c21 opened a new pull request #34640:
URL: https://github.com/apache/spark/pull/34640
<!--
Thanks for sending a pull request! Here are some tips for you:
1. If this is your first time, please read our contributor guidelines:
https://spark.apache.org/contributing.html
2. Ensure you have added or run the appropriate tests for your PR:
https://spark.apache.org/developer-tools.html
3. If the PR is unfinished, add '[WIP]' in your PR title, e.g.,
'[WIP][SPARK-XXXX] Your PR title ...'.
4. Be sure to keep the PR description updated to reflect all changes.
5. Please write your PR title to summarize what this PR proposes.
6. If possible, provide a concise example to reproduce the issue for a
faster review.
7. If you want to add a new configuration, please read the guideline first
for naming configurations in
'core/src/main/scala/org/apache/spark/internal/config/ConfigEntry.scala'.
8. If you want to add or modify an error type or message, please read the
guideline first in
'core/src/main/resources/error/README.md'.
-->
### What changes were proposed in this pull request?
<!--
Please clarify what changes you are proposing. The purpose of this section
is to outline the changes and how this PR fixes the issue.
If possible, please consider writing useful notes for better and faster
reviews in your PR. See the examples below.
1. If you refactor some codes with changing classes, showing the class
hierarchy will help reviewers.
2. If you fix some SQL features, you can provide some references of other
DBMSes.
3. If there is design documentation, please add the link.
4. If there is a discussion in the mailing list, please add the link.
-->
This PR is to introduce a new expression in Spark - `ZOrder`. The motivation
is Z-order enables to sort tuples in a way, to allow efficiently data skipping
for columnar file format (Parquet and ORC).
For query with filter on combination of multiple columns, example:
```sql
SELECT *
FROM table
WHERE x = 0 OR y = 0
```
Parquet/ORC cannot skip file/row-groups efficiently when reading, even
though the table is sorted (locally or globally) on any columns. However when
table is Z-order sorted on multiple columns, Parquet/ORC can skip
file/row-groups efficiently when reading. We should add the feature in Spark to
allow OSS Spark users benefitted in running these queries.
With this PR, user can do Z-order sort when writing the table with followed
syntax:
```sql
INSERT INTO t
SELECT ...
FROM ...
SORT BY ZORDER(x, y, ...)
```
or
```sql
INSERT INTO t
SELECT ...
FROM ...
ORDER BY ZORDER(x, y, ...)
```
Then when reading the table with filter on `x` and `y`, the performance can
be improved by skipping more files and row-groups. More details below for micro
benchmark.
This PR adds the support for Z-order on integer types (byte, short, int, and
long). For other data types such as float and string will be added as followup.
Code-gen support for expression will be also added later.
### Why are the changes needed?
<!--
Please clarify why the changes are needed. For instance,
1. If you propose a new API, clarify the use case for a new API.
2. If you fix a bug, you can clarify why it is a bug.
-->
To improve the query performance when filtering on multiple columns. Seeing
1x-6x run-time improvement in micro benchmark below.
```scala
override def runBenchmarkSuite(mainArgs: Array[String]): Unit = {
def prepareTable(dir: File, numRows: Int): Unit = {
import spark.implicits._
val df = spark.range(numRows).map(_ => (Random.nextLong,
Random.nextLong))
.toDF("x", "y")
val zorderedDf = df.sort(Column(ZOrder(Seq($"x".expr, $"y".expr))))
saveAsTable(df, dir, "")
saveAsTable(zorderedDf, dir, "ZOrder")
}
def saveAsTable(df: DataFrame, dir: File, suffix: String): Unit = {
val blockSize =
org.apache.parquet.hadoop.ParquetWriter.DEFAULT_PAGE_SIZE
val orcPath = dir.getCanonicalPath + "/orc" + suffix
val parquetPath = dir.getCanonicalPath + "/parquet" + suffix
df.write.mode("overwrite")
.option("orc.dictionary.key.threshold", 0.8)
.option("orc.compress.size", blockSize)
.option("orc.stripe.size", blockSize).orc(orcPath)
spark.read.orc(orcPath).createOrReplaceTempView("orcTable" + suffix)
df.write.mode("overwrite")
.option("parquet.block.size", blockSize).parquet(parquetPath)
spark.read.parquet(parquetPath).createOrReplaceTempView("parquetTable"
+ suffix)
}
def withTempTable(tableNames: String*)(f: => Unit): Unit = {
try f finally tableNames.foreach(spark.catalog.dropTempView)
}
runBenchmark(s"ZOrder") {
withTempPath { dir =>
withTempTable("orcTable", "parquetTable", "orcTableZOrder",
"parquetTableZOrder") {
prepareTable(dir, 1024 * 1024 * 15)
val benchmark = new Benchmark("zorder", 1024 * 1024 * 15,
minNumIters = 5, output = output)
benchmark.addCase("Parquet no sort") { _ =>
spark.sql(s"SELECT * FROM parquetTable WHERE x = 0 OR y =
0").noop()
}
benchmark.addCase("Parquet z-order sort on (x, y)") { _ =>
spark.sql(s"SELECT * FROM parquetTableZOrder WHERE x = 0 OR y =
0").noop()
}
benchmark.addCase("ORC no sort") { _ =>
spark.sql(s"SELECT * FROM orcTable WHERE x = 0 OR y = 0").noop()
}
benchmark.addCase("ORC z-order sort on (x, y)") { _ =>
spark.sql(s"SELECT * FROM orcTableZOrder WHERE x = 0 OR y =
0").noop()
}
benchmark.run()
}
}
}
}
```
* Compare the performance between reading table having no sort, and table
having local Z-order sort on `(x, y)`.
Seeing 6x run-time improvement for Parquet, and 1x for ORC:
```
Java HotSpot(TM) 64-Bit Server VM 1.8.0_181-b13 on Mac OS X 10.16
Intel(R) Core(TM) i9-9980HK CPU @ 2.40GHz
zorder: Best Time(ms) Avg Time(ms)
Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
Parquet no sort 274 287
11 57.5 17.4 1.0X
Parquet z-order sort on (x, y) 37 41
2 420.2 2.4 7.3X
ORC no sort 674 754
47 23.3 42.8 0.4X
ORC z-order sort on (x, y) 262 282
11 59.9 16.7 1.0X
```
* Compare the performance between reading table having no sort, and table
having local sort on `(x, y)`.
No performance improvement as expected.
```
Java HotSpot(TM) 64-Bit Server VM 1.8.0_181-b13 on Mac OS X 10.16
Intel(R) Core(TM) i9-9980HK CPU @ 2.40GHz
zorder: Best Time(ms) Avg Time(ms)
Stdev(ms) Rate(M/s) Per Row(ns) Relative
------------------------------------------------------------------------------------------------------------------------
Parquet no sort 285 319
29 55.1 18.1 1.0X
Parquet sort on (x, y) 278 290
10 56.5 17.7 1.0X
ORC no sort 823 842
21 19.1 52.4 0.3X
ORC sort on (x, y) 748 760
16 21.0 47.6 0.4X
```
### Does this PR introduce _any_ user-facing change?
<!--
Note that it means *any* user-facing change including all aspects such as
the documentation fix.
If yes, please clarify the previous behavior and the change this PR proposes
- provide the console output, description and/or an example to show the
behavior difference if possible.
If possible, please also clarify if this is a user-facing change compared to
the released Spark versions or within the unreleased branches such as master.
If no, write 'No'.
-->
Yes. The added expression can be used by user - `zorder`.
### How was this patch tested?
<!--
If tests were added, say they were added here. Please make sure to add some
test cases that check the changes thoroughly including negative and positive
cases if possible.
If it was tested in a way different from regular unit tests, please clarify
how you tested step by step, ideally copy and paste-able, so that other
reviewers can test and check, and descendants can verify in the future.
If tests were not added, please describe why they were not added and/or why
it was difficult to add.
If benchmark tests were added, please run the benchmarks in GitHub Actions
for the consistent environment, and the instructions could accord to:
https://spark.apache.org/developer-tools.html#github-workflow-benchmarks.
-->
Added unit test in `ZOrderExpressionSuite.scala`.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
