = requestedSchema.getPaths().get(i);
...
- throw new IOException("Required column is missing in data file. Col: " + colPath);
+ throw new IOException("Required column is missing in data file. Col: " + Arrays.toString(colPath));
```
## How was this patch tested?
Manual.
Author: Dongjoon Hyun <dongjoon@apache.org>
Closes#12041 from dongjoon-hyun/fix_exception_message_with_string_array.
## What changes were proposed in this pull request?
Renames SQL option `spark.sql.parquet.fileScan` since now all `HadoopFsRelation` based data sources are being migrated to `FileScanRDD` code path.
## How was this patch tested?
None.
Author: Cheng Lian <lian@databricks.com>
Closes#12003 from liancheng/spark-14208-option-renaming.
## What changes were proposed in this pull request?
This PR implements buildReader for json data source and enable it in the new data source code path.
## How was this patch tested?
existing tests
Author: Wenchen Fan <wenchen@databricks.com>
Closes#11960 from cloud-fan/json.
## What changes were proposed in this pull request?
This adds a metric to parquet scans that measures the time in just the scan phase. This is
only possible when the scan returns ColumnarBatches, otherwise the overhead is too high.
This combined with the pipeline metric lets us easily see what percent of the time was
in the scan.
Author: Nong Li <nong@databricks.com>
Closes#12007 from nongli/spark-14210.
## What changes were proposed in this pull request?
This improves the Filter codegen for NULLs by deferring loading the values for IsNotNull.
Instead of generating code like:
boolean isNull = ...
int value = ...
if (isNull) continue;
we will generate:
boolean isNull = ...
if (isNull) continue;
int value = ...
This is useful since retrieving the values can be non-trivial (they can be dictionary encoded
among other things). This currently only works when the attribute comes from the column batch
but could be extended to other cases in the future.
## How was this patch tested?
On tpcds q55, this fixes the regression from introducing the IsNotNull predicates.
```
TPCDS Snappy: Best/Avg Time(ms) Rate(M/s) Per Row(ns)
--------------------------------------------------------------------------------
q55 4564 / 5036 25.2 39.6
q55 4064 / 4340 28.3 35.3
```
Author: Nong Li <nong@databricks.com>
Closes#11792 from nongli/spark-13981.
## What changes were proposed in this pull request?
After DataFrame and Dataset are merged, the trait `Queryable` becomes unnecessary as it has only one implementation. We should remove it.
## How was this patch tested?
existing tests.
Author: Wenchen Fan <wenchen@databricks.com>
Closes#12001 from cloud-fan/df-ds.
## What changes were proposed in this pull request?
Session catalog was added in #11750. However, it doesn't really support temporary functions properly; right now we only store the metadata in the form of `CatalogFunction`, but this doesn't make sense for temporary functions because there is no class name.
This patch moves the `FunctionRegistry` into the `SessionCatalog`. With this, the user can call `catalog.createTempFunction` and `catalog.lookupFunction` to use the function they registered previously. This is currently still dead code, however.
## How was this patch tested?
`SessionCatalogSuite`.
Author: Andrew Or <andrew@databricks.com>
Closes#11972 from andrewor14/temp-functions.
## What changes were proposed in this pull request?
Extract the workaround for HADOOP-10622 introduced by #11940 into UninterruptibleThread so that we can test and reuse it.
## How was this patch tested?
Unit tests
Author: Shixiong Zhu <shixiong@databricks.com>
Closes#11971 from zsxwing/uninterrupt.
## What changes were proposed in this pull request?
This patch addresses the remaining comments left in #11750 and #11918 after they are merged. For a full list of changes in this patch, just trace the commits.
## How was this patch tested?
`SessionCatalogSuite` and `CatalogTestCases`
Author: Andrew Or <andrew@databricks.com>
Closes#12006 from andrewor14/session-catalog-followup.
#### What changes were proposed in this pull request?
This PR adds all the current Spark SQL DDL commands to the new ANTLR 4 based SQL parser.
I have found a few inconsistencies in the current commands:
- Function has an alias field. This is actually the class name of the function.
- Partition specifications should contain nulls in some commands, and contain `None`s in others.
- `AlterTableSkewedLocation`: Should defines which columns have skewed values, and should allow us to define storage for each skewed combination of values. We currently only allow one value per field.
- `AlterTableSetFileFormat`: Should only have one file format, it currently supports both.
I have implemented all these comments like they were, and I propose to improve them in follow-up PRs.
#### How was this patch tested?
The existing DDLCommandSuite.
cc rxin andrewor14 yhuai
Author: Herman van Hovell <hvanhovell@questtec.nl>
Closes#12011 from hvanhovell/SPARK-14086.
## What changes were proposed in this pull request?
Currently, for the key that can not fit within a long, we build a hash map for UnsafeHashedRelation, it's converted to BytesToBytesMap after serialization and deserialization. We should build a BytesToBytesMap directly to have better memory efficiency.
In order to do that, BytesToBytesMap should support multiple (K,V) pair with the same K, Location.putNewKey() is renamed to Location.append(), which could append multiple values for the same key (same Location). `Location.newValue()` is added to find the next value for the same key.
## How was this patch tested?
Existing tests. Added benchmark for broadcast hash join with duplicated keys.
Author: Davies Liu <davies@databricks.com>
Closes#11870 from davies/map2.
### What changes were proposed in this pull request?
The current ANTLR3 parser is quite complex to maintain and suffers from code blow-ups. This PR introduces a new parser that is based on ANTLR4.
This parser is based on the [Presto's SQL parser](https://github.com/facebook/presto/blob/master/presto-parser/src/main/antlr4/com/facebook/presto/sql/parser/SqlBase.g4). The current implementation can parse and create Catalyst and SQL plans. Large parts of the HiveQl DDL and some of the DML functionality is currently missing, the plan is to add this in follow-up PRs.
This PR is a work in progress, and work needs to be done in the following area's:
- [x] Error handling should be improved.
- [x] Documentation should be improved.
- [x] Multi-Insert needs to be tested.
- [ ] Naming and package locations.
### How was this patch tested?
Catalyst and SQL unit tests.
Author: Herman van Hovell <hvanhovell@questtec.nl>
Closes#11557 from hvanhovell/ngParser.
#### What changes were proposed in this pull request?
This PR is to provide native parsing support for two DDL commands: ```Describe Database``` and ```Alter Database Set Properties```
Based on the Hive DDL document:
https://cwiki.apache.org/confluence/display/Hive/LanguageManual+DDL
##### 1. ALTER DATABASE
**Syntax:**
```SQL
ALTER (DATABASE|SCHEMA) database_name SET DBPROPERTIES (property_name=property_value, ...)
```
- `ALTER DATABASE` is to add new (key, value) pairs into `DBPROPERTIES`
##### 2. DESCRIBE DATABASE
**Syntax:**
```SQL
DESCRIBE DATABASE [EXTENDED] db_name
```
- `DESCRIBE DATABASE` shows the name of the database, its comment (if one has been set), and its root location on the filesystem. When `extended` is true, it also shows the database's properties
#### How was this patch tested?
Added the related test cases to `DDLCommandSuite`
Author: gatorsmile <gatorsmile@gmail.com>
Author: xiaoli <lixiao1983@gmail.com>
Author: Xiao Li <xiaoli@Xiaos-MacBook-Pro.local>
This patch had conflicts when merged, resolved by
Committer: Yin Huai <yhuai@databricks.com>
Closes#11977 from gatorsmile/parseAlterDatabase.
## What changes were proposed in this pull request?
This PR implements `FileFormat.buildReader()` for our ORC data source. It also fixed several minor styling issues related to `HadoopFsRelation` planning code path.
Note that `OrcNewInputFormat` doesn't rely on `OrcNewSplit` for creating `OrcRecordReader`s, plain `FileSplit` is just fine. That's why we can simply create the record reader with the help of `OrcNewInputFormat` and `FileSplit`.
## How was this patch tested?
Existing test cases should do the work
Author: Cheng Lian <lian@databricks.com>
Closes#11936 from liancheng/spark-14116-build-reader-for-orc.
### What changes were proposed in this pull request?
Based on the Hive DDL document https://cwiki.apache.org/confluence/display/Hive/LanguageManual+DDL
The syntax of DDL command for Drop Database is
```SQL
DROP (DATABASE|SCHEMA) [IF EXISTS] database_name [RESTRICT|CASCADE];
```
- If `IF EXISTS` is not specified, the default behavior is to issue a warning message if `database_name` does't exist
- `RESTRICT` is the default behavior.
This PR is to provide a native parsing support for `DROP DATABASE`.
#### How was this patch tested?
Added a test case `DDLCommandSuite`
Author: gatorsmile <gatorsmile@gmail.com>
Closes#11962 from gatorsmile/parseDropDatabase.
## What changes were proposed in this pull request?
1. merge consumeChild into consume()
2. always generate code for input variables and UnsafeRow, a plan can use eight of them.
## How was this patch tested?
Existing tests.
Author: Davies Liu <davies@databricks.com>
Closes#11975 from davies/gen_refactor.
## What changes were proposed in this pull request?
This PR fixes some newly added java-lint errors(unused-imports, line-lengsth).
## How was this patch tested?
Pass the Jenkins tests.
Author: Dongjoon Hyun <dongjoon@apache.org>
Closes#11968 from dongjoon-hyun/SPARK-14167.
## What changes were proposed in this pull request?
HDFSMetadataLog uses newer FileContext API to achieve atomic renaming. However, FileContext implementations may not exist for many scheme for which there may be FileSystem implementations. In those cases, rather than failing completely, we should fallback to the FileSystem based implementation, and log warning that there may be file consistency issues in case the log directory is concurrently modified.
In addition I have also added more tests to increase the code coverage.
## How was this patch tested?
Unit test.
Tested on cluster with custom file system.
Author: Tathagata Das <tathagata.das1565@gmail.com>
Closes#11925 from tdas/SPARK-14109.
## What changes were proposed in this pull request?
This PR moves flume back to Spark as per the discussion in the dev mail-list.
## How was this patch tested?
Existing Jenkins tests.
Author: Shixiong Zhu <shixiong@databricks.com>
Closes#11895 from zsxwing/move-flume-back.
## What changes were proposed in this pull request?
There is a potential dead-lock in Hadoop Shell.runCommand before 2.5.0 ([HADOOP-10622](https://issues.apache.org/jira/browse/HADOOP-10622)). If we interrupt some thread running Shell.runCommand, we may hit this issue.
This PR adds some protecion to prevent from interrupting the microBatchThread when we may run into Shell.runCommand. There are two places will call Shell.runCommand now:
- offsetLog.add
- FileStreamSource.getOffset
They will create a file using HDFS API and call Shell.runCommand to set the file permission.
## How was this patch tested?
Existing unit tests.
Author: Shixiong Zhu <shixiong@databricks.com>
Closes#11940 from zsxwing/workaround-for-HADOOP-10622.
## What changes were proposed in this pull request?
StateStoreCoordinator.reportActiveInstance is async, so subsequence state checks must be in eventually.
## How was this patch tested?
Jenkins tests
Author: Tathagata Das <tathagata.das1565@gmail.com>
Closes#11924 from tdas/state-store-flaky-fix.
## What changes were proposed in this pull request?
This PR is a minor cleanup task as part of https://issues.apache.org/jira/browse/SPARK-14008 to explicitly identify/catch the `UnsupportedOperationException` while initializing the vectorized parquet reader. Other exceptions will simply be thrown back to `SqlNewHadoopPartition`.
## How was this patch tested?
N/A (cleanup only; no new functionality added)
Author: Sameer Agarwal <sameer@databricks.com>
Closes#11950 from sameeragarwal/parquet-cleanup.
## What changes were proposed in this pull request?
As we have `CreateArray` and `CreateStruct`, we should also have `CreateMap`. This PR adds the `CreateMap` expression, and the DataFrame API, and python API.
## How was this patch tested?
various new tests.
Author: Wenchen Fan <wenchen@databricks.com>
Closes#11879 from cloud-fan/create_map.
## What changes were proposed in this pull request?
We ran into a problem today debugging some class loading problem during deserialization, and JVM was masking the underlying exception which made it very difficult to debug. We can however log the exceptions using try/catch ourselves in serialization/deserialization. The good thing is that all these methods are already using Utils.tryOrIOException, so we can just put the try catch and logging in a single place.
## How was this patch tested?
A logging change with a manual test.
Author: Reynold Xin <rxin@databricks.com>
Closes#11951 from rxin/SPARK-14149.
## What changes were proposed in this pull request?
This reopens#11836, which was merged but promptly reverted because it introduced flaky Hive tests.
## How was this patch tested?
See `CatalogTestCases`, `SessionCatalogSuite` and `HiveContextSuite`.
Author: Andrew Or <andrew@databricks.com>
Closes#11938 from andrewor14/session-catalog-again.
## What changes were proposed in this pull request?
Dataset has two variants of groupByKey, one for untyped and the other for typed. It actually doesn't make as much sense to have an untyped API here, since apps that want to use untyped APIs should just use the groupBy "DataFrame" API.
## How was this patch tested?
This patch removes a method, and removes the associated tests.
Author: Reynold Xin <rxin@databricks.com>
Closes#11949 from rxin/SPARK-14145.
## What changes were proposed in this pull request?
unionAll has been deprecated in SPARK-14088.
## How was this patch tested?
Should be covered by all existing tests.
Author: Reynold Xin <rxin@databricks.com>
Closes#11946 from rxin/SPARK-14142.
#### What changes were proposed in this pull request?
This PR is to support group by position in SQL. For example, when users input the following query
```SQL
select c1 as a, c2, c3, sum(*) from tbl group by 1, 3, c4
```
The ordinals are recognized as the positions in the select list. Thus, `Analyzer` converts it to
```SQL
select c1, c2, c3, sum(*) from tbl group by c1, c3, c4
```
This is controlled by the config option `spark.sql.groupByOrdinal`.
- When true, the ordinal numbers in group by clauses are treated as the position in the select list.
- When false, the ordinal numbers are ignored.
- Only convert integer literals (not foldable expressions). If found foldable expressions, ignore them.
- When the positions specified in the group by clauses correspond to the aggregate functions in select list, output an exception message.
- star is not allowed to use in the select list when users specify ordinals in group by
Note: This PR is taken from https://github.com/apache/spark/pull/10731. When merging this PR, please give the credit to zhichao-li
Also cc all the people who are involved in the previous discussion: rxin cloud-fan marmbrus yhuai hvanhovell adrian-wang chenghao-intel tejasapatil
#### How was this patch tested?
Added a few test cases for both positive and negative test cases.
Author: gatorsmile <gatorsmile@gmail.com>
Author: xiaoli <lixiao1983@gmail.com>
Author: Xiao Li <xiaoli@Xiaos-MacBook-Pro.local>
Closes#11846 from gatorsmile/groupByOrdinal.
#### What changes were proposed in this pull request?
This PR is to support star expansion in hash. For example,
```SQL
val structDf = testData2.select("a", "b").as("record")
structDf.select(hash($"*")
```
In addition, it refactors the codes for the rule `ResolveStar` and fixes a regression for star expansion in group by when using SQL API. For example,
```SQL
SELECT * FROM testData2 group by a, b
```
cc cloud-fan Now, the code for star resolution is much cleaner. The coverage is better. Could you check if this refactoring is good? Thanks!
#### How was this patch tested?
Added a few test cases to cover it.
Author: gatorsmile <gatorsmile@gmail.com>
Closes#11904 from gatorsmile/starResolution.
## What changes were proposed in this pull request?
`SessionCatalog`, introduced in #11750, is a catalog that keeps track of temporary functions and tables, and delegates metastore operations to `ExternalCatalog`. This functionality overlaps a lot with the existing `analysis.Catalog`.
As of this commit, `SessionCatalog` and `ExternalCatalog` will no longer be dead code. There are still things that need to be done after this patch, namely:
- SPARK-14013: Properly implement temporary functions in `SessionCatalog`
- SPARK-13879: Decide which DDL/DML commands to support natively in Spark
- SPARK-?????: Implement the ones we do want to support through `SessionCatalog`.
- SPARK-?????: Merge SQL/HiveContext
## How was this patch tested?
This is largely a refactoring task so there are no new tests introduced. The particularly relevant tests are `SessionCatalogSuite` and `ExternalCatalogSuite`.
Author: Andrew Or <andrew@databricks.com>
Author: Yin Huai <yhuai@databricks.com>
Closes#11836 from andrewor14/use-session-catalog.
This PR adds a new `Sink` implementation that writes out Parquet files. In order to correctly handle partial failures while maintaining exactly once semantics, the files for each batch are written out to a unique directory and then atomically appended to a metadata log. When a parquet based `DataSource` is initialized for reading, we first check for this log directory and use it instead of file listing when present.
Unit tests are added, as well as a stress test that checks the answer after non-deterministic injected failures.
Author: Michael Armbrust <michael@databricks.com>
Closes#11897 from marmbrus/fileSink.
## What changes were proposed in this pull request?
In this PR, I am implementing a new abstraction for management of streaming state data - State Store. It is a key-value store for persisting running aggregates for aggregate operations in streaming dataframes. The motivation and design is discussed here.
https://docs.google.com/document/d/1-ncawFx8JS5Zyfq1HAEGBx56RDet9wfVp_hDM8ZL254/edit#
## How was this patch tested?
- [x] Unit tests
- [x] Cluster tests
**Coverage from unit tests**
<img width="952" alt="screen shot 2016-03-21 at 3 09 40 pm" src="https://cloud.githubusercontent.com/assets/663212/13935872/fdc8ba86-ef76-11e5-93e8-9fa310472c7b.png">
## TODO
- [x] Fix updates() iterator to avoid duplicate updates for same key
- [x] Use Coordinator in ContinuousQueryManager
- [x] Plugging in hadoop conf and other confs
- [x] Unit tests
- [x] StateStore object lifecycle and methods
- [x] StateStoreCoordinator communication and logic
- [x] StateStoreRDD fault-tolerance
- [x] StateStoreRDD preferred location using StateStoreCoordinator
- [ ] Cluster tests
- [ ] Whether preferred locations are set correctly
- [ ] Whether recovery works correctly with distributed storage
- [x] Basic performance tests
- [x] Docs
Author: Tathagata Das <tathagata.das1565@gmail.com>
Closes#11645 from tdas/state-store.
## What changes were proposed in this pull request?
This PR adds support for TimestampType in the vectorized parquet reader
## How was this patch tested?
1. `VectorizedColumnReader` initially had a gating condition on `primitiveType.getPrimitiveTypeName() == PrimitiveType.PrimitiveTypeName.INT96)` that made us fall back on parquet-mr for handling timestamps. This condition is now removed.
2. The `ParquetHadoopFsRelationSuite` (that tests for all supported hive types -- including `TimestampType`) fails when the gating condition is removed (https://github.com/apache/spark/pull/11808) and should now pass with this change. Similarly, the `ParquetHiveCompatibilitySuite.SPARK-10177 timestamp` test that fails when the gating condition is removed, should now pass as well.
3. Added tests in `HadoopFsRelationTest` that test both the dictionary encoded and non-encoded versions across all supported datatypes.
Author: Sameer Agarwal <sameer@databricks.com>
Closes#11882 from sameeragarwal/timestamp-parquet.
## What changes were proposed in this pull request?
This PR rollback some changes in #11274 , which introduced some performance regression when do a simple aggregation on parquet scan with one integer column.
Does not really understand how this change introduce this huge impact, maybe related show JIT compiler inline functions. (saw very different stats from profiling).
## How was this patch tested?
Manually run the parquet reader benchmark, before this change:
```
Intel(R) Core(TM) i7-4558U CPU 2.80GHz
Int and String Scan: Best/Avg Time(ms) Rate(M/s) Per Row(ns) Relative
-------------------------------------------------------------------------------------------
SQL Parquet Vectorized 2391 / 3107 43.9 22.8 1.0X
```
After this change
```
Java HotSpot(TM) 64-Bit Server VM 1.7.0_60-b19 on Mac OS X 10.9.5
Intel(R) Core(TM) i7-4558U CPU 2.80GHz
Int and String Scan: Best/Avg Time(ms) Rate(M/s) Per Row(ns) Relative
-------------------------------------------------------------------------------------------
SQL Parquet Vectorized 2032 / 2626 51.6 19.4 1.0X```
Author: Davies Liu <davies@databricks.com>
Closes#11912 from davies/fix_regression.
This patch refactors the `MemoryStore` so that it can be tested without needing to construct / mock an entire `BlockManager`.
- The block manager's serialization- and compression-related methods have been moved from `BlockManager` to `SerializerManager`.
- `BlockInfoManager `is now passed directly to classes that need it, rather than being passed via the `BlockManager`.
- The `MemoryStore` now calls `dropFromMemory` via a new `BlockEvictionHandler` interface rather than directly calling the `BlockManager`. This change helps to enforce a narrow interface between the `MemoryStore` and `BlockManager` functionality and makes this interface easier to mock in tests.
- Several of the block unrolling tests have been moved from `BlockManagerSuite` into a new `MemoryStoreSuite`.
Author: Josh Rosen <joshrosen@databricks.com>
Closes#11899 from JoshRosen/reduce-memorystore-blockmanager-coupling.
## What changes were proposed in this pull request?
This PR does the renaming as suggested by marmbrus in [this comment][1].
## How was this patch tested?
Existing tests.
[1]: 6d37e1eb90 (commitcomment-16654694)
Author: Cheng Lian <lian@databricks.com>
Closes#11889 from liancheng/spark-13817-follow-up.
## What changes were proposed in this pull request?
Try to fix a flaky hang
## How was this patch tested?
Existing Jenkins test
Author: Shixiong Zhu <shixiong@databricks.com>
Closes#11909 from zsxwing/hotfix2.
## What changes were proposed in this pull request?
1. Deprecated unionAll. It is pretty confusing to have both "union" and "unionAll" when the two do the same thing in Spark but are different in SQL.
2. Rename reduce in KeyValueGroupedDataset to reduceGroups so it is more consistent with rest of the functions in KeyValueGroupedDataset. Also makes it more obvious what "reduce" and "reduceGroups" mean. Previously it was confusing because it could be reducing a Dataset, or just reducing groups.
3. Added a "name" function, which is more natural to name columns than "as" for non-SQL users.
4. Remove "subtract" function since it is just an alias for "except".
## How was this patch tested?
All changes should be covered by existing tests. Also added couple test cases to cover "name".
Author: Reynold Xin <rxin@databricks.com>
Closes#11908 from rxin/SPARK-14088.
## What changes were proposed in this pull request?
This PR updates `sql/README.md` according to the latest console output and removes some unused imports in `sql` module. This is done by manually, so there is no guarantee to remove all unused imports.
## How was this patch tested?
Manual.
Author: Dongjoon Hyun <dongjoon@apache.org>
Closes#11907 from dongjoon-hyun/update_sql_module.
## What changes were proposed in this pull request?
This fix tries to fix several SQL test warnings under the sql/core/src/test directory. The fixed warnings includes "[unchecked]", "[rawtypes]", and "[varargs]".
## How was this patch tested?
All existing tests passed.
Author: Yong Tang <yong.tang.github@outlook.com>
Closes#11857 from yongtang/SPARK-13401.
## What changes were proposed in this pull request?
Fix an issue that DataFrameReaderWriterSuite may hang forever.
## How was this patch tested?
Existing tests.
Author: Shixiong Zhu <shixiong@databricks.com>
Closes#11902 from zsxwing/hotfix.
## What changes were proposed in this pull request?
This patch moves StringToColumn implicit class into SQLImplicits. This was kept in SQLContext.implicits object for binary backward compatibility, in the Spark 1.x series. It makes more sense for this API to be in SQLImplicits since that's the single class that defines all the SQL implicits.
## How was this patch tested?
Should be covered by existing unit tests.
Author: Reynold Xin <rxin@databricks.com>
Author: Wenchen Fan <wenchen@databricks.com>
Closes#11878 from rxin/SPARK-14060.
## What changes were proposed in this pull request?
This patch changed the return type for SQLContext.range from `Dataset[Long]` (Scala primitive) to `Dataset[java.lang.Long]` (Java boxed long).
Previously, SPARK-13894 changed the return type of range from `Dataset[Row]` to `Dataset[Long]`. The problem is that due to https://issues.scala-lang.org/browse/SI-4388, Scala compiles primitive types in generics into just Object, i.e. range at bytecode level now just returns `Dataset[Object]`. This is really bad for Java users because they are losing type safety and also need to add a type cast every time they use range.
Talked to Jason Zaugg from Lightbend (Typesafe) who suggested the best approach is to return `Dataset[java.lang.Long]`. The downside is that when Scala users want to explicitly type a closure used on the dataset returned by range, they would need to use `java.lang.Long` instead of the Scala `Long`.
## How was this patch tested?
The signature change should be covered by existing unit tests and API tests. I also added a new test case in DatasetSuite for range.
Author: Reynold Xin <rxin@databricks.com>
Closes#11880 from rxin/SPARK-14063.
This PR relaxes the requirements of a `Sink` for structured streaming to only require idempotent appending of data. Previously the `Sink` needed to be able to transactionally append data while recording an opaque offset indicated how far in a stream we have processed.
In order to do this, a new write-ahead-log has been added to stream execution, which records the offsets that will are present in each batch. The log is created in the newly added `checkpointLocation`, which defaults to `${spark.sql.streaming.checkpointLocation}/${queryName}` but can be overriden by setting `checkpointLocation` in `DataFrameWriter`.
In addition to making sinks easier to write the addition of batchIds and a checkpoint location is done in anticipation of integration with the the `StateStore` (#11645).
Author: Michael Armbrust <michael@databricks.com>
Closes#11804 from marmbrus/batchIds.
SPARK-13774: IllegalArgumentException: Can not create a Path from an empty string for incorrect file path
**Overview:**
- If a non-existent path is given in this call
``
scala> sqlContext.read.format("csv").load("file-path-is-incorrect.csv")
``
it throws the following error:
`java.lang.IllegalArgumentException: Can not create a Path from an empty string` …..
`It gets called from inferSchema call in org.apache.spark.sql.execution.datasources.DataSource.resolveRelation`
- The purpose of this JIRA is to throw a better error message.
- With the fix, you will now get a _Path does not exist_ error message.
```
scala> sqlContext.read.format("csv").load("file-path-is-incorrect.csv")
org.apache.spark.sql.AnalysisException: Path does not exist: file:/Users/ksunitha/trunk/spark/file-path-is-incorrect.csv;
at org.apache.spark.sql.execution.datasources.DataSource$$anonfun$12.apply(DataSource.scala:215)
at org.apache.spark.sql.execution.datasources.DataSource$$anonfun$12.apply(DataSource.scala:204)
...
at org.apache.spark.sql.execution.datasources.DataSource.resolveRelation(DataSource.scala:204)
at org.apache.spark.sql.DataFrameReader.load(DataFrameReader.scala:131)
at org.apache.spark.sql.DataFrameReader.load(DataFrameReader.scala:141)
... 49 elided
```
**Details**
_Changes include:_
- Check if path exists or not in resolveRelation in DataSource, and throw an AnalysisException with message like “Path does not exist: $path”
- AnalysisException is thrown similar to the exceptions thrown in resolveRelation.
- The glob path and the non glob path is checked with minimal calls to path exists. If the globPath is empty, then it is a nonexistent glob pattern and an error will be thrown. In the scenario that it is not globPath, it is necessary to only check if the first element in the Seq is valid or not.
_Test modifications:_
- Changes went in for 3 tests to account for this error checking.
- SQLQuerySuite:test("run sql directly on files") – Error message needed to be updated.
- 2 tests failed in MetastoreDataSourcesSuite because they had a dummy path and so test is modified to give a tempdir and allow it to move past so it can continue to test the codepath it meant to test
_New Tests:_
2 new tests are added to DataFrameSuite to validate that glob and non-glob path will throw the new error message.
_Testing:_
Unit tests were run with the fix.
**Notes/Questions to reviewers:**
- There is some code duplication in DataSource.scala in resolveRelation method and also createSource with respect to getting the paths. I have not made any changes to the createSource codepath. Should we make the change there as well ?
- From other JIRAs, I know there is restructuring and changes going on in this area, not sure how that will affect these changes, but since this seemed like a starter issue, I looked into it. If we prefer not to add the overhead of the checks, or if there is a better place to do so, let me know.
I would appreciate your review. Thanks for your time and comments.
Author: Sunitha Kambhampati <skambha@us.ibm.com>
Closes#11775 from skambha/improve_errmsg.
## What changes were proposed in this pull request?
https://issues.apache.org/jira/browse/SPARK-13953
Currently, JSON data source creates a new field in `PERMISSIVE` mode for storing malformed string.
This field can be renamed via `spark.sql.columnNameOfCorruptRecord` option but it is a global configuration.
This PR make that option can be applied per read and can be specified via `option()`. This will overwrites `spark.sql.columnNameOfCorruptRecord` if it is set.
## How was this patch tested?
Unit tests were used and `./dev/run_tests` for coding style tests.
Author: hyukjinkwon <gurwls223@gmail.com>
Closes#11881 from HyukjinKwon/SPARK-13953.
## What changes were proposed in this pull request?
As we have completed the `SQLBuilder`, we can safely turn on native view by default.
## How was this patch tested?
existing tests.
Author: Wenchen Fan <wenchen@databricks.com>
Closes#11872 from cloud-fan/native-view.
This PR add implements the new `buildReader` interface for the Parquet `FileFormat`. An simple implementation of `FileScanRDD` is also included.
This code should be tested by the many existing tests for parquet.
Author: Michael Armbrust <michael@databricks.com>
Author: Sameer Agarwal <sameer@databricks.com>
Author: Nong Li <nong@databricks.com>
Closes#11709 from marmbrus/parquetReader.
## What changes were proposed in this pull request?
This patch adds support for reading `DecimalTypes` with high (> 18) precision in `VectorizedColumnReader`
## How was this patch tested?
1. `VectorizedColumnReader` initially had a gating condition on `primitiveType.getDecimalMetadata().getPrecision() > Decimal.MAX_LONG_DIGITS()` that made us fall back on parquet-mr for handling high-precision decimals. This condition is now removed.
2. In particular, the `ParquetHadoopFsRelationSuite` (that tests for all supported hive types -- including `DecimalType(25, 5)`) fails when the gating condition is removed (https://github.com/apache/spark/pull/11808) and should now pass with this change.
Author: Sameer Agarwal <sameer@databricks.com>
Closes#11869 from sameeragarwal/bigdecimal-parquet.
This PR resolves two issues:
First, expanding * inside aggregate functions of structs when using Dataframe/Dataset APIs. For example,
```scala
structDf.groupBy($"a").agg(min(struct($"record.*")))
```
Second, it improves the error messages when having invalid star usage when using Dataframe/Dataset APIs. For example,
```scala
pagecounts4PartitionsDS
.map(line => (line._1, line._3))
.toDF()
.groupBy($"_1")
.agg(sum("*") as "sumOccurances")
```
Before the fix, the invalid usage will issue a confusing error message, like:
```
org.apache.spark.sql.AnalysisException: cannot resolve '_1' given input columns _1, _2;
```
After the fix, the message is like:
```
org.apache.spark.sql.AnalysisException: Invalid usage of '*' in function 'sum'
```
cc: rxin nongli cloud-fan
Author: gatorsmile <gatorsmile@gmail.com>
Closes#11208 from gatorsmile/sumDataSetResolution.
## What changes were proposed in this pull request?
This patch merges DatasetHolder and DataFrameHolder. This makes more sense because DataFrame/Dataset are now one class.
In addition, fixed some minor issues with pull request #11732.
## How was this patch tested?
Updated existing unit tests that test these implicits.
Author: Reynold Xin <rxin@databricks.com>
Closes#11737 from rxin/SPARK-13898.
## What changes were proposed in this pull request?
WholeStageCodegen naturally breaks the execution into pipelines that are easier to
measure duration. This is more granular than the task timings (a task can be multiple
pipelines) and is integrated with the web ui.
We currently report total time (across all tasks), min/mask/median to get a sense of how long each is taking.
## How was this patch tested?
Manually tested looking at the web ui.
Author: Nong Li <nong@databricks.com>
Closes#11741 from nongli/spark-13916.
## What changes were proposed in this pull request?
There is only one exception: `PythonUDF`. However, I don't think the `PythonUDF#` prefix is useful, as we can only create python udf under python context. This PR removes the `PythonUDF#` prefix from `PythonUDF.toString`, so that it doesn't need to overrde `sql`.
## How was this patch tested?
existing tests.
Author: Wenchen Fan <wenchen@databricks.com>
Closes#11859 from cloud-fan/tmp.
## What changes were proposed in this pull request?
This PR generates code that get a value in each column from ```ColumnVector``` instead of creating ```InternalRow``` when ```ColumnarBatch``` is accessed. This PR improves benchmark program by up to 15%.
This PR consists of two parts:
1. Get an ```ColumnVector ``` by using ```ColumnarBatch.column()``` method
2. Get a value of each column by using ```rdd_col${COLIDX}.getInt(ROWIDX)``` instead of ```rdd_row.getInt(COLIDX)```
This is a motivated example.
````
sqlContext.conf.setConfString(SQLConf.PARQUET_VECTORIZED_READER_ENABLED.key, "true")
sqlContext.conf.setConfString(SQLConf.WHOLESTAGE_CODEGEN_ENABLED.key, "true")
val values = 10
withTempPath { dir =>
withTempTable("t1", "tempTable") {
sqlContext.range(values).registerTempTable("t1")
sqlContext.sql("select id % 2 as p, cast(id as INT) as id from t1")
.write.partitionBy("p").parquet(dir.getCanonicalPath)
sqlContext.read.parquet(dir.getCanonicalPath).registerTempTable("tempTable")
sqlContext.sql("select sum(p) from tempTable").collect
}
}
````
The original code
````java
...
/* 072 */ while (!shouldStop() && rdd_batchIdx < numRows) {
/* 073 */ InternalRow rdd_row = rdd_batch.getRow(rdd_batchIdx++);
/* 074 */ /*** CONSUME: TungstenAggregate(key=[], functions=[(sum(cast(p#4 as bigint)),mode=Partial,isDistinct=false)], output=[sum#10L]) */
/* 075 */ /* input[0, int] */
/* 076 */ boolean rdd_isNull = rdd_row.isNullAt(0);
/* 077 */ int rdd_value = rdd_isNull ? -1 : (rdd_row.getInt(0));
...
````
The code generated by this PR
````java
/* 072 */ while (!shouldStop() && rdd_batchIdx < numRows) {
/* 073 */ org.apache.spark.sql.execution.vectorized.ColumnVector rdd_col0 = rdd_batch.column(0);
/* 074 */ /*** CONSUME: TungstenAggregate(key=[], functions=[(sum(cast(p#4 as bigint)),mode=Partial,isDistinct=false)], output=[sum#10L]) */
/* 075 */ /* input[0, int] */
/* 076 */ boolean rdd_isNull = rdd_col0.getIsNull(rdd_batchIdx);
/* 077 */ int rdd_value = rdd_isNull ? -1 : (rdd_col0.getInt(rdd_batchIdx));
...
/* 128 */ rdd_batchIdx++;
/* 129 */ }
/* 130 */ if (shouldStop()) return;
````
Performance
Without this PR
````
model name : Intel(R) Xeon(R) CPU E5-2667 v2 3.30GHz
Partitioned Table: Best/Avg Time(ms) Rate(M/s) Per Row(ns) Relative
-------------------------------------------------------------------------------------------
Read data column 434 / 488 36.3 27.6 1.0X
Read partition column 302 / 346 52.1 19.2 1.4X
Read both columns 588 / 643 26.8 37.4 0.7X
````
With this PR
````
model name : Intel(R) Xeon(R) CPU E5-2667 v2 3.30GHz
Partitioned Table: Best/Avg Time(ms) Rate(M/s) Per Row(ns) Relative
-------------------------------------------------------------------------------------------
Read data column 392 / 516 40.1 24.9 1.0X
Read partition column 256 / 318 61.4 16.3 1.5X
Read both columns 523 / 539 30.1 33.3 0.7X
````
## How was this patch tested?
Tested by existing test suites and benchmark
Author: Kazuaki Ishizaki <ishizaki@jp.ibm.com>
Closes#11636 from kiszk/SPARK-13805.
## What changes were proposed in this pull request?
This PR try acquire the memory for hash map in shuffled hash join, fail the task if there is no enough memory (otherwise it could OOM the executor).
It also removed unused HashedRelation.
## How was this patch tested?
Existing unit tests. Manual tests with TPCDS Q78.
Author: Davies Liu <davies@databricks.com>
Closes#11826 from davies/cleanup_hash2.
## What changes were proposed in this pull request?
Ad-hoc Dataset API ScalaDoc fixes
## How was this patch tested?
By building and checking ScalaDoc locally.
Author: Cheng Lian <lian@databricks.com>
Closes#11862 from liancheng/ds-doc-fixes.
## What changes were proposed in this pull request?
When we validate an encoder, we may call `dataType` on unresolved expressions. This PR fix the validation so that we will resolve attributes first.
## How was this patch tested?
a new test in `DatasetSuite`
Author: Wenchen Fan <wenchen@databricks.com>
Closes#11816 from cloud-fan/encoder.
#### What changes were proposed in this pull request?
This PR is to support order by position in SQL, e.g.
```SQL
select c1, c2, c3 from tbl order by 1 desc, 3
```
should be equivalent to
```SQL
select c1, c2, c3 from tbl order by c1 desc, c3 asc
```
This is controlled by config option `spark.sql.orderByOrdinal`.
- When true, the ordinal numbers are treated as the position in the select list.
- When false, the ordinal number in order/sort By clause are ignored.
- Only convert integer literals (not foldable expressions). If found foldable expressions, ignore them
- This also works with select *.
**Question**: Do we still need sort by columns that contain zero reference? In this case, it will have no impact on the sorting results. IMO, we should not allow users do it. rxin cloud-fan marmbrus yhuai hvanhovell
-- Update: In these cases, they are ignored in this case.
**Note**: This PR is taken from https://github.com/apache/spark/pull/10731. When merging this PR, please give the credit to zhichao-li
Also cc all the people who are involved in the previous discussion: adrian-wang chenghao-intel tejasapatil
#### How was this patch tested?
Added a few test cases for both positive and negative test cases.
Author: gatorsmile <gatorsmile@gmail.com>
Closes#11815 from gatorsmile/orderByPosition.
## What changes were proposed in this pull request?
This PR adds some proper periods and spaces to Spark CLI help messages and SQL/YARN conf docs for consistency.
## How was this patch tested?
Manual.
Author: Dongjoon Hyun <dongjoon@apache.org>
Closes#11848 from dongjoon-hyun/add_proper_period_and_space.
## What changes were proposed in this pull request?
[Spark Coding Style Guide](https://cwiki.apache.org/confluence/display/SPARK/Spark+Code+Style+Guide) has 100-character limit on lines, but it's disabled for Java since 11/09/15. This PR enables **LineLength** checkstyle again. To help that, this also introduces **RedundantImport** and **RedundantModifier**, too. The following is the diff on `checkstyle.xml`.
```xml
- <!-- TODO: 11/09/15 disabled - the lengths are currently > 100 in many places -->
- <!--
<module name="LineLength">
<property name="max" value="100"/>
<property name="ignorePattern" value="^package.*|^import.*|a href|href|http://|https://|ftp://"/>
</module>
- -->
<module name="NoLineWrap"/>
<module name="EmptyBlock">
<property name="option" value="TEXT"/>
-167,5 +164,7
</module>
<module name="CommentsIndentation"/>
<module name="UnusedImports"/>
+ <module name="RedundantImport"/>
+ <module name="RedundantModifier"/>
```
## How was this patch tested?
Currently, `lint-java` is disabled in Jenkins. It needs a manual test.
After passing the Jenkins tests, `dev/lint-java` should passes locally.
Author: Dongjoon Hyun <dongjoon@apache.org>
Closes#11831 from dongjoon-hyun/SPARK-14011.
## What changes were proposed in this pull request?
Currently, there is no way to control the behaviour when fails to parse corrupt records in JSON data source .
This PR adds the support for parse modes just like CSV data source. There are three modes below:
- `PERMISSIVE` : When it fails to parse, this sets `null` to to field. This is a default mode when it has been this mode.
- `DROPMALFORMED`: When it fails to parse, this drops the whole record.
- `FAILFAST`: When it fails to parse, it just throws an exception.
This PR also make JSON data source share the `ParseModes` in CSV data source.
## How was this patch tested?
Unit tests were used and `./dev/run_tests` for code style tests.
Author: hyukjinkwon <gurwls223@gmail.com>
Closes#11756 from HyukjinKwon/SPARK-13764.
## What changes were proposed in this pull request?
Previously, Dataset.groupBy returns a GroupedData, and Dataset.groupByKey returns a GroupedDataset. The naming is very similar, and unfortunately does not convey the real differences between the two.
Assume we are grouping by some keys (K). groupByKey is a key-value style group by, in which the schema of the returned dataset is a tuple of just two fields: key and value. groupBy, on the other hand, is a relational style group by, in which the schema of the returned dataset is flattened and contain |K| + |V| fields.
This pull request also removes the experimental tag from RelationalGroupedDataset. It has been with DataFrame since 1.3, and we have enough confidence now to stabilize it.
## How was this patch tested?
This is a rename to improve API understandability. Should be covered by all existing tests.
Author: Reynold Xin <rxin@databricks.com>
Closes#11841 from rxin/SPARK-13897.
## What changes were proposed in this pull request?
500L << 20 is actually pretty close to 32-bit int limit. I was trying to increase this to 500L << 23 and got negative numbers instead.
## How was this patch tested?
I'm only modifying test code.
Author: Reynold Xin <rxin@databricks.com>
Closes#11839 from rxin/SPARK-14018.
## What changes were proposed in this pull request?
This is a minor followup on https://github.com/apache/spark/pull/11799 that extracts out the `VectorizedColumnReader` from `VectorizedParquetRecordReader` into its own file.
## How was this patch tested?
N/A (refactoring only)
Author: Sameer Agarwal <sameer@databricks.com>
Closes#11834 from sameeragarwal/rename.
## What changes were proposed in this pull request?
This PR cleans up the new parquet record reader with the following changes:
1. Removes the non-vectorized parquet reader code from `UnsafeRowParquetRecordReader`.
2. Removes the non-vectorized column reader code from `ColumnReader`.
3. Renames `UnsafeRowParquetRecordReader` to `VectorizedParquetRecordReader` and `ColumnReader` to `VectorizedColumnReader`
4. Deprecate `PARQUET_UNSAFE_ROW_RECORD_READER_ENABLED`
## How was this patch tested?
Refactoring only; Existing tests should reveal any problems.
Author: Sameer Agarwal <sameer@databricks.com>
Closes#11799 from sameeragarwal/vectorized-parquet.
## What changes were proposed in this pull request?
ShuffledHashJoin (also outer join) is removed in 1.6, in favor of SortMergeJoin, which is more robust and also fast.
ShuffledHashJoin is still useful in this case: 1) one table is much smaller than the other one, then cost to build a hash table on smaller table is smaller than sorting the larger table 2) any partition of the small table could fit in memory.
This PR brings back ShuffledHashJoin, basically revert #9645, and fix the conflict. Also merging outer join and left-semi join into the same class. This PR does not implement full outer join, because it's not implemented efficiently (requiring build hash table on both side).
A simple benchmark (one table is 5x smaller than other one) show that ShuffledHashJoin could be 2X faster than SortMergeJoin.
## How was this patch tested?
Added new unit tests for ShuffledHashJoin.
Author: Davies Liu <davies@databricks.com>
Closes#11788 from davies/shuffle_join.
## What changes were proposed in this pull request?
This patch updates documentations for Datasets. I also updated some internal documentation for exchange/broadcast.
## How was this patch tested?
Just documentation/api stability update.
Author: Reynold Xin <rxin@databricks.com>
Closes#11814 from rxin/dataset-docs.
## What changes were proposed in this pull request?
JIRA: https://issues.apache.org/jira/browse/SPARK-13930
Recently the fast serialization has been introduced to collecting DataFrame/Dataset (#11664). The same technology can be used on collect limit operator too.
## How was this patch tested?
Add a benchmark for collect limit to `BenchmarkWholeStageCodegen`.
Without this patch:
model name : Westmere E56xx/L56xx/X56xx (Nehalem-C)
collect limit: Best/Avg Time(ms) Rate(M/s) Per Row(ns) Relative
-------------------------------------------------------------------------------------------
collect limit 1 million 3413 / 3768 0.3 3255.0 1.0X
collect limit 2 millions 9728 / 10440 0.1 9277.3 0.4X
With this patch:
model name : Westmere E56xx/L56xx/X56xx (Nehalem-C)
collect limit: Best/Avg Time(ms) Rate(M/s) Per Row(ns) Relative
-------------------------------------------------------------------------------------------
collect limit 1 million 833 / 1284 1.3 794.4 1.0X
collect limit 2 millions 3348 / 4005 0.3 3193.3 0.2X
Author: Liang-Chi Hsieh <simonh@tw.ibm.com>
Closes#11759 from viirya/execute-take.
## What changes were proposed in this pull request?
This PR revises Dataset API ScalaDoc. All public methods are divided into the following groups
* `groupname basic`: Basic Dataset functions
* `groupname action`: Actions
* `groupname untypedrel`: Untyped Language Integrated Relational Queries
* `groupname typedrel`: Typed Language Integrated Relational Queries
* `groupname func`: Functional Transformations
* `groupname rdd`: RDD Operations
* `groupname output`: Output Operations
`since` tag and sample code are also updated. We may want to add more sample code for typed APIs.
## How was this patch tested?
Documentation change. Checked by building unidoc locally.
Author: Cheng Lian <lian@databricks.com>
Closes#11769 from liancheng/spark-13826-ds-api-doc.
## What changes were proposed in this pull request?
Support queries that JOIN tables with USING clause.
SELECT * from table1 JOIN table2 USING <column_list>
USING clause can be used as a means to simplify the join condition
when :
1) Equijoin semantics is desired and
2) The column names in the equijoin have the same name.
We already have the support for Natural Join in Spark. This PR makes
use of the already existing infrastructure for natural join to
form the join condition and also the projection list.
## How was the this patch tested?
Have added unit tests in SQLQuerySuite, CatalystQlSuite, ResolveNaturalJoinSuite
Author: Dilip Biswal <dbiswal@us.ibm.com>
Closes#11297 from dilipbiswal/spark-13427.
## What changes were proposed in this pull request?
Logging was made private in Spark 2.0. If we move it, then users would be able to create a Logging trait themselves to avoid changing their own code.
## How was this patch tested?
existing tests.
Author: Wenchen Fan <wenchen@databricks.com>
Closes#11764 from cloud-fan/logger.
Because ClassTags are available when constructing ShuffledRDD we can use them to automatically use Kryo for shuffle serialization when the RDD's types are known to be compatible with Kryo.
This patch introduces `SerializerManager`, a component which picks the "best" serializer for a shuffle given the elements' ClassTags. It will automatically pick a Kryo serializer for ShuffledRDDs whose key, value, and/or combiner types are primitives, arrays of primitives, or strings. In the future we can use this class as a narrow extension point to integrate specialized serializers for other types, such as ByteBuffers.
In a planned followup patch, I will extend the BlockManager APIs so that we're able to use similar automatic serializer selection when caching RDDs (this is a little trickier because the ClassTags need to be threaded through many more places).
Author: Josh Rosen <joshrosen@databricks.com>
Closes#11755 from JoshRosen/automatically-pick-best-serializer.
## What changes were proposed in this pull request?
Since developer API of plug-able parser has been removed in #10801 , docs should be updated accordingly.
## How was this patch tested?
This patch will not affect the real code path.
Author: Daoyuan Wang <daoyuan.wang@intel.com>
Closes#11758 from adrian-wang/spark12855.
## What changes were proposed in this pull request?
This PR removes three minor duplicated lines. First one is making the following unreachable code warning.
```
JoinSuite.scala:52: unreachable code
[warn] case j: BroadcastHashJoin => j
```
The other two are just consecutive repetitions in `Seq` of MiMa filters.
## How was this patch tested?
Pass the existing Jenkins test.
Author: Dongjoon Hyun <dongjoon@apache.org>
Closes#11773 from dongjoon-hyun/remove_duplicated_line.
## What changes were proposed in this pull request?
Fix expression generation for optional types.
Standard Java reflection causes issues when dealing with synthetic Scala objects (things that do not map to Java and thus contain a dollar sign in their name). This patch introduces Scala reflection in such cases.
This patch also adds a regression test for Dataset's handling of classes defined in package objects (which was the initial purpose of this PR).
## How was this patch tested?
A new test in ExpressionEncoderSuite that tests optional inner classes and a regression test for Dataset's handling of package objects.
Author: Jakob Odersky <jakob@odersky.com>
Closes#11708 from jodersky/SPARK-13118-package-objects.
## What changes were proposed in this pull request?
We need to copy the UnsafeRow since a Join could produce multiple rows from single input rows. We could avoid that if there is no join (or the join will not produce multiple rows) inside WholeStageCodegen.
Updated the benchmark for `collect`, we could see 20-30% speedup.
## How was this patch tested?
existing unit tests.
Author: Davies Liu <davies@databricks.com>
Closes#11740 from davies/avoid_copy2.
## What changes were proposed in this pull request?
This https://github.com/apache/spark/pull/2400 added the support to parse JSON rows wrapped with an array. However, this throws an exception when the given data contains array data and struct data in the same field as below:
```json
{"a": {"b": 1}}
{"a": []}
```
and the schema is given as below:
```scala
val schema =
StructType(
StructField("a", StructType(
StructField("b", StringType) :: Nil
)) :: Nil)
```
- **Before**
```scala
sqlContext.read.schema(schema).json(path).show()
```
```scala
Exception in thread "main" org.apache.spark.SparkException: Job aborted due to stage failure: Task 7 in stage 0.0 failed 4 times, most recent failure: Lost task 7.3 in stage 0.0 (TID 10, 192.168.1.170): java.lang.ClassCastException: org.apache.spark.sql.types.GenericArrayData cannot be cast to org.apache.spark.sql.catalyst.InternalRow
at org.apache.spark.sql.catalyst.expressions.BaseGenericInternalRow$class.getStruct(rows.scala:50)
at org.apache.spark.sql.catalyst.expressions.GenericMutableRow.getStruct(rows.scala:247)
at org.apache.spark.sql.catalyst.expressions.GeneratedClass$SpecificPredicate.eval(Unknown Source)
...
```
- **After**
```scala
sqlContext.read.schema(schema).json(path).show()
```
```bash
+----+
| a|
+----+
| [1]|
|null|
+----+
```
For other data types, in this case it converts the given values are `null` but only this case emits an exception.
This PR makes the support for wrapped rows applied only at the top level.
## How was this patch tested?
Unit tests were used and `./dev/run_tests` for code style tests.
Author: hyukjinkwon <gurwls223@gmail.com>
Closes#11752 from HyukjinKwon/SPARK-3308-follow-up.
## What changes were proposed in this pull request?
Narrow down the parameter type of `UserDefinedType#serialize()`. Currently, the parameter type is `Any`, however it would logically make more sense to narrow it down to the type of the actual user defined type.
## How was this patch tested?
Existing tests were successfully run on local machine.
Author: Jakob Odersky <jakob@odersky.com>
Closes#11379 from jodersky/SPARK-11011-udt-types.
## What changes were proposed in this pull request?
https://issues.apache.org/jira/browse/SPARK-13894
Change the return type of the `SQLContext.range` API from `DataFrame` to `Dataset`.
## How was this patch tested?
No additional unit test required.
Author: Cheng Hao <hao.cheng@intel.com>
Closes#11730 from chenghao-intel/range.
## What changes were proposed in this pull request?
There is a feature of hive SQL called multi-insert. For example:
```
FROM src
INSERT OVERWRITE TABLE dest1
SELECT key + 1
INSERT OVERWRITE TABLE dest2
SELECT key WHERE key > 2
INSERT OVERWRITE TABLE dest3
SELECT col EXPLODE(arr) exp AS col
...
```
We partially support it currently, with some limitations: 1) WHERE can't reference columns produced by LATERAL VIEW. 2) It's not executed eagerly, i.e. `sql("...multi-insert clause...")` won't take place right away like other commands, e.g. CREATE TABLE.
This PR removes these limitations and make us fully support multi-insert.
## How was this patch tested?
new tests in `SQLQuerySuite`
Author: Wenchen Fan <wenchen@databricks.com>
Closes#11754 from cloud-fan/lateral-view.
## What changes were proposed in this pull request?
Follow up to https://github.com/apache/spark/pull/11657
- Also update `String.getBytes("UTF-8")` to use `StandardCharsets.UTF_8`
- And fix one last new Coverity warning that turned up (use of unguarded `wait()` replaced by simpler/more robust `java.util.concurrent` classes in tests)
- And while we're here cleaning up Coverity warnings, just fix about 15 more build warnings
## How was this patch tested?
Jenkins tests
Author: Sean Owen <sowen@cloudera.com>
Closes#11725 from srowen/SPARK-13823.2.
## What changes were proposed in this pull request?
https://issues.apache.org/jira/browse/SPARK-13899
This PR makes CSV data source produce `InternalRow` instead of `Row`.
Basically, this resembles JSON data source. It uses the same codes for casting.
## How was this patch tested?
Unit tests were used within IDE and code style was checked by `./dev/run_tests`.
Author: hyukjinkwon <gurwls223@gmail.com>
Closes#11717 from HyukjinKwon/SPARK-13899.
## What changes were proposed in this pull request?
This PR brings codegen support for broadcast left-semi join.
## How was this patch tested?
Existing tests. Added benchmark, the result show 7X speedup.
Author: Davies Liu <davies@databricks.com>
Closes#11742 from davies/gen_semi.
## What changes were proposed in this pull request?
This PR just move some code from SortMergeOuterJoin into SortMergeJoin.
This is for support codegen for outer join.
## How was this patch tested?
existing tests.
Author: Davies Liu <davies@databricks.com>
Closes#11743 from davies/gen_smjouter.
## What changes were proposed in this pull request?
This patch changes DataFrameReader.text()'s return type from DataFrame to Dataset[String].
Closes#11731.
## How was this patch tested?
Updated existing integration tests to reflect the change.
Author: Reynold Xin <rxin@databricks.com>
Closes#11739 from rxin/SPARK-13895.
## What changes were proposed in this pull request?
Change the return type of toJson in Dataset class
## How was this patch tested?
No additional unit test required.
Author: Stavros Kontopoulos <stavros.kontopoulos@typesafe.com>
Closes#11732 from skonto/fix_toJson.
## What changes were proposed in this pull request?
Our internal code can go through SessionState.catalog and SessionState.analyzer. This brings two small benefits:
1. Reduces internal dependency on SQLContext.
2. Removes 2 public methods in Java (Java does not obey package private visibility).
More importantly, according to the design in SPARK-13485, we'd need to claim this catalog function for the user-facing public functions, rather than having an internal field.
## How was this patch tested?
Existing unit/integration test code.
Author: Reynold Xin <rxin@databricks.com>
Closes#11716 from rxin/SPARK-13893.
## What changes were proposed in this pull request?
Use method 'testQuietly' to avoid ContinuousQuerySuite flooding the console logs with garbage
Make ContinuousQuerySuite not output logs to the console. The logs will still output to unit-tests.log.
## How was this patch tested?
Just check Jenkins output.
Author: Xin Ren <iamshrek@126.com>
Closes#11703 from keypointt/SPARK-13660.
## What changes were proposed in this pull request?
In general it is better for internal classes to not depend on the external class (in this case SQLContext) to reduce coupling between user-facing APIs and the internal implementations. This patch removes SQLContext dependency from some internal classes such as SparkPlanner, SparkOptimizer.
As part of this patch, I also removed the following internal methods from SQLContext:
```
protected[sql] def functionRegistry: FunctionRegistry
protected[sql] def optimizer: Optimizer
protected[sql] def sqlParser: ParserInterface
protected[sql] def planner: SparkPlanner
protected[sql] def continuousQueryManager
protected[sql] def prepareForExecution: RuleExecutor[SparkPlan]
```
## How was this patch tested?
Existing unit/integration tests.
Author: Reynold Xin <rxin@databricks.com>
Closes#11712 from rxin/sqlContext-planner.
## What changes were proposed in this pull request?
When initial creating `CVSSuite.scala` in SPARK-12833, there was a typo on `scalastyle:on`: `scalstyle:on`. So, it turns off ScalaStyle checking for the rest of the file mistakenly. So, it can not find a violation on the code of `SPARK-12668` added recently. This issue fixes the existing escaping correctly and adds a new escaping for `SPARK-12668` code like the following.
```scala
test("test aliases sep and encoding for delimiter and charset") {
+ // scalastyle:off
val cars = sqlContext
...
.load(testFile(carsFile8859))
+ // scalastyle:on
```
This will prevent future potential problems, too.
## How was this patch tested?
Pass the Jenkins test.
Author: Dongjoon Hyun <dongjoon@apache.org>
Closes#11700 from dongjoon-hyun/SPARK-13870.
## What changes were proposed in this pull request?
This patch removes DescribeCommand's dependency on LogicalPlan. After this patch, DescribeCommand simply accepts a TableIdentifier. It minimizes the dependency, and blocks my next patch (removes SQLContext dependency from SparkPlanner).
## How was this patch tested?
Should be covered by existing unit tests and Hive compatibility tests that run describe table.
Author: Reynold Xin <rxin@databricks.com>
Closes#11710 from rxin/SPARK-13884.
## What changes were proposed in this pull request?
When we call DataFrame/Dataset.collect(), Java serializer (or Kryo Serializer) will be used to serialize the UnsafeRows in executor, then deserialize them into UnsafeRows in driver. Java serializer (and Kyro serializer) are slow on millions rows, because they try to find out the same rows, but usually there is no same rows.
This PR will serialize the UnsafeRows as byte array by packing them together, then Java serializer (or Kyro serializer) serialize the bytes very fast (there are fewer blocks and byte array are not compared by content).
The UnsafeRow format is highly compressible, the serialized bytes are also compressed (configurable by spark.io.compression.codec).
## How was this patch tested?
Existing unit tests.
Add a benchmark for collect, before this patch:
```
Intel(R) Core(TM) i7-4558U CPU 2.80GHz
collect: Best/Avg Time(ms) Rate(M/s) Per Row(ns) Relative
-------------------------------------------------------------------------------------------
collect 1 million 3991 / 4311 0.3 3805.7 1.0X
collect 2 millions 10083 / 10637 0.1 9616.0 0.4X
collect 4 millions 29551 / 30072 0.0 28182.3 0.1X
```
```
Intel(R) Core(TM) i7-4558U CPU 2.80GHz
collect: Best/Avg Time(ms) Rate(M/s) Per Row(ns) Relative
-------------------------------------------------------------------------------------------
collect 1 million 775 / 1170 1.4 738.9 1.0X
collect 2 millions 1153 / 1758 0.9 1099.3 0.7X
collect 4 millions 4451 / 5124 0.2 4244.9 0.2X
```
We can see about 5-7X speedup.
Author: Davies Liu <davies@databricks.com>
Closes#11664 from davies/serialize_row.
## What changes were proposed in this pull request?
Avoid the copy in HashedRelation, since most of the HashedRelation are built with Array[Row], added the copy() for LeftSemiJoinHash. This could help to reduce the memory consumption for Broadcast join.
## How was this patch tested?
Existing tests.
Author: Davies Liu <davies@databricks.com>
Closes#11666 from davies/remove_copy.
## What changes were proposed in this pull request?
1. Rename DataFrame.scala Dataset.scala, since the class is now named Dataset.
2. Remove LegacyFunctions. It was introduced in Spark 1.6 for backward compatibility, and can be removed in Spark 2.0.
## How was this patch tested?
Should be covered by existing unit/integration tests.
Author: Reynold Xin <rxin@databricks.com>
Closes#11704 from rxin/SPARK-13880.
## What changes were proposed in this pull request?
- Add a MetadataLog interface for metadata reliably storage.
- Add HDFSMetadataLog as a MetadataLog implementation based on HDFS.
- Update FileStreamSource to use HDFSMetadataLog instead of managing metadata by itself.
## How was this patch tested?
unit tests
Author: Shixiong Zhu <shixiong@databricks.com>
Closes#11625 from zsxwing/metadata-log.
## What changes were proposed in this pull request?
We introduced some local operators in org.apache.spark.sql.execution.local package but never fully wired the engine to actually use these. We still plan to implement a full local mode, but it's probably going to be fairly different from what the current iterator-based local mode would look like. Based on what we know right now, we might want a push-based columnar version of these operators.
Let's just remove them for now, and we can always re-introduced them in the future by looking at branch-1.6.
## How was this patch tested?
This is simply dead code removal.
Author: Reynold Xin <rxin@databricks.com>
Closes#11705 from rxin/SPARK-13882.
This PR adds a new strategy, `FileSourceStrategy`, that can be used for planning scans of collections of files that might be partitioned or bucketed.
Compared with the existing planning logic in `DataSourceStrategy` this version has the following desirable properties:
- It removes the need to have `RDD`, `broadcastedHadoopConf` and other distributed concerns in the public API of `org.apache.spark.sql.sources.FileFormat`
- Partition column appending is delegated to the format to avoid an extra copy / devectorization when appending partition columns
- It minimizes the amount of data that is shipped to each executor (i.e. it does not send the whole list of files to every worker in the form of a hadoop conf)
- it natively supports bucketing files into partitions, and thus does not require coalescing / creating a `UnionRDD` with the correct partitioning.
- Small files are automatically coalesced into fewer tasks using an approximate bin-packing algorithm.
Currently only a testing source is planned / tested using this strategy. In follow-up PRs we will port the existing formats to this API.
A stub for `FileScanRDD` is also added, but most methods remain unimplemented.
Other minor cleanups:
- partition pruning is pushed into `FileCatalog` so both the new and old code paths can use this logic. This will also allow future implementations to use indexes or other tricks (i.e. a MySQL metastore)
- The partitions from the `FileCatalog` now propagate information about file sizes all the way up to the planner so we can intelligently spread files out.
- `Array` -> `Seq` in some internal APIs to avoid unnecessary `toArray` calls
- Rename `Partition` to `PartitionDirectory` to differentiate partitions used earlier in pruning from those where we have already enumerated the files and their sizes.
Author: Michael Armbrust <michael@databricks.com>
Closes#11646 from marmbrus/fileStrategy.
Addressing outstanding comments in #11573.
Jenkins, new test case in `DDLCommandSuite`
Author: Andrew Or <andrew@databricks.com>
Closes#11667 from andrewor14/ddl-parser-followups.
If a _SUCCESS appears in the inner partitioning dir, partition discovery will treat that _SUCCESS file as a data file. Then, partition discovery will fail because it finds that the dir structure is not valid. We should ignore those `_SUCCESS` files.
In future, it is better to ignore all files/dirs starting with `_` or `.`. This PR does not make this change. I am thinking about making this change simple, so we can consider of getting it in branch 1.6.
To ignore all files/dirs starting with `_` or `, the main change is to let ParquetRelation have another way to get metadata files. Right now, it relies on FileStatusCache's cachedLeafStatuses, which returns file statuses of both metadata files (e.g. metadata files used by parquet) and data files, which requires more changes.
https://issues.apache.org/jira/browse/SPARK-13207
Author: Yin Huai <yhuai@databricks.com>
Closes#11088 from yhuai/SPARK-13207.
## What changes were proposed in this pull request?
This PR fixes 135 typos over 107 files:
* 121 typos in comments
* 11 typos in testcase name
* 3 typos in log messages
## How was this patch tested?
Manual.
Author: Dongjoon Hyun <dongjoon@apache.org>
Closes#11689 from dongjoon-hyun/fix_more_typos.
## What changes were proposed in this pull request?
- Fixes calls to `new String(byte[])` or `String.getBytes()` that rely on platform default encoding, to use UTF-8
- Same for `InputStreamReader` and `OutputStreamWriter` constructors
- Standardizes on UTF-8 everywhere
- Standardizes specifying the encoding with `StandardCharsets.UTF-8`, not the Guava constant or "UTF-8" (which means handling `UnuspportedEncodingException`)
- (also addresses the other remaining Coverity scan issues, which are pretty trivial; these are separated into commit 1deecd8d9c )
## How was this patch tested?
Jenkins tests
Author: Sean Owen <sowen@cloudera.com>
Closes#11657 from srowen/SPARK-13823.
## What changes were proposed in this pull request?
fix typo in DataSourceRegister
## How was this patch tested?
found when going through latest code
Author: Jacky Li <jacky.likun@huawei.com>
Closes#11686 from jackylk/patch-12.
## What changes were proposed in this pull request?
This PR removes two methods, `collectRows()` and `takeRows()`, from `Dataset[T]`. These methods were added in PR #11443, and were later considered not useful.
## How was this patch tested?
Existing tests should do the work.
Author: Cheng Lian <lian@databricks.com>
Closes#11678 from liancheng/remove-collect-rows-and-take-rows.
PR #11443 added an extra `plan: Option[LogicalPlan]` argument to `AnalysisException` and attached partially analyzed plan to thrown `AnalysisException` in `QueryExecution.assertAnalyzed()`. However, the original stack trace wasn't properly inherited. This PR fixes this issue by inheriting the stack trace.
A test case is added to verify that the first entry of `AnalysisException` stack trace isn't from `QueryExecution`.
Author: Cheng Lian <lian@databricks.com>
Closes#11677 from liancheng/analysis-exception-stacktrace.
## What changes were proposed in this pull request?
This PR split the PhysicalRDD into two classes, PhysicalRDD and PhysicalScan. PhysicalRDD is used for DataFrames that is created from existing RDD. PhysicalScan is used for DataFrame that is created from data sources. This enable use to apply different optimization on both of them.
Also fix the problem for sameResult() on two DataSourceScan.
Also fix the equality check to toString for `In`. It's better to use Seq there, but we can't break this public API (sad).
## How was this patch tested?
Existing tests. Manually tested with TPCDS query Q59 and Q64, all those duplicated exchanges can be re-used now, also saw there are 40+% performance improvement (saving half of the scan).
Author: Davies Liu <davies@databricks.com>
Closes#11514 from davies/existing_rdd.
## What changes were proposed in this pull request?
This patch is ported over from viirya's changes in #11048. Currently for most DDLs we just pass the query text directly to Hive. Instead, we should parse these commands ourselves and in the future (not part of this patch) use the `HiveCatalog` to process these DDLs. This is a pretext to merging `SQLContext` and `HiveContext`.
Note: As of this patch we still pass the query text to Hive. The difference is that we now parse the commands ourselves so in the future we can just use our own catalog.
## How was this patch tested?
Jenkins, new `DDLCommandSuite`, which comprises of about 40% of the changes here.
Author: Andrew Or <andrew@databricks.com>
Closes#11573 from andrewor14/parser-plus-plus.
This is needed to avoid odd compiler errors when building just the
sql package with maven, because of odd interactions between scalac
and shaded classes.
Author: Marcelo Vanzin <vanzin@cloudera.com>
Closes#11640 from vanzin/SPARK-13780.
## What changes were proposed in this pull request?
PR #11443 temporarily disabled MiMA check, this PR re-enables it.
One extra change is that `object DataFrame` is also removed. The only purpose of introducing `object DataFrame` was to use it as an internal factory for creating `Dataset[Row]`. By replacing this internal factory with `Dataset.newDataFrame`, both `DataFrame` and `DataFrame$` are entirely removed from the API, so that we can simply put a `MissingClassProblem` filter in `MimaExcludes.scala` for most DataFrame API changes.
## How was this patch tested?
Tested by MiMA check triggered by Jenkins.
Author: Cheng Lian <lian@databricks.com>
Closes#11656 from liancheng/re-enable-mima.
#### What changes were proposed in this pull request?
`projectList` is useless. Its value is always the same as the child.output. Remove it from the class `Window`. Removal can simplify the codes in Analyzer and Optimizer.
This PR is based on the discussion started by cloud-fan in a separate PR:
https://github.com/apache/spark/pull/5604#discussion_r55140466
This PR also eliminates useless `Window`.
cloud-fan yhuai
#### How was this patch tested?
Existing test cases cover it.
Author: gatorsmile <gatorsmile@gmail.com>
Author: xiaoli <lixiao1983@gmail.com>
Author: Xiao Li <xiaoli@Xiaos-MacBook-Pro.local>
Closes#11565 from gatorsmile/removeProjListWindow.
## What changes were proposed in this pull request?
This PR unifies DataFrame and Dataset by migrating existing DataFrame operations to Dataset and make `DataFrame` a type alias of `Dataset[Row]`.
Most Scala code changes are source compatible, but Java API is broken as Java knows nothing about Scala type alias (mostly replacing `DataFrame` with `Dataset<Row>`).
There are several noticeable API changes related to those returning arrays:
1. `collect`/`take`
- Old APIs in class `DataFrame`:
```scala
def collect(): Array[Row]
def take(n: Int): Array[Row]
```
- New APIs in class `Dataset[T]`:
```scala
def collect(): Array[T]
def take(n: Int): Array[T]
def collectRows(): Array[Row]
def takeRows(n: Int): Array[Row]
```
Two specialized methods `collectRows` and `takeRows` are added because Java doesn't support returning generic arrays. Thus, for example, `DataFrame.collect(): Array[T]` actually returns `Object` instead of `Array<T>` from Java side.
Normally, Java users may fall back to `collectAsList` and `takeAsList`. The two new specialized versions are added to avoid performance regression in ML related code (but maybe I'm wrong and they are not necessary here).
1. `randomSplit`
- Old APIs in class `DataFrame`:
```scala
def randomSplit(weights: Array[Double], seed: Long): Array[DataFrame]
def randomSplit(weights: Array[Double]): Array[DataFrame]
```
- New APIs in class `Dataset[T]`:
```scala
def randomSplit(weights: Array[Double], seed: Long): Array[Dataset[T]]
def randomSplit(weights: Array[Double]): Array[Dataset[T]]
```
Similar problem as above, but hasn't been addressed for Java API yet. We can probably add `randomSplitAsList` to fix this one.
1. `groupBy`
Some original `DataFrame.groupBy` methods have conflicting signature with original `Dataset.groupBy` methods. To distinguish these two, typed `Dataset.groupBy` methods are renamed to `groupByKey`.
Other noticeable changes:
1. Dataset always do eager analysis now
We used to support disabling DataFrame eager analysis to help reporting partially analyzed malformed logical plan on analysis failure. However, Dataset encoders requires eager analysi during Dataset construction. To preserve the error reporting feature, `AnalysisException` now takes an extra `Option[LogicalPlan]` argument to hold the partially analyzed plan, so that we can check the plan tree when reporting test failures. This plan is passed by `QueryExecution.assertAnalyzed`.
## How was this patch tested?
Existing tests do the work.
## TODO
- [ ] Fix all tests
- [ ] Re-enable MiMA check
- [ ] Update ScalaDoc (`since`, `group`, and example code)
Author: Cheng Lian <lian@databricks.com>
Author: Yin Huai <yhuai@databricks.com>
Author: Wenchen Fan <wenchen@databricks.com>
Author: Cheng Lian <liancheng@users.noreply.github.com>
Closes#11443 from liancheng/ds-to-df.
## What changes were proposed in this pull request?
Since the opening curly brace, '{', has many usages as discussed in [SPARK-3854](https://issues.apache.org/jira/browse/SPARK-3854), this PR adds a ScalaStyle rule to prevent '){' pattern for the following majority pattern and fixes the code accordingly. If we enforce this in ScalaStyle from now, it will improve the Scala code quality and reduce review time.
```
// Correct:
if (true) {
println("Wow!")
}
// Incorrect:
if (true){
println("Wow!")
}
```
IntelliJ also shows new warnings based on this.
## How was this patch tested?
Pass the Jenkins ScalaStyle test.
Author: Dongjoon Hyun <dongjoon@apache.org>
Closes#11637 from dongjoon-hyun/SPARK-3854.
## What changes were proposed in this pull request?
ContinuousQueryManager is sometimes flaky on Jenkins. I could not reproduce it on my machine, so I guess it about the waiting times which causes problems if Jenkins is loaded. I have increased the wait time in the hope that it will be less flaky.
## How was this patch tested?
I reran the unit test many times on a loop in my machine. I am going to run it a few time in Jenkins, that's the real test.
Author: Tathagata Das <tathagata.das1565@gmail.com>
Closes#11638 from tdas/cqm-flaky-test.
## What changes were proposed in this pull request?
We should reuse an object similar to the other non-primitive type getters. For
a query that computes averages over decimal columns, this shows a 10% speedup
on overall query times.
## How was this patch tested?
Existing tests and this benchmark
```
TPCDS Snappy: Best/Avg Time(ms) Rate(M/s) Per Row(ns)
--------------------------------------------------------------------------------
q27-agg (master) 10627 / 11057 10.8 92.3
q27-agg (this patch) 9722 / 9832 11.8 84.4
```
Author: Nong Li <nong@databricks.com>
Closes#11624 from nongli/spark-13790.
JIRA: https://issues.apache.org/jira/browse/SPARK-13636
## What changes were proposed in this pull request?
As shown in the wholestage codegen verion of Sort operator, when Sort is top of Exchange (or other operator that produce UnsafeRow), we will create variables from UnsafeRow, than create another UnsafeRow using these variables. We should avoid the unnecessary unpack and pack variables from UnsafeRows.
## How was this patch tested?
All existing wholestage codegen tests should be passed.
Author: Liang-Chi Hsieh <viirya@gmail.com>
Closes#11484 from viirya/direct-consume-unsaferow.
## What changes were proposed in this pull request?
According to #11627 , this PR replace `DataFrameWriter.stream()` with `startStream()` in comments of `ContinuousQueryListener.java`.
## How was this patch tested?
Manual. (It changes on comments.)
Author: Dongjoon Hyun <dongjoon@apache.org>
Closes#11629 from dongjoon-hyun/minor_rename.
## What changes were proposed in this pull request?
The new name makes it more obvious with the verb "start" that we are actually starting some execution.
## How was this patch tested?
This is just a rename. Existing unit tests should cover it.
Author: Reynold Xin <rxin@databricks.com>
Closes#11627 from rxin/SPARK-13794.
## What changes were proposed in this pull request?
https://issues.apache.org/jira/browse/SPARK-13766
This PR makes the file extensions (written by internal datasource) consistent.
**Before**
- TEXT, CSV and JSON
```
[.COMPRESSION_CODEC_NAME]
```
- Parquet
```
[.COMPRESSION_CODEC_NAME].parquet
```
- ORC
```
.orc
```
**After**
- TEXT, CSV and JSON
```
.txt[.COMPRESSION_CODEC_NAME]
.csv[.COMPRESSION_CODEC_NAME]
.json[.COMPRESSION_CODEC_NAME]
```
- Parquet
```
[.COMPRESSION_CODEC_NAME].parquet
```
- ORC
```
[.COMPRESSION_CODEC_NAME].orc
```
When the compression codec is set,
- For Parquet and ORC, each still stays in Parquet and ORC format but just have compressed data internally. So, I think it is okay to name `.parquet` and `.orc` at the end.
- For Text, CSV and JSON, each does not stays in each format but it has different data format according to compression codec. So, each has the names `.json`, `.csv` and `.txt` before the compression extension.
## How was this patch tested?
Unit tests are used and `./dev/run_tests` for coding style tests.
Author: hyukjinkwon <gurwls223@gmail.com>
Closes#11604 from HyukjinKwon/SPARK-13766.
## What changes were proposed in this pull request?
Fix this use case, which was already fixed in SPARK-10548 in 1.6 but was broken in master due to #9264:
```
(1 to 100).par.foreach { _ => sc.parallelize(1 to 5).map { i => (i, i) }.toDF("a", "b").count() }
```
This threw `IllegalArgumentException` consistently before this patch. For more detail, see the JIRA.
## How was this patch tested?
New test in `SQLExecutionSuite`.
Author: Andrew Or <andrew@databricks.com>
Closes#11586 from andrewor14/fix-concurrent-sql.
#### What changes were proposed in this pull request?
Remove all the deterministic conditions in a [[Filter]] that are contained in the Child's Constraints.
For example, the first query can be simplified to the second one.
```scala
val queryWithUselessFilter = tr1
.where("tr1.a".attr > 10 || "tr1.c".attr < 10)
.join(tr2.where('d.attr < 100), Inner, Some("tr1.a".attr === "tr2.a".attr))
.where(
("tr1.a".attr > 10 || "tr1.c".attr < 10) &&
'd.attr < 100 &&
"tr2.a".attr === "tr1.a".attr)
```
```scala
val query = tr1
.where("tr1.a".attr > 10 || "tr1.c".attr < 10)
.join(tr2.where('d.attr < 100), Inner, Some("tr1.a".attr === "tr2.a".attr))
```
#### How was this patch tested?
Six test cases are added.
Author: gatorsmile <gatorsmile@gmail.com>
Closes#11406 from gatorsmile/FilterRemoval.
## What changes were proposed in this pull request?
It’s possible to have common parts in a query, for example, self join, it will be good to avoid the duplicated part to same CPUs and memory (Broadcast or cache).
Exchange will materialize the underlying RDD by shuffle or collect, it’s a great point to check duplicates and reuse them. Duplicated exchanges means they generate exactly the same result inside a query.
In order to find out the duplicated exchanges, we should be able to compare SparkPlan to check that they have same results or not. We already have that for LogicalPlan, so we should move that into QueryPlan to make it available for SparkPlan.
Once we can find the duplicated exchanges, we should replace all of them with same SparkPlan object (could be wrapped by ReusedExchage for explain), then the plan tree become a DAG. Since all the planner only work with tree, so this rule should be the last one for the entire planning.
After the rule, the plan will looks like:
```
WholeStageCodegen
: +- Project [id#0L]
: +- BroadcastHashJoin [id#0L], [id#2L], Inner, BuildRight, None
: :- Project [id#0L]
: : +- BroadcastHashJoin [id#0L], [id#1L], Inner, BuildRight, None
: : :- Range 0, 1, 4, 1024, [id#0L]
: : +- INPUT
: +- INPUT
:- BroadcastExchange HashedRelationBroadcastMode(true,List(id#1L),List(id#1L))
: +- WholeStageCodegen
: : +- Range 0, 1, 4, 1024, [id#1L]
+- ReusedExchange [id#2L], BroadcastExchange HashedRelationBroadcastMode(true,List(id#1L),List(id#1L))
```
For three ways SortMergeJoin,
```
== Physical Plan ==
WholeStageCodegen
: +- Project [id#0L]
: +- SortMergeJoin [id#0L], [id#4L], None
: :- INPUT
: +- INPUT
:- WholeStageCodegen
: : +- Project [id#0L]
: : +- SortMergeJoin [id#0L], [id#3L], None
: : :- INPUT
: : +- INPUT
: :- WholeStageCodegen
: : : +- Sort [id#0L ASC], false, 0
: : : +- INPUT
: : +- Exchange hashpartitioning(id#0L, 200), None
: : +- WholeStageCodegen
: : : +- Range 0, 1, 4, 33554432, [id#0L]
: +- WholeStageCodegen
: : +- Sort [id#3L ASC], false, 0
: : +- INPUT
: +- ReusedExchange [id#3L], Exchange hashpartitioning(id#0L, 200), None
+- WholeStageCodegen
: +- Sort [id#4L ASC], false, 0
: +- INPUT
+- ReusedExchange [id#4L], Exchange hashpartitioning(id#0L, 200), None
```
If the same ShuffleExchange or BroadcastExchange, execute()/executeBroadcast() will be called by different parents, they should cached the RDD/Broadcast, return the same one for all the parents.
## How was this patch tested?
Added some unit tests for this. Had done some manual tests on TPCDS query Q59 and Q64, we can see some exchanges are re-used (this requires a change in PhysicalRDD to for sameResult, is be done in #11514 ).
Author: Davies Liu <davies@databricks.com>
Closes#11403 from davies/dedup.
## What changes were proposed in this pull request?
If there are many branches in a CaseWhen expression, the generated code could go above the 64K limit for single java method, will fail to compile. This PR change it to fallback to interpret mode if there are more than 20 branches.
This PR is based on #11243 and #11221, thanks to joehalliwell
Closes#11243Closes#11221
## How was this patch tested?
Add a test with 50 branches.
Author: Davies Liu <davies@databricks.com>
Closes#11592 from davies/fix_when.
## What changes were proposed in this pull request?
In order to make `docs/examples` (and other related code) more simple/readable/user-friendly, this PR replaces existing codes like the followings by using `diamond` operator.
```
- final ArrayList<Product2<Object, Object>> dataToWrite =
- new ArrayList<Product2<Object, Object>>();
+ final ArrayList<Product2<Object, Object>> dataToWrite = new ArrayList<>();
```
Java 7 or higher supports **diamond** operator which replaces the type arguments required to invoke the constructor of a generic class with an empty set of type parameters (<>). Currently, Spark Java code use mixed usage of this.
## How was this patch tested?
Manual.
Pass the existing tests.
Author: Dongjoon Hyun <dongjoon@apache.org>
Closes#11541 from dongjoon-hyun/SPARK-13702.
## What changes were proposed in this pull request?
This issue fixes the following potential bugs and Java coding style detected by Coverity and Checkstyle.
- Implement both null and type checking in equals functions.
- Fix wrong type casting logic in SimpleJavaBean2.equals.
- Add `implement Cloneable` to `UTF8String` and `SortedIterator`.
- Remove dereferencing before null check in `AbstractBytesToBytesMapSuite`.
- Fix coding style: Add '{}' to single `for` statement in mllib examples.
- Remove unused imports in `ColumnarBatch` and `JavaKinesisStreamSuite`.
- Remove unused fields in `ChunkFetchIntegrationSuite`.
- Add `stop()` to prevent resource leak.
Please note that the last two checkstyle errors exist on newly added commits after [SPARK-13583](https://issues.apache.org/jira/browse/SPARK-13583).
## How was this patch tested?
manual via `./dev/lint-java` and Coverity site.
Author: Dongjoon Hyun <dongjoon@apache.org>
Closes#11530 from dongjoon-hyun/SPARK-13692.
This PR replaces #9925 which had issues with CI. **Please see the original PR for any previous discussions.**
## What changes were proposed in this pull request?
Deprecate the SparkSQL column operator !== and use =!= as an alternative.
Fixes subtle issues related to operator precedence (basically, !== does not have the same priority as its logical negation, ===).
## How was this patch tested?
All currently existing tests.
Author: Jakob Odersky <jodersky@gmail.com>
Closes#11588 from jodersky/SPARK-7286.
## Motivation
CSV data source was contributed by Databricks. It is the inlined version of https://github.com/databricks/spark-csv. The data source name was `com.databricks.spark.csv`. As a result there are many tables created on older versions of spark with that name as the source. For backwards compatibility we should keep the old name.
## Proposed changes
`com.databricks.spark.csv` was added to list of `backwardCompatibilityMap` in `ResolvedDataSource.scala`
## Tests
A unit test was added to `CSVSuite` to parse a csv file using the old name.
Author: Hossein <hossein@databricks.com>
Closes#11589 from falaki/SPARK-13754.
## What changes were proposed in this pull request?
This PR fix the sizeInBytes of HadoopFsRelation.
## How was this patch tested?
Added regression test for that.
Author: Davies Liu <davies@databricks.com>
Closes#11590 from davies/fix_sizeInBytes.
When generating Graphviz DOT files in the SQL query visualization we need to escape double-quotes inside node labels. This is a followup to #11309, which fixed a similar graph in Spark Core's DAG visualization.
Author: Josh Rosen <joshrosen@databricks.com>
Closes#11587 from JoshRosen/graphviz-escaping.
## What changes were proposed in this pull request?
If a filter predicate or a join condition consists of `IsNotNull` checks, we should reorder these checks such that these non-nullability checks are evaluated before the rest of the predicates.
For e.g., if a filter predicate is of the form `a > 5 && isNotNull(b)`, we should rewrite this as `isNotNull(b) && a > 5` during physical plan generation.
## How was this patch tested?
new unit tests that verify the physical plan for both filters and joins in `ReorderedPredicateSuite`
Author: Sameer Agarwal <sameer@databricks.com>
Closes#11511 from sameeragarwal/reorder-isnotnull.
Follow-up to #11509, that simply refactors the interface that we use when resolving a pluggable `DataSource`.
- Multiple functions share the same set of arguments so we make this a case class, called `DataSource`. Actual resolution is now done by calling a function on this class.
- Instead of having multiple methods named `apply` (some of which do writing some of which do reading) we now explicitly have `resolveRelation()` and `write(mode, df)`.
- Get rid of `Array[String]` since this is an internal API and was forcing us to awkwardly call `toArray` in a bunch of places.
Author: Michael Armbrust <michael@databricks.com>
Closes#11572 from marmbrus/dataSourceResolution.
## What changes were proposed in this pull request?
This PR change the way how we generate the code for the output variables passing from a plan to it's parent.
Right now, they are generated before call consume() of it's parent. It's not efficient, if the parent is a Filter or Join, which could filter out most the rows, the time to access some of the columns that are not used by the Filter or Join are wasted.
This PR try to improve this by defering the access of columns until they are actually used by a plan. After this PR, a plan does not need to generate code to evaluate the variables for output, just passing the ExprCode to its parent by `consume()`. In `parent.consumeChild()`, it will check the output from child and `usedInputs`, generate the code for those columns that is part of `usedInputs` before calling `doConsume()`.
This PR also change the `if` from
```
if (cond) {
xxx
}
```
to
```
if (!cond) continue;
xxx
```
The new one could help to reduce the nested indents for multiple levels of Filter and BroadcastHashJoin.
It also added some comments for operators.
## How was the this patch tested?
Unit tests. Manually ran TPCDS Q55, this PR improve the performance about 30% (scale=10, from 2.56s to 1.96s)
Author: Davies Liu <davies@databricks.com>
Closes#11274 from davies/gen_defer.
## What changes were proposed in this pull request?
When we add more DDL parsing logic in the future, SparkQl will become very big. To keep it smaller, we'll introduce helper "parser objects", e.g. one to parse alter table commands. However, these parser objects will need to access some helper methods that exist in CatalystQl. The proposal is to move those methods to an isolated ParserUtils object.
This is based on viirya's changes in #11048. It prefaces the bigger fix for SPARK-13139 to make the diff of that patch smaller.
## How was this patch tested?
No change in functionality, so just Jenkins.
Author: Andrew Or <andrew@databricks.com>
Closes#11529 from andrewor14/parser-utils.
`HadoopFsRelation` is used for reading most files into Spark SQL. However today this class mixes the concerns of file management, schema reconciliation, scan building, bucketing, partitioning, and writing data. As a result, many data sources are forced to reimplement the same functionality and the various layers have accumulated a fair bit of inefficiency. This PR is a first cut at separating this into several components / interfaces that are each described below. Additionally, all implementations inside of Spark (parquet, csv, json, text, orc, svmlib) have been ported to the new API `FileFormat`. External libraries, such as spark-avro will also need to be ported to work with Spark 2.0.
### HadoopFsRelation
A simple `case class` that acts as a container for all of the metadata required to read from a datasource. All discovery, resolution and merging logic for schemas and partitions has been removed. This an internal representation that no longer needs to be exposed to developers.
```scala
case class HadoopFsRelation(
sqlContext: SQLContext,
location: FileCatalog,
partitionSchema: StructType,
dataSchema: StructType,
bucketSpec: Option[BucketSpec],
fileFormat: FileFormat,
options: Map[String, String]) extends BaseRelation
```
### FileFormat
The primary interface that will be implemented by each different format including external libraries. Implementors are responsible for reading a given format and converting it into `InternalRow` as well as writing out an `InternalRow`. A format can optionally return a schema that is inferred from a set of files.
```scala
trait FileFormat {
def inferSchema(
sqlContext: SQLContext,
options: Map[String, String],
files: Seq[FileStatus]): Option[StructType]
def prepareWrite(
sqlContext: SQLContext,
job: Job,
options: Map[String, String],
dataSchema: StructType): OutputWriterFactory
def buildInternalScan(
sqlContext: SQLContext,
dataSchema: StructType,
requiredColumns: Array[String],
filters: Array[Filter],
bucketSet: Option[BitSet],
inputFiles: Array[FileStatus],
broadcastedConf: Broadcast[SerializableConfiguration],
options: Map[String, String]): RDD[InternalRow]
}
```
The current interface is based on what was required to get all the tests passing again, but still mixes a couple of concerns (i.e. `bucketSet` is passed down to the scan instead of being resolved by the planner). Additionally, scans are still returning `RDD`s instead of iterators for single files. In a future PR, bucketing should be removed from this interface and the scan should be isolated to a single file.
### FileCatalog
This interface is used to list the files that make up a given relation, as well as handle directory based partitioning.
```scala
trait FileCatalog {
def paths: Seq[Path]
def partitionSpec(schema: Option[StructType]): PartitionSpec
def allFiles(): Seq[FileStatus]
def getStatus(path: Path): Array[FileStatus]
def refresh(): Unit
}
```
Currently there are two implementations:
- `HDFSFileCatalog` - based on code from the old `HadoopFsRelation`. Infers partitioning by recursive listing and caches this data for performance
- `HiveFileCatalog` - based on the above, but it uses the partition spec from the Hive Metastore.
### ResolvedDataSource
Produces a logical plan given the following description of a Data Source (which can come from DataFrameReader or a metastore):
- `paths: Seq[String] = Nil`
- `userSpecifiedSchema: Option[StructType] = None`
- `partitionColumns: Array[String] = Array.empty`
- `bucketSpec: Option[BucketSpec] = None`
- `provider: String`
- `options: Map[String, String]`
This class is responsible for deciding which of the Data Source APIs a given provider is using (including the non-file based ones). All reconciliation of partitions, buckets, schema from metastores or inference is done here.
### DataSourceAnalysis / DataSourceStrategy
Responsible for analyzing and planning reading/writing of data using any of the Data Source APIs, including:
- pruning the files from partitions that will be read based on filters.
- appending partition columns*
- applying additional filters when a data source can not evaluate them internally.
- constructing an RDD that is bucketed correctly when required*
- sanity checking schema match-up and other analysis when writing.
*In the future we should do that following:
- Break out file handling into its own Strategy as its sufficiently complex / isolated.
- Push the appending of partition columns down in to `FileFormat` to avoid an extra copy / unvectorization.
- Use a custom RDD for scans instead of `SQLNewNewHadoopRDD2`
Author: Michael Armbrust <michael@databricks.com>
Author: Wenchen Fan <wenchen@databricks.com>
Closes#11509 from marmbrus/fileDataSource.
## What changes were proposed in this pull request?
https://issues.apache.org/jira/browse/SPARK-13442
This PR adds the support for inferring `BooleanType` for schema.
It supports to infer case-insensitive `true` / `false` as `BooleanType`.
Unittests were added for `CSVInferSchemaSuite` and `CSVSuite` for end-to-end test.
## How was the this patch tested?
This was tested with unittests and with `dev/run_tests` for coding style
Author: hyukjinkwon <gurwls223@gmail.com>
Closes#11315 from HyukjinKwon/SPARK-13442.
## What changes were proposed in this pull request?
This PR adds an optimizer rule to eliminate reading (unnecessary) NULL values if they are not required for correctness by inserting `isNotNull` filters is the query plan. These filters are currently inserted beneath existing `Filter` and `Join` operators and are inferred based on their data constraints.
Note: While this optimization is applicable to all types of join, it primarily benefits `Inner` and `LeftSemi` joins.
## How was this patch tested?
1. Added a new `NullFilteringSuite` that tests for `IsNotNull` filters in the query plan for joins and filters. Also, tests interaction with the `CombineFilters` optimizer rules.
2. Test generated ExpressionTrees via `OrcFilterSuite`
3. Test filter source pushdown logic via `SimpleTextHadoopFsRelationSuite`
cc yhuai nongli
Author: Sameer Agarwal <sameer@databricks.com>
Closes#11372 from sameeragarwal/gen-isnotnull.
## What changes were proposed in this pull request?
It's weird that expressions don't always have all the expressions in it. This PR marks `QueryPlan.expressions` final to forbid sub classes overriding it to exclude some expressions. Currently only `Generate` override it, we can use `producedAttributes` to fix the unresolved attribute problem for it.
Note that this PR doesn't fix the problem in #11497
## How was this patch tested?
existing tests.
Author: Wenchen Fan <wenchen@databricks.com>
Closes#11532 from cloud-fan/generate.
## What changes were proposed in this pull request?
(Please fill in changes proposed in this fix)
Currently, the parquet reader returns rows one by one which is bad for performance. This patch
updates the reader to directly return ColumnarBatches. This is only enabled with whole stage
codegen, which is the only operator currently that is able to consume ColumnarBatches (instead
of rows). The current implementation is a bit of a hack to get this to work and we should do
more refactoring of these low level interfaces to make this work better.
## How was this patch tested?
```
Results:
TPCDS: Best/Avg Time(ms) Rate(M/s) Per Row(ns)
---------------------------------------------------------------------------------
q55 (before) 8897 / 9265 12.9 77.2
q55 5486 / 5753 21.0 47.6
```
Author: Nong Li <nong@databricks.com>
Closes#11435 from nongli/spark-13255.
## What changes were proposed in this pull request?
This patch simply moves things to existing package `o.a.s.sql.catalyst.parser` in an effort to reduce the size of the diff in #11048. This is conceptually the same as a recently merged patch #11482.
## How was this patch tested?
Jenkins.
Author: Andrew Or <andrew@databricks.com>
Closes#11506 from andrewor14/parser-package.
## What changes were proposed in this pull request?
A test suite added for the bug fix -SPARK 12941; for the mapping of the StringType to corresponding in Oracle
## How was this patch tested?
manual tests done
(Please explain how this patch was tested. E.g. unit tests, integration tests, manual tests)
(If this patch involves UI changes, please attach a screenshot; otherwise, remove this)
Author: thomastechs <thomas.sebastian@tcs.com>
Author: THOMAS SEBASTIAN <thomas.sebastian@tcs.com>
Closes#11489 from thomastechs/thomastechs-12941-master-new.
## What changes were proposed in this pull request?
This PR support visualization for subquery in SQL web UI, also improve the explain of subquery, especially when it's used together with whole stage codegen.
For example:
```python
>>> sqlContext.range(100).registerTempTable("range")
>>> sqlContext.sql("select id / (select sum(id) from range) from range where id > (select id from range limit 1)").explain(True)
== Parsed Logical Plan ==
'Project [unresolvedalias(('id / subquery#9), None)]
: +- 'SubqueryAlias subquery#9
: +- 'Project [unresolvedalias('sum('id), None)]
: +- 'UnresolvedRelation `range`, None
+- 'Filter ('id > subquery#8)
: +- 'SubqueryAlias subquery#8
: +- 'GlobalLimit 1
: +- 'LocalLimit 1
: +- 'Project [unresolvedalias('id, None)]
: +- 'UnresolvedRelation `range`, None
+- 'UnresolvedRelation `range`, None
== Analyzed Logical Plan ==
(id / scalarsubquery()): double
Project [(cast(id#0L as double) / cast(subquery#9 as double)) AS (id / scalarsubquery())#11]
: +- SubqueryAlias subquery#9
: +- Aggregate [(sum(id#0L),mode=Complete,isDistinct=false) AS sum(id)#10L]
: +- SubqueryAlias range
: +- Range 0, 100, 1, 4, [id#0L]
+- Filter (id#0L > subquery#8)
: +- SubqueryAlias subquery#8
: +- GlobalLimit 1
: +- LocalLimit 1
: +- Project [id#0L]
: +- SubqueryAlias range
: +- Range 0, 100, 1, 4, [id#0L]
+- SubqueryAlias range
+- Range 0, 100, 1, 4, [id#0L]
== Optimized Logical Plan ==
Project [(cast(id#0L as double) / cast(subquery#9 as double)) AS (id / scalarsubquery())#11]
: +- SubqueryAlias subquery#9
: +- Aggregate [(sum(id#0L),mode=Complete,isDistinct=false) AS sum(id)#10L]
: +- Range 0, 100, 1, 4, [id#0L]
+- Filter (id#0L > subquery#8)
: +- SubqueryAlias subquery#8
: +- GlobalLimit 1
: +- LocalLimit 1
: +- Project [id#0L]
: +- Range 0, 100, 1, 4, [id#0L]
+- Range 0, 100, 1, 4, [id#0L]
== Physical Plan ==
WholeStageCodegen
: +- Project [(cast(id#0L as double) / cast(subquery#9 as double)) AS (id / scalarsubquery())#11]
: : +- Subquery subquery#9
: : +- WholeStageCodegen
: : : +- TungstenAggregate(key=[], functions=[(sum(id#0L),mode=Final,isDistinct=false)], output=[sum(id)#10L])
: : : +- INPUT
: : +- Exchange SinglePartition, None
: : +- WholeStageCodegen
: : : +- TungstenAggregate(key=[], functions=[(sum(id#0L),mode=Partial,isDistinct=false)], output=[sum#14L])
: : : +- Range 0, 1, 4, 100, [id#0L]
: +- Filter (id#0L > subquery#8)
: : +- Subquery subquery#8
: : +- CollectLimit 1
: : +- WholeStageCodegen
: : : +- Project [id#0L]
: : : +- Range 0, 1, 4, 100, [id#0L]
: +- Range 0, 1, 4, 100, [id#0L]
```
The web UI looks like:
This PR also change the tree structure of WholeStageCodegen to make it consistent than others. Before this change, Both WholeStageCodegen and InputAdapter hold a references to the same plans, those could be updated without notify another, causing problems, this is discovered by #11403 .
## How was this patch tested?
Existing tests, also manual tests with the example query, check the explain and web UI.
Author: Davies Liu <davies@databricks.com>
Closes#11417 from davies/viz_subquery.
## What changes were proposed in this pull request?
Make ContinuousQueryManagerSuite not output logs to the console. The logs will still output to `unit-tests.log`.
I also updated `SQLListenerMemoryLeakSuite` to use `quietly` to avoid changing the log level which won't output logs to `unit-tests.log`.
## How was this patch tested?
Just check Jenkins output.
Author: Shixiong Zhu <shixiong@databricks.com>
Closes#11439 from zsxwing/quietly-ContinuousQueryManagerSuite.
## What changes were proposed in this pull request?
This patch simply moves things to a new package in an effort to reduce the size of the diff in #11048. Currently the new package only has one file, but in the future we'll add many new commands in SPARK-13139.
## How was this patch tested?
Jenkins.
Author: Andrew Or <andrew@databricks.com>
Closes#11482 from andrewor14/commands-package.
## What changes were proposed in this pull request?
This PR fixes typos in comments and testcase name of code.
## How was this patch tested?
manual.
Author: Dongjoon Hyun <dongjoon@apache.org>
Closes#11481 from dongjoon-hyun/minor_fix_typos_in_code.
## What changes were proposed in this pull request?
This PR adds the support to specify compression codecs for both ORC and Parquet.
## How was this patch tested?
unittests within IDE and code style tests with `dev/run_tests`.
Author: hyukjinkwon <gurwls223@gmail.com>
Closes#11464 from HyukjinKwon/SPARK-13543.
## What changes were proposed in this pull request?
Fixes compile problem due to inadvertent use of `Option.contains`, only in Scala 2.11. The change should have been to replace `Option.exists(_ == x)` with `== Some(x)`. Replacing exists with contains only makes sense for collections. Replacing use of `Option.exists` still makes sense though as it's misleading.
## How was this patch tested?
Jenkins tests / compilation
(If this patch involves UI changes, please attach a screenshot; otherwise, remove this)
Author: Sean Owen <sowen@cloudera.com>
Closes#11493 from srowen/SPARK-13423.2.
## What changes were proposed in this pull request?
After SPARK-6990, `dev/lint-java` keeps Java code healthy and helps PR review by saving much time.
This issue aims remove unused imports from Java/Scala code and add `UnusedImports` checkstyle rule to help developers.
## How was this patch tested?
```
./dev/lint-java
./build/sbt compile
```
Author: Dongjoon Hyun <dongjoon@apache.org>
Closes#11438 from dongjoon-hyun/SPARK-13583.
## What changes were proposed in this pull request?
Make some cross-cutting code improvements according to static analysis. These are individually up for discussion since they exist in separate commits that can be reverted. The changes are broadly:
- Inner class should be static
- Mismatched hashCode/equals
- Overflow in compareTo
- Unchecked warnings
- Misuse of assert, vs junit.assert
- get(a) + getOrElse(b) -> getOrElse(a,b)
- Array/String .size -> .length (occasionally, -> .isEmpty / .nonEmpty) to avoid implicit conversions
- Dead code
- tailrec
- exists(_ == ) -> contains find + nonEmpty -> exists filter + size -> count
- reduce(_+_) -> sum map + flatten -> map
The most controversial may be .size -> .length simply because of its size. It is intended to avoid implicits that might be expensive in some places.
## How was the this patch tested?
Existing Jenkins unit tests.
Author: Sean Owen <sowen@cloudera.com>
Closes#11292 from srowen/SPARK-13423.
JIRA: https://issues.apache.org/jira/browse/SPARK-13466
## What changes were proposed in this pull request?
With column pruning rule in optimizer, some Project operators will become redundant. We should remove these redundant Projects.
For an example query:
val input = LocalRelation('key.int, 'value.string)
val query =
Project(Seq($"x.key", $"y.key"),
Join(
SubqueryAlias("x", input),
BroadcastHint(SubqueryAlias("y", input)), Inner, None))
After the first run of column pruning, it would like:
Project(Seq($"x.key", $"y.key"),
Join(
Project(Seq($"x.key"), SubqueryAlias("x", input)),
Project(Seq($"y.key"), <-- inserted by the rule
BroadcastHint(SubqueryAlias("y", input))),
Inner, None))
Actually we don't need the outside Project now. This patch will remove it:
Join(
Project(Seq($"x.key"), SubqueryAlias("x", input)),
Project(Seq($"y.key"),
BroadcastHint(SubqueryAlias("y", input))),
Inner, None)
## How was the this patch tested?
Unit test is added into ColumnPruningSuite.
Author: Liang-Chi Hsieh <viirya@gmail.com>
Closes#11341 from viirya/remove-redundant-project.
## What changes were proposed in this pull request?
This pr to make the short names of compression codecs in `ParquetRelation` consistent against other ones. This pr comes from #11324.
## How was this patch tested?
Add more tests in `TextSuite`.
Author: Takeshi YAMAMURO <linguin.m.s@gmail.com>
Closes#11408 from maropu/SPARK-13528.
## What changes were proposed in this pull request?
Also updated the other benchmarks when the default to use vectorized decode was flipped.
Author: Nong Li <nong@databricks.com>
Closes#11454 from nongli/benchmark.
## What changes were proposed in this pull request?
In order to tell OutputStream that the task has failed or not, we should call the failure callbacks BEFORE calling writer.close().
## How was this patch tested?
Added new unit tests.
Author: Davies Liu <davies@databricks.com>
Closes#11450 from davies/callback.
#### What changes were proposed in this pull request?
This PR is to prune unnecessary columns when the operator is `MapPartitions`. The solution is to add an extra `Project` in the child node.
For the other two operators `AppendColumns` and `MapGroups`, it sounds doable. More discussions are required. The major reason is the current implementation of the `inputPlan` of `groupBy` is based on the child of `AppendColumns`. It might be a bug? Thus, will submit a separate PR.
#### How was this patch tested?
Added a test case in ColumnPruningSuite to verify the rule. Added another test case in DatasetSuite.scala to verify the data.
Author: gatorsmile <gatorsmile@gmail.com>
Closes#11460 from gatorsmile/datasetPruningNew.
Rows with null values in partition column are not included in the results because none of the partition
where clause specify is null predicate on the partition column. This fix adds is null predicate on the partition column to the first JDBC partition where clause.
Example:
JDBCPartition(THEID < 1 or THEID is null, 0),JDBCPartition(THEID >= 1 AND THEID < 2,1),
JDBCPartition(THEID >= 2, 2)
Author: sureshthalamati <suresh.thalamati@gmail.com>
Closes#11063 from sureshthalamati/nullable_jdbc_part_col_spark-13167.
## What changes were proposed in this pull request?
Broadcast left semi join without joining keys is already supported in BroadcastNestedLoopJoin, it has the same implementation as LeftSemiJoinBNL, we should remove that.
## How was this patch tested?
Updated unit tests.
Author: Davies Liu <davies@databricks.com>
Closes#11448 from davies/remove_bnl.
## What changes were proposed in this pull request?
This PR defer the resolution from a id of dictionary to value until the column is actually accessed (inside getInt/getLong), this is very useful for those columns and rows that are filtered out. It's also useful for binary type, we will not need to copy all the byte arrays.
This PR also change the underlying type for small decimal that could be fit within a Int, in order to use getInt() to lookup the value from IntDictionary.
## How was this patch tested?
Manually test TPCDS Q7 with scale factor 10, saw about 30% improvements (after PR #11274).
Author: Davies Liu <davies@databricks.com>
Closes#11437 from davies/decode_dict.
JIRA: https://issues.apache.org/jira/browse/SPARK-13511
## What changes were proposed in this pull request?
Current limit operator doesn't support wholestage codegen. This is open to add support for it.
In the `doConsume` of `GlobalLimit` and `LocalLimit`, we use a count term to count the processed rows. Once the row numbers catches the limit number, we set the variable `stopEarly` of `BufferedRowIterator` newly added in this pr to `true` that indicates we want to stop processing remaining rows. Then when the wholestage codegen framework checks `shouldStop()`, it will stop the processing of the row iterator.
Before this, the executed plan for a query `sqlContext.range(N).limit(100).groupBy().sum()` is:
TungstenAggregate(key=[], functions=[(sum(id#5L),mode=Final,isDistinct=false)], output=[sum(id)#6L])
+- TungstenAggregate(key=[], functions=[(sum(id#5L),mode=Partial,isDistinct=false)], output=[sum#9L])
+- GlobalLimit 100
+- Exchange SinglePartition, None
+- LocalLimit 100
+- Range 0, 1, 1, 524288000, [id#5L]
After add wholestage codegen support:
WholeStageCodegen
: +- TungstenAggregate(key=[], functions=[(sum(id#40L),mode=Final,isDistinct=false)], output=[sum(id)#41L])
: +- TungstenAggregate(key=[], functions=[(sum(id#40L),mode=Partial,isDistinct=false)], output=[sum#44L])
: +- GlobalLimit 100
: +- INPUT
+- Exchange SinglePartition, None
+- WholeStageCodegen
: +- LocalLimit 100
: +- Range 0, 1, 1, 524288000, [id#40L]
## How was this patch tested?
A test is added into BenchmarkWholeStageCodegen.
Author: Liang-Chi Hsieh <viirya@gmail.com>
Closes#11391 from viirya/wholestage-limit.
## What changes were proposed in this pull request?
This PR adds support for implementing whole state codegen for sort. Builds heaving on nongli 's PR: https://github.com/apache/spark/pull/11008 (which actually implements the feature), and adds the following changes on top:
- [x] Generated code updates peak execution memory metrics
- [x] Unit tests in `WholeStageCodegenSuite` and `SQLMetricsSuite`
## How was this patch tested?
New unit tests in `WholeStageCodegenSuite` and `SQLMetricsSuite`. Further, all existing sort tests should pass.
Author: Sameer Agarwal <sameer@databricks.com>
Author: Nong Li <nong@databricks.com>
Closes#11359 from sameeragarwal/sort-codegen.
https://issues.apache.org/jira/browse/SPARK-13507https://issues.apache.org/jira/browse/SPARK-13509
## What changes were proposed in this pull request?
This PR adds the support to write CSV data directly by a single call to the given path.
Several unitests were added for each functionality.
## How was this patch tested?
This was tested with unittests and with `dev/run_tests` for coding style
Author: hyukjinkwon <gurwls223@gmail.com>
Author: Hyukjin Kwon <gurwls223@gmail.com>
Closes#11389 from HyukjinKwon/SPARK-13507-13509.
## What changes were proposed in this pull request?
Nested classes defined within Scala objects are translated into Java static nested classes. Unlike inner classes, they don't need outer scopes. But the analyzer still thinks that an outer scope is required.
This PR fixes this issue simply by checking whether a nested class is static before looking up its outer scope.
## How was this patch tested?
A test case is added to `DatasetSuite`. It checks contents of a Dataset whose element type is a nested class declared in a Scala object.
Author: Cheng Lian <lian@databricks.com>
Closes#11421 from liancheng/spark-13540-object-as-outer-scope.
JIRA: https://issues.apache.org/jira/browse/SPARK-13537
## What changes were proposed in this pull request?
In readBytes of VectorizedPlainValuesReader, we use buffer[offset] to access bytes in buffer. It is incorrect because offset is added with Platform.BYTE_ARRAY_OFFSET when initialization. We should fix it.
## How was this patch tested?
`ParquetHadoopFsRelationSuite` sometimes (depending on the randomly generated data) will be [failed](https://amplab.cs.berkeley.edu/jenkins/job/SparkPullRequestBuilder/52136/consoleFull) by this bug. After applying this, the test can be passed.
I added a test to `ParquetHadoopFsRelationSuite` with the data which will fail without this patch.
The error exception:
[info] ParquetHadoopFsRelationSuite:
[info] - test all data types - StringType (440 milliseconds)
[info] - test all data types - BinaryType (434 milliseconds)
[info] - test all data types - BooleanType (406 milliseconds)
20:59:38.618 ERROR org.apache.spark.executor.Executor: Exception in task 0.0 in stage 2597.0 (TID 67966)
java.lang.ArrayIndexOutOfBoundsException: 46
at org.apache.spark.sql.execution.datasources.parquet.VectorizedPlainValuesReader.readBytes(VectorizedPlainValuesReader.java:88)
Author: Liang-Chi Hsieh <viirya@gmail.com>
Closes#11418 from viirya/fix-readbytes.
## What changes were proposed in this pull request?
This creates a `SessionState`, which groups a few fields that existed in `SQLContext`. Because `HiveContext` extends `SQLContext` we also need to make changes there. This is mainly a cleanup task that will soon pave the way for merging the two contexts.
## How was this patch tested?
Existing unit tests; this patch introduces no change in behavior.
Author: Andrew Or <andrew@databricks.com>
Closes#11405 from andrewor14/refactor-session.
## What changes were proposed in this pull request?
Fix readBytes in VectorizedPlainValuesReader. This fixes a copy and paste issue.
## How was this patch tested?
Ran ParquetHadoopFsRelationSuite which failed before this.
Author: Nong Li <nong@databricks.com>
Closes#11414 from nongli/spark-13533.
JIRA: https://issues.apache.org/jira/browse/SPARK-13530
## What changes were proposed in this pull request?
By enabling vectorized parquet scanner by default, the unit test `ParquetHadoopFsRelationSuite` based on `HadoopFsRelationTest` will be failed due to the lack of short type support in `UnsafeRowParquetRecordReader`. We should fix it.
The error exception:
[info] ParquetHadoopFsRelationSuite:
[info] - test all data types - StringType (499 milliseconds)
[info] - test all data types - BinaryType (447 milliseconds)
[info] - test all data types - BooleanType (520 milliseconds)
[info] - test all data types - ByteType (418 milliseconds)
00:22:58.920 ERROR org.apache.spark.executor.Executor: Exception in task 0.0 in stage 124.0 (TID 1949)
org.apache.commons.lang.NotImplementedException: Unimplemented type: ShortType
at org.apache.spark.sql.execution.datasources.parquet.UnsafeRowParquetRecordReader$ColumnReader.readIntBatch(UnsafeRowParquetRecordReader.java:769)
at org.apache.spark.sql.execution.datasources.parquet.UnsafeRowParquetRecordReader$ColumnReader.readBatch(UnsafeRowParquetRecordReader.java:640)
at org.apache.spark.sql.execution.datasources.parquet.UnsafeRowParquetRecordReader$ColumnReader.access$000(UnsafeRowParquetRecordReader.java:461)
at org.apache.spark.sql.execution.datasources.parquet.UnsafeRowParquetRecordReader.nextBatch(UnsafeRowParquetRecordReader.java:224)
## How was this patch tested?
The unit test `ParquetHadoopFsRelationSuite` based on `HadoopFsRelationTest` will be [failed](https://amplab.cs.berkeley.edu/jenkins/job/SparkPullRequestBuilder/52110/consoleFull) due to the lack of short type support in UnsafeRowParquetRecordReader. By adding this support, the test can be passed.
Author: Liang-Chi Hsieh <viirya@gmail.com>
Closes#11412 from viirya/add-shorttype-support.
## What changes were proposed in this pull request?
Change the default of the flag to enable this feature now that the implementation is complete.
## How was this patch tested?
The new parquet reader should be a drop in, so will be exercised by the existing tests.
Author: Nong Li <nong@databricks.com>
Closes#11397 from nongli/spark-13518.
## What changes were proposed in this pull request?
This patch includes these performance fixes:
- Remove unnecessary setNotNull() calls. The NULL bits are cleared already.
- Speed up RLE group decoding
- Speed up dictionary decoding by decoding NULLs directly into the result.
## How was this patch tested?
(Please explain how this patch was tested. E.g. unit tests, integration tests, manual tests)
In addition to the updated benchmarks, on TPCDS, the result of these changes
running Q55 (sf40) is:
```
TPCDS: Best/Avg Time(ms) Rate(M/s) Per Row(ns)
---------------------------------------------------------------------------------
q55 (Before) 6398 / 6616 18.0 55.5
q55 (After) 4983 / 5189 23.1 43.3
```
Author: Nong Li <nong@databricks.com>
Closes#11375 from nongli/spark-13499.
## What changes were proposed in this pull request?
Currently, BroadcastNestedLoopJoin is implemented for worst case, it's too slow, very easy to hang forever. This PR will create fast path for some joinType and buildSide, also improve the worst case (will use much less memory than before).
Before this PR, one task requires O(N*K) + O(K) in worst cases, N is number of rows from one partition of streamed table, it could hang the job (because of GC).
In order to workaround this for InnerJoin, we have to disable auto-broadcast, switch to CartesianProduct: This could be workaround for InnerJoin, see https://forums.databricks.com/questions/6747/how-do-i-get-a-cartesian-product-of-a-huge-dataset.html
In this PR, we will have fast path for these joins :
InnerJoin with BuildLeft or BuildRight
LeftOuterJoin with BuildRight
RightOuterJoin with BuildLeft
LeftSemi with BuildRight
These fast paths are all stream based (take one pass on streamed table), required O(1) memory.
All other join types and build types will take two pass on streamed table, one pass to find the matched rows that includes streamed part, which require O(1) memory, another pass to find the rows from build table that does not have a matched row from streamed table, which required O(K) memory, K is the number rows from build side, one bit per row, should be much smaller than the memory for broadcast. The following join types work in this way:
LeftOuterJoin with BuildLeft
RightOuterJoin with BuildRight
FullOuterJoin with BuildLeft or BuildRight
LeftSemi with BuildLeft
This PR also added tests for all the join types for BroadcastNestedLoopJoin.
After this PR, for InnerJoin with one small table, BroadcastNestedLoopJoin should be faster than CartesianProduct, we don't need that workaround anymore.
## How was the this patch tested?
Added unit tests.
Author: Davies Liu <davies@databricks.com>
Closes#11328 from davies/nested_loop.
## What changes were proposed in this pull request?
This is another try of PR #11323.
This PR removes DataFrame RDD operations except for `foreach` and `foreachPartitions` (they are actions rather than transformations). Original calls are now replaced by calls to methods of `DataFrame.rdd`.
PR #11323 was reverted because it introduced a regression: both `DataFrame.foreach` and `DataFrame.foreachPartitions` wrap underlying RDD operations with `withNewExecutionId` to track Spark jobs. But they are removed in #11323.
## How was the this patch tested?
No extra tests are added. Existing tests should do the work.
Author: Cheng Lian <lian@databricks.com>
Closes#11388 from liancheng/remove-df-rdd-ops.
## What changes were proposed in this pull request?
This patch creates the public API for runtime configuration and an implementation for it. The public runtime configuration includes configs for existing SQL, as well as Hadoop Configuration.
This new interface is currently dead code. It will be added to SQLContext and a session entry point to Spark when we add that.
## How was this patch tested?
a new unit test suite
Author: Reynold Xin <rxin@databricks.com>
Closes#11378 from rxin/SPARK-13487.
## What changes were proposed in this pull request?
This Pull request is used for the fix SPARK-12941, creating a data type mapping to Oracle for the corresponding data type"Stringtype" from dataframe. This PR is for the master branch fix, where as another PR is already tested with the branch 1.4
## How was the this patch tested?
(Please explain how this patch was tested. E.g. unit tests, integration tests, manual tests)
This patch was tested using the Oracle docker .Created a new integration suite for the same.The oracle.jdbc jar was to be downloaded from the maven repository.Since there was no jdbc jar available in the maven repository, the jar was downloaded from oracle site manually and installed in the local; thus tested. So, for SparkQA test case run, the ojdbc jar might be manually placed in the local maven repository(com/oracle/ojdbc6/11.2.0.2.0) while Spark QA test run.
Author: thomastechs <thomas.sebastian@tcs.com>
Closes#11306 from thomastechs/master.
This pr added benchmark codes for Encoder#compress().
Also, it replaced the benchmark results with new ones because the output format of `Benchmark` changed.
Author: Takeshi YAMAMURO <linguin.m.s@gmail.com>
Closes#11236 from maropu/CompressionSpike.
## Motivation
As a pre-requisite to off-heap caching of blocks, we need a mechanism to prevent pages / blocks from being evicted while they are being read. With on-heap objects, evicting a block while it is being read merely leads to memory-accounting problems (because we assume that an evicted block is a candidate for garbage-collection, which will not be true during a read), but with off-heap memory this will lead to either data corruption or segmentation faults.
## Changes
### BlockInfoManager and reader/writer locks
This patch adds block-level read/write locks to the BlockManager. It introduces a new `BlockInfoManager` component, which is contained within the `BlockManager`, holds the `BlockInfo` objects that the `BlockManager` uses for tracking block metadata, and exposes APIs for locking blocks in either shared read or exclusive write modes.
`BlockManager`'s `get*()` and `put*()` methods now implicitly acquire the necessary locks. After a `get()` call successfully retrieves a block, that block is locked in a shared read mode. A `put()` call will block until it acquires an exclusive write lock. If the write succeeds, the write lock will be downgraded to a shared read lock before returning to the caller. This `put()` locking behavior allows us store a block and then immediately turn around and read it without having to worry about it having been evicted between the write and the read, which will allow us to significantly simplify `CacheManager` in the future (see #10748).
See `BlockInfoManagerSuite`'s test cases for a more detailed specification of the locking semantics.
### Auto-release of locks at the end of tasks
Our locking APIs support explicit release of locks (by calling `unlock()`), but it's not always possible to guarantee that locks will be released prior to the end of the task. One reason for this is our iterator interface: since our iterators don't support an explicit `close()` operator to signal that no more records will be consumed, operations like `take()` or `limit()` don't have a good means to release locks on their input iterators' blocks. Another example is broadcast variables, whose block locks can only be released at the end of the task.
To address this, `BlockInfoManager` uses a pair of maps to track the set of locks acquired by each task. Lock acquisitions automatically record the current task attempt id by obtaining it from `TaskContext`. When a task finishes, code in `Executor` calls `BlockInfoManager.unlockAllLocksForTask(taskAttemptId)` to free locks.
### Locking and the MemoryStore
In order to prevent in-memory blocks from being evicted while they are being read, the `MemoryStore`'s `evictBlocksToFreeSpace()` method acquires write locks on blocks which it is considering as candidates for eviction. These lock acquisitions are non-blocking, so a block which is being read will not be evicted. By holding write locks until the eviction is performed or skipped (in case evicting the blocks would not free enough memory), we avoid a race where a new reader starts to read a block after the block has been marked as an eviction candidate but before it has been removed.
### Locking and remote block transfer
This patch makes small changes to to block transfer and network layer code so that locks acquired by the BlockTransferService are released as soon as block transfer messages are consumed and released by Netty. This builds on top of #11193, a bug fix related to freeing of network layer ManagedBuffers.
## FAQ
- **Why not use Java's built-in [`ReadWriteLock`](https://docs.oracle.com/javase/7/docs/api/java/util/concurrent/locks/ReadWriteLock.html)?**
Our locks operate on a per-task rather than per-thread level. Under certain circumstances a task may consist of multiple threads, so using `ReadWriteLock` would mean that we might call `unlock()` from a thread which didn't hold the lock in question, an operation which has undefined semantics. If we could rely on Java 8 classes, we might be able to use [`StampedLock`](https://docs.oracle.com/javase/8/docs/api/java/util/concurrent/locks/StampedLock.html) to work around this issue.
- **Why not detect "leaked" locks in tests?**:
See above notes about `take()` and `limit`.
Author: Josh Rosen <joshrosen@databricks.com>
Closes#10705 from JoshRosen/pin-pages.
## What changes were proposed in this pull request?
This PR removes DataFrame RDD operations. Original calls are now replaced by calls to methods of `DataFrame.rdd`.
## How was the this patch tested?
No extra tests are added. Existing tests should do the work.
Author: Cheng Lian <lian@databricks.com>
Closes#11323 from liancheng/remove-df-rdd-ops.
## What changes were proposed in this pull request?
This patch moves SQLConf into org.apache.spark.sql.internal package to make it very explicit that it is internal. Soon I will also submit more API work that creates implementations of interfaces in this internal package.
## How was this patch tested?
If it compiles, then the refactoring should work.
Author: Reynold Xin <rxin@databricks.com>
Closes#11363 from rxin/SPARK-13486.
## What changes were proposed in this pull request?
This PR mostly rewrite the ColumnPruning rule to support most of the SQL logical plans (except those for Dataset).
This PR also fix a bug in Generate, it should always output UnsafeRow, added an regression test for that.
## How was this patch tested?
This is test by unit tests, also manually test with TPCDS Q78, which could prune all unused columns successfully, improved the performance by 78% (from 22s to 12s).
Author: Davies Liu <davies@databricks.com>
Closes#11354 from davies/fix_column_pruning.
## What changes were proposed in this pull request?
* Scala DataFrameStatFunctions: Added version of approxQuantile taking a List instead of an Array, for Python compatbility
* Python DataFrame and DataFrameStatFunctions: Added approxQuantile
## How was this patch tested?
* unit test in sql/tests.py
Documentation was copied from the existing approxQuantile exactly.
Author: Joseph K. Bradley <joseph@databricks.com>
Closes#11356 from jkbradley/approx-quantile-python.
This PR adds a new abstraction called an `ExpressionSet` which attempts to canonicalize expressions to remove cosmetic differences. Deterministic expressions that are in the set after canonicalization will always return the same answer given the same input (i.e. false positives should not be possible). However, it is possible that two canonical expressions that are not equal will in fact return the same answer given any input (i.e. false negatives are possible).
```scala
val set = AttributeSet('a + 1 :: 1 + 'a :: Nil)
set.iterator => Iterator('a + 1)
set.contains('a + 1) => true
set.contains(1 + 'a) => true
set.contains('a + 2) => false
```
Other relevant changes include:
- Since this concept overlaps with the existing `semanticEquals` and `semanticHash`, those functions are also ported to this new infrastructure.
- A memoized `canonicalized` version of the expression is added as a `lazy val` to `Expression` and is used by both `semanticEquals` and `ExpressionSet`.
- A set of unit tests for `ExpressionSet` are added
- Tests which expect `semanticEquals` to be less intelligent than it now is are updated.
As a followup, we should consider auditing the places where we do `O(n)` `semanticEquals` operations and replace them with `ExpressionSet`. We should also consider consolidating `AttributeSet` as a specialized factory for an `ExpressionSet.`
Author: Michael Armbrust <michael@databricks.com>
Closes#11338 from marmbrus/expressionSet.
Some parts of the engine rely on UnsafeRow which the vectorized parquet scanner does not want
to produce. This add a conversion in Physical RDD. In the case where codegen is used (and the
scan is the start of the pipeline), there is no requirement to use UnsafeRow. This patch adds
update PhysicallRDD to support codegen, which eliminates the need for the UnsafeRow conversion
in all cases.
The result of these changes for TPCDS-Q19 at the 10gb sf reduces the query time from 9.5 seconds
to 6.5 seconds.
Author: Nong Li <nong@databricks.com>
Closes#11141 from nongli/spark-13250.
## What changes were proposed in this pull request?
When we pass a Python function to JVM side, we also need to send its context, e.g. `envVars`, `pythonIncludes`, `pythonExec`, etc. However, it's annoying to pass around so many parameters at many places. This PR abstract python function along with its context, to simplify some pyspark code and make the logic more clear.
## How was the this patch tested?
by existing unit tests.
Author: Wenchen Fan <wenchen@databricks.com>
Closes#11342 from cloud-fan/python-clean.
JIRA: https://issues.apache.org/jira/browse/SPARK-13383
## What changes were proposed in this pull request?
When we do column pruning in Optimizer, we put additional Project on top of a logical plan. However, when we already wrap a BroadcastHint on a logical plan, the added Project will hide BroadcastHint after later execution.
We should take care of BroadcastHint when we do column pruning.
## How was the this patch tested?
Unit test is added.
Author: Liang-Chi Hsieh <viirya@gmail.com>
Closes#11260 from viirya/keep-broadcasthint.
