Lecture Tutorial GraalVM

Goals of this tutorial:

• Compare GraalVM native image and JVM using a basic SpringBoot application
• Examine differences in size, execution time, and resource consumption
• Determine scenarios where GraalVM native image is more advantageous than JVM and vice versa

Prerequisites

Download and install git for your operating system.

Docker requires WSL (Windows Subsystem for Linux)

wsl --install

Download and install Docker Desktop on Windows or Docker Engine and Docker Compose on Linux.

Download and install Python 3.11) (or later).

Because the two SpringBoot applications "graalvm-demo-book" and "graalvm-demo-book-utilizer" are built inside of Docker, there is no need to install GraalVM or Java locally.

Project Structure

lecture-graalvm
├── graalvm-demo-book
│   ├── Main Application
│   └── Insert books into a MongoDB
│
├── graalvm-demo-book-utilizer
│   ├── Utilizer Application
│   └── Create a load-test of the Main Application
│
├── load-test
│   ├── Python Scripts
│   └── Run multiple load-tests and evalute the results
│
└── monitoring
    ├── Grafana and Prometheus config
    └── Preconfigured Grafana dashboard with container monitoring

The Book Demo Application

The Project Structure outlines the two key applications: graalvm-demo-book and graalvm-demo-book-utilizer.

The graalvm-demo-book application manages books and provides Rest endpoints to add, delete and get books. This is the application we use for the comparison.

The graalvm-demo-book-utilizer application creates multiple load tests for the graalvm-demo-book app.

The components of a book are as follows:

Name	Type	Description
id	String	Unique ID for the Book
title	String	The book's title
author	String	The book's author
pageCount	Integer	The book's page count

A simplified diagram of the main components.

+-------------------+       +-----------------------+
|                   |       |                       |
| graalvm-demo-book |<------| graalvm-demo-utilizer |
|                   |       |                       | 
|    POST /books    |       |    POST /load-test    |   
|    insert books   |       |    - configure url,   |
|                   |       |       endpoint        |
+---------|---------+       |    - set numBooks,    |
          |                 |       numRequests,    |
          |                 |       numUsers        |
          v                 |                       |
 +-----------------+        +-----------------------+
 |                 |
 |    MongoDB      |
 |   Store books   |
 |                 |
 +-----------------+

1. Demo Application with JVM

First, we have start with a normal JVM and look how our demo application behaves.

You can use any command line tool (Terminal, PowerShell, Bash, ...)

1.1 Build Start JVM demo app

Clone the GitHub repository

git clone https://github.com/envite-consulting/showcase-graalvm.git

Navigate to the directory (the entire rest of this tutorial can be executed from this directory)

cd showcase-graalvm

Prepare folder structure (Linux only)

mkdir target && chmod -R g+rX,o+rX target
chmod -R g+rX,o+rX monitoring

Pull all images, e.g. mongo db

docker compose pull --ignore-buildable

Build JVM image:

docker compose build graalvm-demo-book-jvm

Dockerfile: graalvm-demo-book/Dockerfile.jvm

• Docker multi-stage build
• SpringBoot layered image

Run app and mongodb

docker compose up -d graalvm-demo-book-jvm 

Show running containers

docker compose ps

View start log and find the startup time of the application

docker compose logs -f graalvm-demo-book-jvm

1.2 Play with Book API

On Windows use Git console or another bash like terminal to run the following curl commands. Alternatively you can install VS Code Plugin "Rest client" or ItelliJ's "Services" und run the http requests via requests.http

Create a new book

time curl -v -XPOST -d'{"id":"978-3-8477-1359-3","title":"Nils Holgerssons wunderbare Reise durch Schweden","author":"Selma Lagerlöf","pageCount":704}' -H'Content-Type: application/json; charset=utf-8' http://localhost:8080/books

How long did the first request take?

If you want, you can play a little bit with the API.

curl http://localhost:8080/books/978-3-8477-1359-3

curl http://localhost:8080/books

curl -v -XPOST http://localhost:8080/books/bulk \
-H 'Content-Type: application/json; charset=utf-8' \
--data-binary @- << EOF
[
  {"id":"978-0-345-40946-1","title":"The Demon-Haunted World","author":"Carl Sagan, Ann Druyan","pageCount":480},
  {"id":"978-0-345-53943-4","title":"Cosmos","author":"Carl Sagan","pageCount":432}
]
EOF

curl http://localhost:8080/books/bulk/978-0-345-40946-1,978-0-345-53943-4

1.3 Discussion

What did you discover? Is all good or do you think there are some issues especially related to Cloud?

2. GraalVM Native Image

What is GraalVM?

• AOT: Ahead-of-time compilation
• Low Memory Footprint
• Self-contained executables
• Reduced startup time

--> How does it solve our problem? Compilation to native binary moves stuff from runtime to compile time.

2.1 Build Start demo app build with GraalVM native image

Build native image:

docker compose build graalvm-demo-book-native

Dockerfile: graalvm-demo-book/Dockerfile.native

• Docker multi-stage build
• ./mvnw -Pnative,musl native:compile # what happens here?

Run app and mongodb

docker compose up -d graalvm-demo-book-native 

Show running containers

docker compose ps

View start log and find the startup time of the application

docker compose logs -f graalvm-demo-book-native

2.2 Execute Request against Book API

Create a new book

time curl -v -XPOST -d'{"id":"978-3-8477-1359-3","title":"Nils Holgerssons wunderbare Reise durch Schweden","author":"Selma Lagerlöf","pageCount":704}' -H'Content-Type: application/json; charset=utf-8' http://localhost:8084/books

How long did the first request take?

curl http://localhost:8084/books/978-3-8477-1359-3

2.3 Discussion

What did you discover? What is better now?

4. Is GraalVM really always better?

4.1 Docker Image Layers

Dive is a tool for exploring a docker image and layer contents.

Let's use it to analyze our two Docker images.

JVM:

docker run --rm -it \
    -v /var/run/docker.sock:/var/run/docker.sock \
    wagoodman/dive:latest graalvm-demo-book-jvm:0.1.0

Native:

docker run --rm -it \
    -v /var/run/docker.sock:/var/run/docker.sock \
    wagoodman/dive:latest graalvm-demo-book-native:0.1.0

Which differences do you see? Do you see problems, which one is better?

4.2 Analysis of resource consumption and behaviour over long run

Stop Application

Stop all running containers

docker compose down -v

Start Monitoring

docker compose up -d prometheus cadvisor grafana

Open Grafana: http://localhost:3000/d/edonzk2655t6oc/lecture-graalvm

Start the two containers again

docker compose up -d graalvm-demo-book-jvm graalvm-demo-book-native

Which differences do you see on the Grafana Dashboard?

Run utilizer

On Windows use Git console or another bash like terminal to run the following curl commands. Alternatively you can install VS Code Plugin "Rest client" or ItelliJ's "Services" und run the http requests via requests_load-test.http.

These are the parameters of graalvm-demo-book-utilizer

Name	Type	Description
webClientUrl	String	The URL of the service to be load-tested
webClientEndpoint	String	The endpoint of the service
numberOfBooks	Integer	The number of books
numberOfRequests	Integer	The number of requests
concurrentUsers	Integer	The number of simulated concurrent users via parallel threading

Each simulated user (via parallel threading) will send the specified number of requests (numberOfRequests).

User
├── Request-1
│   └── numberOfBooks books
│    
├── Request-2
│   └── numberOfBooks books
...
└── Request-n

The total number of books that are written to the mongoDB are: concurrentUsers * numberOfRequests * numberOfBooks.

Build the utilizer:

docker compose build graalvm-demo-book-utilizer

Start the utilizer

docker compose up -d graalvm-demo-book-utilizer

Run a simple load-test on graalvm-demo-book-jvm

curl -v -XPOST http://localhost:8085/load-test \
-H 'Content-Type: application/json; charset=utf-8' \
--data-binary @- << EOF
{
    "webClientUrl": "http://graalvm-demo-book-jvm:8080",
    "webClientEndpoint": "/books/bulk",
    "numberOfBooks": 20,
    "numberOfRequests": 10,
    "concurrentUsers": 1
}
EOF

Run the same load-test on graalvm-demo-book-native

curl -v -XPOST http://localhost:8085/load-test \
-H 'Content-Type: application/json; charset=utf-8' \
--data-binary @- << EOF
{
    "webClientUrl": "http://graalvm-demo-book-native:8080",
    "webClientEndpoint": "/books/bulk",
    "numberOfBooks": 20,
    "numberOfRequests": 10,
    "concurrentUsers": 1
}
EOF

Try changing the parameters and compare the results.

Which differences do you see?

Run python load-test

Create Python environment and install requirements (Windows)

python -m 'venv' .venv
.venv/Scripts/activate
pip install -r ./load-test/requirements.txt

Create Python environment and install requirements (Linux)

python -m 'venv' .venv
source .venv/bin/activate
pip install -r ./load-test/requirements.txt

Before running the load test, we should restart the containers. (Why?)

docker compose down -v graalvm-demo-book-jvm graalvm-demo-book-native prometheus
docker compose up -d prometheus

Wait some seconds and then start the book apps:

docker compose up -d graalvm-demo-book-jvm graalvm-demo-book-native

Hint: On Linux, if you have a CPU with performance and efficience cores, you can pin the containers to a specific CPU-set in the compose.yaml file. For example if you have a Intel CPU with 4 performance cores and you want to ensure that the load test runs on them, set cpuset to "0-7".

graalvm-demo-book-jvm:
    container_name: graalvm-demo-book-jvm
    image: graalvm-demo-book-jvm:0.1.0
    cpuset: "0-7"

Run load-test

python ./load-test/send_requests.py --urls "jvm,native" --runs 150

Evaluate the load-test

python ./load-test/requests_eval.py

Which differences do you see? During and after the load-test?

For which kind of workload would you use which variant?

4.3 Further Discussion

• JVM vs. GraalVM native: advantages and disadvantages
• In relation to Cloud Computing?

Additional Topics:

• JVM Optimizatios: Jlink, CDS, AOT
• GraalVM AOT, JIT