quarkus: High memory utilization using Quarkus Native

@JBailes and company this is also why I have petitioned Baeldung to take down their article or make it accurate as its leading to misinformation out in the space about Spring Boot vs Quarkus. I have used both heavily and I dread working in Spring Boot and I enjoy every second of working in Quarkus. 😄

I cannot reproduce the behaviour. I used this sample-application of mine (github.com) with the following modifications:

• limit the container to 30M 35M of memory (limit and reservation)
• build the application natively:
  • change image of animal-service to Containerfile.native
  • Build the application native: ./mvnw --define native --define quarkus.native.container-runtime=podman clean package
• deploy the database: docker-compose --file local-deployment/docker-compose up -d postgres
• create db-schema: ./mvnw flyway:migrate
• deploy the application: docker-compose --file local-deployment/docker-compose up -d --build
• in a separate terminal, start watching the container stats: podman stats animal-service --interval 1
• start wrk tests: wrk/run.sh 1 4 2m
• let the tests finish, observe the container stats during test execution

condense script to do all deployment-steps above in one go:

# Build the application:              
./mvnw clean package
# Stop previous service (if any):
docker-compose --file local-deployment/docker-compose.yml down
# Start the database service:
docker-compose --file local-deployment/docker-compose.yml up -d postgres
# Create Database schema:
./mvnw flyway:migrate
# Start the application
docker-compose --file local-deployment/docker-compose.yml up -d --build
# Run wrk tests:
wrk/run.sh 1 4 2m

Throughput on my machine was around 4.9k requests/second. podman stats... reports a consumption of ~25 MB of memory. When I took a look in the container itself (echo $(podman exec animal-service '/bin/sh' '-c' 'cat /proc/1/status') | sed -e 's|.*\(..RSS: [[:digit:]]\+ ..\).*|\1|''), I saw that the output (28141 kB) is in line with what podman stats reported. Just for good measure, I ran the wrk-tests for 10 minutes, to verify stability¹.

I experimented a little bit more with memory limits. Going to 30M results in:

• the cpu having a load of ~75%, and
• almost no load on the database container, i.e. almost nothing comes through.
• This was also reflected in the throughput, which was a miserable 10 requests/s

There is soem performance to be gained when we increase memory to 40M. There is on more performance gain with more than 40M memory.

EDIT I changed the application to include a timer job. It seems that this pushed the application “over the barrier”, so it now needs 35M of RAM intead of 30M. I adjusted the text accordingly.

<footnotesize>¹This is layman’s was to verify. A real verification would run much longer.</footnotesize>

@melloware I think you’ve hit the nail on the head, that explains the results I was getting as I see the same behavior. I also suspect Quarkus caches much more aggressively then Spring Boot, as I am seeing Quarkus use substantially (~30%) less CPU then Spring. This evening I’m planning on playing around more with the memory settings within Quarkus as well as limiting memory as I have a suspicion allowing Quarkus unlimited RAM may have it consuming more RAM then expected.