Compressing native executables with UPX
UPX is is an open-source, portable, high-performance executable packer initially created in 1996. It takes an executable as input and produces a compressed executable. Readers from a certain age may have already used UPX a long time ago when programs needed to be compressed to fit on a floppy disk. More than 20 years later, UPX is still rocking.
In Quarkus 2.6, we integrated UPX compression in the Quarkus build. So, Quarkus can automatically produce a compressed executable. This post explains how you can use this new feature. But, before going further, you need to understand that nothing comes for free. The compressed executable is smaller on disk, but its memory usage is higher. So make sure to read the "big warning" section before using it.
Getting you ready
Quarkus can compress your executable if you have installed UPX on your system or use an in-container build.
UPX is available on most operating systems. So, you should be able to download it from their release page. UPX is cross-platform. So you can compress a Linux 64 bits executable even from a macOS or Windows machine.
If installing UPX is not an option for you, you can ask Quarkus to build your native executable using an in-container build (with -Dquarkus.native.container-build=true
).
You will get a Linux 64 bits executable (perfect for containers but unusable from your machine if you don’t use Linux).
The in-container build provides UPX so that it can compress the executable.
When using an in-container build, you don’t need GraalVM or UPX on your machine.
This option is particularly interesting on CI.
In this post, we will use the first approach. The Building a Native Executable page explains how to use in-container build.
Let’s get something to compress
First, we need an application. To keep things simple, let’s create a new application from https://code.quarkus.io/?a=upx-compression-demo&e=resteasy-reactive-jacksoncode.quarkus.io. This application uses RESTEasy Reactive and its Jackson support, but the compression works with any project, and we won’t even look at the code.
Before using UPX, we need to measure the disk space and memory usage of the non-compressed native executable. To do that, we need the native executable:
> ./mvnw package -Dnative
The resulting executable use ~46 MB of disk space:
.rwxr-xr-x 46M clement 11 Dec 17:44 upx-compression-demo-1.0.0-SNAPSHOT-runner
Now, let’s have a look at the memory consumption. Start the application using:
> ./target/upx-compression-demo-1.0.0-SNAPSHOT-runner
In another terminal, invoke the application using curl and gets its memory usage:
> curl http://localhost:8080/hello
Hello RESTEasy Reactive%
> rss runner
PID 0M COMMAND
3947 21M ./target/upx-compression-demo-1.0.0-SNAPSHOT-runner
So, it takes 21 MB of RSS, basically the amount of RAM it uses.
The rss command is the following function:
|
rss () {
pgrep $1 | xargs ps -o pid,rss,command -p | awk '{$2=int($2/1024)"M";}{ print;}'
}
Check Quarkus - Measuring Performance to learn more about RSS and how to measure it
Configuring the compression
To compress your executable, you need to configure the compression level. The compression goes from 1 to 10:
-
1
: faster compression -
9
: better compression -
10
: best compression (can be slow for big files)
Configure the level from the application.properties
file:
quarkus.native.compression.level=7
That’s all you need to do to enable the compression.
Building the compressed native executable
Let’s regenerate the native executable. This time, because of the configured compression level, Quarkus will compress it:
> ./mvnw package -Dnative
...
...
[INFO] [io.quarkus.deployment.pkg.steps.UpxCompressionBuildStep] Executing /usr/local/bin/upx -7 /Users/clement/Downloads/upx-compression-demo/target/upx-compression-demo-1.0.0-SNAPSHOT-runner
Ultimate Packer for eXecutables
Copyright (C) 1996 - 2020
UPX 3.96 Markus Oberhumer, Laszlo Molnar & John Reiser Jan 23rd 2020
File size Ratio Format Name
-------------------- ------ ----------- -----------
46242352 -> 14774288 31.95% macho/amd64 upx-compression-demo-1.0.0-SNAPSHOT-runner
...
As you can see, this time, it runs UPX to compress the native executable. If you check the size, you should get something around 15 MB:
.rwxr-xr-x 15M clement 11 Dec 18:03 upx-compression-demo-1.0.0-SNAPSHOT-runner
So we went from 46 M to 15 M; this is a considerable gain, even if it still does not fit on a floppy disk.
The BIG warning
However, as said in the introduction, nothing comes for free. Earlier, we also measured the memory usage of the uncompressed executable (21 MB). Let’s compare with the compressed executable.
Run the application with:
> ./target/upx-compression-demo-1.0.0-SNAPSHOT-runner
And, in another terminal, run:
> curl http://localhost:8080/hello
Hello RESTEasy Reactive%
> rss runner
PID 0M COMMAND
4501 57M ./target/upx-compression-demo-1.0.0-SNAPSHOT-runner
57 MB! So, it uses ~2.7 times more RSS than the uncompressed executable. This overhead is because the compressed executable must unpack the program on startup and store it in memory. It can also increase the startup time, but this startup overhead is insignificant most of the time.
Summary
UPX lets you compress your native executables. In Quarkus 2.6, you only need to configure the compression level, and voilà, it compresses it for you.
However, do not think it all comes for free. While the gain on disk space is fantastic, do not ignore the RSS overhead.
UPX compression can benefit CLI tools or environments where disk space is a constrained resource. For long-running applications or microservices, the RSS overhead reduces the deployment density. So, if the storage is not a problem or if the density of deployment is crucial for you, better not compress your executable.