[
https://issues.apache.org/jira/browse/PARQUET-2202?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17693672#comment-17693672
]
ASF GitHub Bot commented on PARQUET-2202:
-----------------------------------------
jerolba opened a new pull request, #1035:
URL: https://github.com/apache/parquet-mr/pull/1035
The [original Jira ticket
](https://issues.apache.org/jira/browse/PARQUET-2202) references to a concrete
bad usage of the `Preconditions.checkArgument` method, where a String is
calculated before validating the check, creating an overhead when in theory
100% of the cases the composed String will not be used.
The proposed solution inlines the call to `Preconditions.checkArgument`, but
the correct approach is to call the `checkArgument` method with a String to
format if the argument check is not valid.
A similar issue ocurrs in the
[constructor](https://github.com/apache/parquet-mr/blob/62b774cd0f0c60cfbe540bbfa60bee15929af5d4/parquet-common/src/main/java/org/apache/parquet/bytes/CapacityByteArrayOutputStream.java#L153)
of the `CapacityByteArrayOutputStream` class, where the error message is
always created using the `String.format`, instead of passing the template
String and its params to the `checkArgument` method. This performance issue is
also visible in a profiling.
The MR fixes both cases, and reviews the usage of `Preconditions` class, to
ensure that error messages are not calculated before checking the boolean
expression.
I've also reviewed `Preconditions` methods to improve the performance. When
you call to a method with a varargs argument, Java internally allocates an
array containing all values. To avoid this allocation when the number of params
is very low, is recomended to overload the method with versions of the method
with different number of arguments. This approach is heavily used in logging
frameworks, or in [Guava Preconditions
implementation](https://github.com/google/guava/blob/4312d949967f3fb245636f66437a00dd8c346d38/guava/src/com/google/common/base/Preconditions.java#L118)
Because nearly 100% of cases are going to check the condition as valid, the
Array associated with varargs call to `Preconditions.strings` method will never
be created.
Because all changes are related with `Preconditions.checkargument`, I've
created a single PR. I can split it in multiple PRs if needed. Each commit of
the PR makes a type of change, but I can squash into a single commit if needed.
Make sure you have checked _all_ steps below.
### Jira
- [x] My PR addresses the following [Parquet
Jira](https://issues.apache.org/jira/browse/PARQUET/) issues and references
them in the PR title. For example, "PARQUET-2202: My Parquet PR"
- https://issues.apache.org/jira/browse/PARQUET-2202
- In case you are adding a dependency, check if the license complies with
the [ASF 3rd Party License
Policy](https://www.apache.org/legal/resolved.html#category-x).
### Tests
- [x] My PR adds test cases to the following unit tests:
1.
https://github.com/jerolba/parquet-mr/blob/review_usage_of_preconditions_checkargument/parquet-common/src/test/java/org/apache/parquet/TestPreconditions.java
### Commits
- [x] My commits all reference Jira issues in their subject lines. In
addition, my commits follow the guidelines from "[How to write a good git
commit message](http://chris.beams.io/posts/git-commit/)":
1. Subject is separated from body by a blank line
1. Subject is limited to 50 characters (not including Jira issue reference)
1. Subject does not end with a period
1. Subject uses the imperative mood ("add", not "adding")
1. Body wraps at 72 characters
1. Body explains "what" and "why", not "how"
### Documentation
- [x] In case of new functionality, my PR adds documentation that describes
how to use it.
- All the public functions and the classes in the PR contain Javadoc that
explain what it does
> Redundant String allocation on the hot path in
> CapacityByteArrayOutputStream.setByte
> ------------------------------------------------------------------------------------
>
> Key: PARQUET-2202
> URL: https://issues.apache.org/jira/browse/PARQUET-2202
> Project: Parquet
> Issue Type: Bug
> Components: parquet-mr
> Affects Versions: 1.12.3
> Reporter: Andrei Pangin
> Priority: Major
> Labels: performance
> Attachments: profile-alloc.png, profile-cpu.png
>
>
> Profiling of a Spark application revealed a performance issue in production:
> {{CapacityByteArrayOutputStream.setByte}} consumed 2.2% of total CPU time and
> made up 4.6% of total allocations. However, in normal case, this method
> should allocate nothing at all.
> Here is an excerpt from async-profiler report.
> CPU profile:
> !profile-cpu.png|width=560!
> Allocation profile:
> !profile-alloc.png|width=560!
> The reason is a {{checkArgument()}} call with an unconditionally constructed
> dynamic String:
> [https://github.com/apache/parquet-mr/blob/62b774cd0f0c60cfbe540bbfa60bee15929af5d4/parquet-common/src/main/java/org/apache/parquet/bytes/CapacityByteArrayOutputStream.java#L303]
> The suggested fix is to move String construction under the condition:
> {code:java}
> if (index >= bytesUsed) {
> throw new IllegalArgumentException("Index: " + index +
> " is >= the current size of: " + bytesUsed);
> }{code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
