Pod OOM - Xmx not beeing respected

Question

I am starting a Java app as follows

JAVA_OPTS=-Xmx7g -Xms512m -Xss1m -XX:+UseCompressedOops -XX:-OmitStackTraceInFastThrow

The pod I am using for this, has

        resources:
          limits:
            cpu: 4096m
            memory: 9Gi
          requests:
            cpu: 1024m
            memory: 512Mi

However, I see once and then, pods being killed with

"Memory cgroup out of memory: Kill process 3284727 (java) score 1982 or sacrifice child
Killed process 3284727 (java) total-vm:15318320kB, anon-rss:9380388kB, file-rss:20180kB, shmem-rss:0kB"

Why does this take place?

How come the memory usage surpasses 9G given that I set Xmx=7G

https://srvaroa.github.io/jvm/kubernetes/memory/docker/oomkiller/2019/05/29/k8s-and-java.html — Yonah Dissen, Sep 18 '19 at 12:00
Maybe a duplicate of https://stackoverflow.com/questions/53451103/java-using-much-more-memory-than-heap-size-or-size-correctly-docker-memory-limi — derkoe, Sep 18 '19 at 12:11

score 1 · Answer 1 · answered Sep 18 '19 at 12:04

With -Xmx you only specify the Java heap size - there is a lot of other memory that the JVM uses (like stack, native memory for the JVM, direct buffers, etc.).

In our experience the correct size for total usage of the JVM is 1.5 to 2 times the heap size but this depends heavily on your use case (for example some applications using direct buffers may have 32GB of RAM using only 1GB of heap).

So run your app with more limits, check the actual usage and then define that + 5% as a container limit.

You should also adapt your request to at least 1Gi with -Xms512m.

Pod OOM - Xmx not beeing respected

1 Answers1