1fc66f6870
### What changes were proposed in this pull request? This PR is to support nested column type in Spark ORC vectorized reader. Currently ORC vectorized reader [does not support nested column type (struct, array and map)](https://github.com/apache/spark/blob/master/sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/orc/OrcFileFormat.scala#L138). We implemented nested column vectorized reader for FB-ORC in our internal fork of Spark. We are seeing performance improvement compared to non-vectorized reader when reading nested columns. In addition, this can also help improve the non-nested column performance when reading non-nested and nested columns together in one query. Before this PR: * `OrcColumnVector` is the implementation class for Spark's `ColumnVector` to wrap Hive's/ORC's `ColumnVector` to read `AtomicType` data. After this PR: * `OrcColumnVector` is an abstract class to keep interface being shared between multiple implementation class of orc column vectors, namely `OrcAtomicColumnVector` (for `AtomicType`), `OrcArrayColumnVector` (for `ArrayType`), `OrcMapColumnVector` (for `MapType`), `OrcStructColumnVector` (for `StructType`). So the original logic to read `AtomicType` data is moved from `OrcColumnVector` to `OrcAtomicColumnVector`. The abstract class of `OrcColumnVector` is needed here because of supporting nested column (i.e. nested column vectors). * A utility method `OrcColumnVectorUtils.toOrcColumnVector` is added to create Spark's `OrcColumnVector` from Hive's/ORC's `ColumnVector`. * A new user-facing config `spark.sql.orc.enableNestedColumnVectorizedReader` is added to control enabling/disabling vectorized reader for nested columns. The default value is false (i.e. disabling by default). For certain tables having deep nested columns, vectorized reader might take too much memory for each sub-column vectors, compared to non-vectorized reader. So providing a config here to work around OOM for query reading wide and deep nested columns if any. We plan to enable it by default on 3.3. Leave it disable in 3.2 in case for any unknown bugs. ### Why are the changes needed? Improve query performance when reading nested columns from ORC file format. Tested with locally adding a small benchmark in `OrcReadBenchmark.scala`. Seeing more than 1x run time improvement. ``` Running benchmark: SQL Nested Column Scan Running case: Native ORC MR Stopped after 2 iterations, 37850 ms Running case: Native ORC Vectorized (Enabled Nested Column) Stopped after 2 iterations, 15892 ms Running case: Native ORC Vectorized (Disabled Nested Column) Stopped after 2 iterations, 37954 ms Running case: Hive built-in ORC Stopped after 2 iterations, 35118 ms Java HotSpot(TM) 64-Bit Server VM 1.8.0_181-b13 on Mac OS X 10.15.7 Intel(R) Core(TM) i9-9980HK CPU 2.40GHz SQL Nested Column Scan: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative ------------------------------------------------------------------------------------------------------------------------------ Native ORC MR 18706 18925 310 0.1 17839.6 1.0X Native ORC Vectorized (Enabled Nested Column) 7625 7946 455 0.1 7271.6 2.5X Native ORC Vectorized (Disabled Nested Column) 18415 18977 796 0.1 17561.5 1.0X Hive built-in ORC 17469 17559 127 0.1 16660.1 1.1X ``` Benchmark: ``` nestedColumnScanBenchmark(1024 * 1024) def nestedColumnScanBenchmark(values: Int): Unit = { val benchmark = new Benchmark(s"SQL Nested Column Scan", values, output = output) withTempPath { dir => withTempTable("t1", "nativeOrcTable", "hiveOrcTable") { import spark.implicits._ spark.range(values).map(_ => Random.nextLong).map { x => val arrayOfStructColumn = (0 until 5).map(i => (x + i, s"$x" * 5)) val mapOfStructColumn = Map( s"$x" -> (x * 0.1, (x, s"$x" * 100)), (s"$x" * 2) -> (x * 0.2, (x, s"$x" * 200)), (s"$x" * 3) -> (x * 0.3, (x, s"$x" * 300))) (arrayOfStructColumn, mapOfStructColumn) }.toDF("col1", "col2") .createOrReplaceTempView("t1") prepareTable(dir, spark.sql(s"SELECT * FROM t1")) benchmark.addCase("Native ORC MR") { _ => withSQLConf(SQLConf.ORC_VECTORIZED_READER_ENABLED.key -> "false") { spark.sql("SELECT SUM(SIZE(col1)), SUM(SIZE(col2)) FROM nativeOrcTable").noop() } } benchmark.addCase("Native ORC Vectorized (Enabled Nested Column)") { _ => spark.sql("SELECT SUM(SIZE(col1)), SUM(SIZE(col2)) FROM nativeOrcTable").noop() } benchmark.addCase("Native ORC Vectorized (Disabled Nested Column)") { _ => withSQLConf(SQLConf.ORC_VECTORIZED_READER_NESTED_COLUMN_ENABLED.key -> "false") { spark.sql("SELECT SUM(SIZE(col1)), SUM(SIZE(col2)) FROM nativeOrcTable").noop() } } benchmark.addCase("Hive built-in ORC") { _ => spark.sql("SELECT SUM(SIZE(col1)), SUM(SIZE(col2)) FROM hiveOrcTable").noop() } benchmark.run() } } } ``` ### Does this PR introduce _any_ user-facing change? No. ### How was this patch tested? Added one simple test in `OrcSourceSuite.scala` to verify correctness. Definitely need more unit tests and add benchmark here, but I want to first collect feedback before crafting more tests. Closes #31958 from c21/orc-vector. Authored-by: Cheng Su <chengsu@fb.com> Signed-off-by: Liang-Chi Hsieh <viirya@gmail.com> |
||
---|---|---|
.github | ||
assembly | ||
bin | ||
binder | ||
build | ||
common | ||
conf | ||
core | ||
data | ||
dev | ||
docs | ||
examples | ||
external | ||
graphx | ||
hadoop-cloud | ||
launcher | ||
licenses | ||
licenses-binary | ||
mllib | ||
mllib-local | ||
project | ||
python | ||
R | ||
repl | ||
resource-managers | ||
sbin | ||
sql | ||
streaming | ||
tools | ||
.asf.yaml | ||
.gitattributes | ||
.gitignore | ||
.sbtopts | ||
appveyor.yml | ||
CONTRIBUTING.md | ||
LICENSE | ||
LICENSE-binary | ||
NOTICE | ||
NOTICE-binary | ||
pom.xml | ||
README.md | ||
scalastyle-config.xml |
Apache Spark
Spark is a unified analytics engine for large-scale data processing. It provides high-level APIs in Scala, Java, Python, and R, and an optimized engine that supports general computation graphs for data analysis. It also supports a rich set of higher-level tools including Spark SQL for SQL and DataFrames, MLlib for machine learning, GraphX for graph processing, and Structured Streaming for stream processing.
Online Documentation
You can find the latest Spark documentation, including a programming guide, on the project web page. This README file only contains basic setup instructions.
Building Spark
Spark is built using Apache Maven. To build Spark and its example programs, run:
./build/mvn -DskipTests clean package
(You do not need to do this if you downloaded a pre-built package.)
More detailed documentation is available from the project site, at "Building Spark".
For general development tips, including info on developing Spark using an IDE, see "Useful Developer Tools".
Interactive Scala Shell
The easiest way to start using Spark is through the Scala shell:
./bin/spark-shell
Try the following command, which should return 1,000,000,000:
scala> spark.range(1000 * 1000 * 1000).count()
Interactive Python Shell
Alternatively, if you prefer Python, you can use the Python shell:
./bin/pyspark
And run the following command, which should also return 1,000,000,000:
>>> spark.range(1000 * 1000 * 1000).count()
Example Programs
Spark also comes with several sample programs in the examples
directory.
To run one of them, use ./bin/run-example <class> [params]
. For example:
./bin/run-example SparkPi
will run the Pi example locally.
You can set the MASTER environment variable when running examples to submit
examples to a cluster. This can be a mesos:// or spark:// URL,
"yarn" to run on YARN, and "local" to run
locally with one thread, or "local[N]" to run locally with N threads. You
can also use an abbreviated class name if the class is in the examples
package. For instance:
MASTER=spark://host:7077 ./bin/run-example SparkPi
Many of the example programs print usage help if no params are given.
Running Tests
Testing first requires building Spark. Once Spark is built, tests can be run using:
./dev/run-tests
Please see the guidance on how to run tests for a module, or individual tests.
There is also a Kubernetes integration test, see resource-managers/kubernetes/integration-tests/README.md
A Note About Hadoop Versions
Spark uses the Hadoop core library to talk to HDFS and other Hadoop-supported storage systems. Because the protocols have changed in different versions of Hadoop, you must build Spark against the same version that your cluster runs.
Please refer to the build documentation at "Specifying the Hadoop Version and Enabling YARN" for detailed guidance on building for a particular distribution of Hadoop, including building for particular Hive and Hive Thriftserver distributions.
Configuration
Please refer to the Configuration Guide in the online documentation for an overview on how to configure Spark.
Contributing
Please review the Contribution to Spark guide for information on how to get started contributing to the project.