Why does my Java process consumes twice memory inside a docker container vs host

Question

I faced an interesting problem trying to analyze a memory consumption in my Java application running on docker container vs host machine.

1. The Java app is web app on the Jetty server 9.4.9
2. Java version : 1.8
3. Host : MAC
4. Docker images: jetty:9.4-jre8
5. The docker daemon is 18.03.1-ce version.

On the host I'm using Yourkit tool to analyze a memory consumption.

For docker container docker stats <docker id/name>

What I'm getting is that on MAC yourkit shows me 50M Non-heap size + ~40M heap size, in total ~ 100M

Whereas, when I deploy and run the same war on a container, the stats shows me 200M

CONTAINER ID        NAME                CPU %               MEM USAGE / LIMIT     MEM %               NET I/O             BLOCK I/O           PIDS
879fb113ca8d        jetty-app           0.19%               214.6MiB / 1.952GiB   10.74%              1.49MB / 88.9kB     31.7MB / 6.42MB     29

Can anyone shed some light on this phenomenon?

Assuming that stats provides wrong results, I tried to limit the memory on a container using --memory flag doesn't help much, I'm getting OOM.

Thanks in advance

Answer 1

You might want to try and measure again, with an openJDK 8u212 or more (April, 16th 2019). (no Oracle JDK, since their license has changed )

See " Docker support in Java 8 — finally! " from Grzegorz Kocur .
Now:

There is no need to use any hacky workarounds in a docker entrypoint, nor setting Xmx as fixed value anymore.

Docker support was also backported to Java 8.
Let's check the newest openjdk image tagged as 8u212. We'll limit the memory to 1G and use 1 CPU:

 docker run -ti --cpus 1 -m 1G openjdk:8u212-jdk 

You can fine-tune the heap-size with new flags (already present in Java 10+, but now back ported to Java 8), and explained here .

-XX:InitialRAMPercentage
-XX:MaxRAMPercentage
-XX:MinRAMPercentage

If for some reason the new JVM behaviour is not desired it can be switched off using - XX:-UseContainerSupport .