Does it work with the JVM?

[danielepolencic]

This is just a question.

Does memory pressure work with the JVM where the memory is allocated upfront?

[MathieuCesbron]

Hello, thanks for the question.

I still have to check, I will update when I am sure of the answer.

[Terraris]

I have the same question :)

[MathieuCesbron]

It looks like it works with the JVM, here is the test I made:

1. Build a dummy java container with a minimum memory limit of 80m

FROM openjdk:17-jdk-alpine

WORKDIR /app

COPY Main.java /app/Main.java

RUN javac Main.java

CMD ["java", "-Xms80m", "Main"]

2. Apply the container in a Pod with Kondense.
3. Check that Kondense will still resize more than 80m

We see that Kondense resizes it to 47Mib, below the 80M limit put in the Dockerfile.

[danielepolencic]

I'm not a Java expert, but I suspect this will not work. Since I filed the original issue, I did some research and garbage-collected languages expect the memory to be available to allocate objects on the heap. If the heap is "small", that leads to more frequent garbage collection, impacting performance.
Kondense can see memory utilization until that point, but I suspect the JVM could try to allocate more objects to the heap without triggering GC (because the heap is not full yet).
I guess it does not help that the Java app does nothing 😅
Again, I'm happy to be proven wrong, but I suspect this holds true for all garbage collected languages.