7539ae59d6
## What changes were proposed in this pull request? `ReadAheadInputStream` was introduced in https://github.com/apache/spark/pull/18317/ to optimize reading spill files from disk. However, from the profiles it seems that the hot path of reading small amounts of data (like readInt) is inefficient - it involves taking locks, and multiple checks. Optimize locking: Lock is not needed when simply accessing the active buffer. Only lock when needing to swap buffers or trigger async reading, or get information about the async state. Optimize short-path single byte reads, that are used e.g. by Java library DataInputStream.readInt. The asyncReader used to call "read" only once on the underlying stream, that never filled the underlying buffer when it was wrapping an LZ4BlockInputStream. If the buffer was returned unfilled, that would trigger the async reader to be triggered to fill the read ahead buffer on each call, because the reader would see that the active buffer is below the refill threshold all the time. However, filling the full buffer all the time could introduce increased latency, so also add an `AtomicBoolean` flag for the async reader to return earlier if there is a reader waiting for data. Remove `readAheadThresholdInBytes` and instead immediately trigger async read when switching the buffers. It allows to simplify code paths, especially the hot one that then only has to check if there is available data in the active buffer, without worrying if it needs to retrigger async read. It seems to have positive effect on perf. ## How was this patch tested? It was noticed as a regression in some workloads after upgrading to Spark 2.3. It was particularly visible on TPCDS Q95 running on instances with fast disk (i3 AWS instances). Running with profiling: * Spark 2.2 - 5.2-5.3 minutes 9.5% in LZ4BlockInputStream.read * Spark 2.3 - 6.4-6.6 minutes 31.1% in ReadAheadInputStream.read * Spark 2.3 + fix - 5.3-5.4 minutes 13.3% in ReadAheadInputStream.read - very slightly slower, practically within noise. We didn't see other regressions, and many workloads in general seem to be faster with Spark 2.3 (not investigated if thanks to async readed, or unrelated). Author: Juliusz Sompolski <julek@databricks.com> Closes #20555 from juliuszsompolski/SPARK-23366. |
||
---|---|---|
.github | ||
assembly | ||
bin | ||
build | ||
common | ||
conf | ||
core | ||
data | ||
dev | ||
docs | ||
examples | ||
external | ||
graphx | ||
hadoop-cloud | ||
launcher | ||
licenses | ||
mllib | ||
mllib-local | ||
project | ||
python | ||
R | ||
repl | ||
resource-managers | ||
sbin | ||
sql | ||
streaming | ||
tools | ||
.gitattributes | ||
.gitignore | ||
.travis.yml | ||
appveyor.yml | ||
CONTRIBUTING.md | ||
LICENSE | ||
NOTICE | ||
pom.xml | ||
README.md | ||
scalastyle-config.xml |
Apache Spark
Spark is a fast and general cluster computing system for Big Data. 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 Spark 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.)
You can build Spark using more than one thread by using the -T option with Maven, see "Parallel builds in Maven 3". 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 1000:
scala> sc.parallelize(1 to 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 1000:
>>> sc.parallelize(range(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.
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" 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.