Skip to content

fbauzac/quarkus-profiling-workshop

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

42 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

hello-world-app Project

This project uses Quarkus, the Supersonic Subatomic Java Framework.

If you want to learn more about Quarkus, please visit its website: https://quarkus.io/ .

Packaging and running the application

The application can be packaged using:

./mvnw package

It produces the quarkus-run.jar file in the target/quarkus-app/ directory. Be aware that it’s not an über-jar as the dependencies are copied into the target/quarkus-app/lib/ directory.

The application is now runnable using java -jar target/quarkus-app/quarkus-run.jar.

If you want to build an über-jar, execute the following command:

./mvnw package -Dquarkus.package.type=uber-jar

The application, packaged as an über-jar, is now runnable using java -jar target/*-runner.jar.

Creating a native executable

You can create a native executable using:

./mvnw package -Pnative

Or, if you don't have GraalVM installed, you can run the native executable build in a container using:

./mvnw package -Pnative -Dquarkus.native.container-build=true

You can then execute your native executable with: ./target/hello-world-app-1.0.0-SNAPSHOT-runner

If you want to learn more about building native executables, please consult https://quarkus.io/guides/maven-tooling.

Related Guides

  • Mutiny (guide): Write reactive applications with the modern Reactive Programming library Mutiny
  • RESTEasy Reactive (guide): A JAX-RS implementation utilizing build time processing and Vert.x. This extension is not compatible with the quarkus-resteasy extension, or any of the extensions that depend on it.

Provided Code

RESTEasy Reactive

Easily start your Reactive RESTful Web Services

Related guide section...

How to load generate HTTP workload

How to install the chosen load generator: https://hyperfoil.io/quickstart/quickstart1.html

Why Hyperfoil? https://www.slideshare.net/InfoQ/how-not-to-measure-latency-60111840

Preparation Steps

Download and unzip the generator:

wget https://github.com/Hyperfoil/Hyperfoil/releases/download/hyperfoil-all-0.24/hyperfoil-0.24.zip \
    && unzip hyperfoil-0.24.zip \
    && cd hyperfoil-0.24

From within the hyperfoil's /bin folder and assuming the Quarkus hello world endpoint to be up and running:

[hyperfoil@in-vm]$ wrk -t 1 -c 10 -d 10s http://localhost:8080/hello

While the benchmark is completed, it will output something like:

Running 10s test @ http://localhost:8080/hello
  1 threads and 10 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency   192.64μs  317.09μs  66.06ms   99.77%
    Req/Sec   45764.73  17108.08  58857.00     90.91
  503412 requests in 10.002s,  70.57MB read
Requests/sec: 50331.13
Transfer/sec:   7.06MB

Is it possible to start a CLI interactive sessionn with hyperfoil and both report in html or perform quick diff between wrk runs.

Which profiler? Async-profiler OBVIOUSLY!

Although async-profiler is awesome, but using it can be made simpler with ap-loader.

Download it:

wget https://github.com/jvm-profiling-tools/ap-loader/releases/download/2.9/ap-loader-all.jar

Let's have fun with the various scripts in the /scripts folder of the project!

Cheat Codes

The benchmarking scripts allow to capture more information by using the proper output format.

e.g.

$ ./benchmark.sh -u persons/agesum -r 4000 -e cache-misses -f jfr

the resulting output file should be named as <quarkus pid>_cache_misses.jfr. It then can be opened (if you dare!! ) with JMC 8.x or can be used the built-in jfr2flame converter to produce flamegraphs with more information e.g. line of code to be blamed, if available

$ java -jar ap-loader-all.jar converter jfr2flame <quarkus pid>_cache_misses.jfr --threads --lines out.html

More informations about converter are part of the command's help or can be found on the async profiler homepage.

How to compare benchmarking results?

Hyperfoil allow to easily compare between two runs, but sadly, given that we've used the CLI versions of wrk/wrk2 (load generation CLI commands) we don't have an easy way to find identifier for such runs.

Leveraging on our knowledge of the benchmarking script, it performs 2 runs:

  • a warmup run with all-out throughput
  • a second run with the configured throughput (if any)

This means that for each benchmarking session it produces 2 runs on in Hyperfoil and the ones to be compared must be taken considering the warming up ones too.

Assuming two benchmarks runs already completed:

# run the hyperfoil CLI
$ hyperfoil-0.23/bin/cli.sh
# start a local session on it
[hyperfoil]$ start-local 
Starting controller in default directory (/tmp/hyperfoil)
Controller started, listening on 127.0.0.1:33161
Connecting to the controller...
Connected to 127.0.0.1:33161!
# now on we can interact with the local env
[hyperfoil@in-vm]$ runs
   RUN_ID  BENCHMARK  STARTED                  TERMINATED               DESCRIPTION
!  0000    wrk        2023/01/17 04:53:15.985  2023/01/17 04:53:31.989             
!  0001    wrk        2023/01/17 04:53:35.261  2023/01/17 04:53:51.267             
# that could be a loooooooong list by the end of the workshop :P
#                           ...
!  0067    wrk        2023/01/18 12:37:06.368  2023/01/18 12:37:32.380             
!  0068    wrk2       2023/01/18 12:37:36.843  2023/01/18 12:38:02.852             
!  0069    wrk        2023/01/18 12:39:00.641  2023/01/18 12:39:26.650             
!  006A    wrk2       2023/01/18 12:39:31.091  2023/01/18 12:39:57.099

As said before, given that each benchmark script run perform 2 hyperfoil runs, the last 2 warmed up runs are 0068 and 006A.

Let's see what compare says:

[hyperfoil@in-vm]$ compare 0068 006A
Comparing runs 0068 and 006A
PHASE        METRIC   REQUESTS    MEAN                 p50                p90                   p99                   p99.9                p99.99
calibration  request  +8(+0.03%)   -5.85 ms(-524.95%)      +0 ns(+0.00%)  -36.20 ms(-7214.30%)   -28.57 ms(-153.52%)  -28.57 ms(-153.52%)  -28.57 ms(-153.52%)
test         request  -8(-0.01%)   -1.45 ms(-257.71%)      +0 ns(+0.00%)    -1.01 ms(-200.41%)  -29.51 ms(-5882.05%)  -16.52 ms(-122.33%)  -16.52 ms(-122.33%)

Is it clear that under the same load (whatever it is), the 2 perform very differently, with 006A getting 1 ms p90 lower latencies, that's an improvement of -200.41% (!!!).

To manually study the dynamics of each run, is it possible to produce an html report for each with:

[hyperfoil@in-vm]$ report 0068 --destination=linkedlist.html

You can then open the html file an check the rate progress, latency distributions and errors.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • HTML 94.1%
  • Java 4.0%
  • Shell 1.9%