Hello World Example
Here is a tiny Java HTTP service used by the JVM memory in Kubernetes slides. The demo is intentionally small so the commands focus on the container, cgroup, heap, and native-memory signals rather than application code.
- HelloWorld.java: Simple single Java file that is a HTTP server
- Dockerfile: Dockerfile for the application
- k8s-deployment.yaml: Kubernetes deployment manifest
The Kubernetes manifest enables a few JVM diagnostics through
JAVA_TOOL_OPTIONS:
env:
- name: JAVA_TOOL_OPTIONS
value: >
-Xlog:os+container=debug
-XX:NativeMemoryTracking=summary
-XX:+UseG1GC
-XX:+FlightRecorder
-XX:StartFlightRecording=filename=/tmp/startup.jfr,dumponexit=true
It also exposes JMX on port 9091 for JDK Mission Control.
Build the Image
Run these commands from this directory:
docker build -t jvmperf.net/java-hello-world:v1 .
If you are using a local Kubernetes runtime such as minikube, kind, or Docker Desktop, make sure the image is available to that cluster before deploying.
Deploy to Kubernetes
Apply the deployment and service:
kubectl apply -f k8s-deployment.yaml
Wait for the pod to become ready:
kubectl rollout status deployment/java-hello-world
kubectl get pods -l app=java-hello-world
For the commands below, capture the pod name once:
POD=$(kubectl get pod -l app=java-hello-world -o jsonpath='{.items[0].metadata.name}')
echo "$POD"
Verify the App
The service is a NodePort, so Kubernetes assigns a local node port for the
HTTP endpoint:
kubectl get svc java-hello-world-service
Use the returned NODE-PORT value with localhost:
NODE_PORT=$(kubectl get svc java-hello-world-service -o jsonpath='{.spec.ports[0].nodePort}')
curl "http://localhost:${NODE_PORT}/"
Expected response:
Hello, World from Kubernetes!
Check the JVM’s Container View
Start with -XshowSettings:system. This is the quick check that the JVM sees the
pod’s cgroup limits rather than sizing itself from the whole node:
kubectl exec "$POD" -- java -XshowSettings:system -version
Look for output shaped like this:
Operating System Metrics:
Provider: cgroupv2
Effective CPU Count: 1
CPU Quota: 100000us
CPU Period: 100000us
Memory Limit: 1.00G
Memory & Swap Limit: 1.00G
The important fields for the slides are:
| Field | Why it matters |
|---|---|
Provider |
Shows whether the JVM is reading cgroup data. |
Effective CPU Count |
Influences GC threads, JIT threads, and app parallelism. |
Memory Limit |
The process budget for heap plus native memory. |
CPU Quota / CPU Period |
Should line up with the deployment CPU limit. |
Inspect Native Memory
Native Memory Tracking is enabled in the manifest with
-XX:NativeMemoryTracking=summary, so you can ask the running JVM for its native
memory map:
kubectl exec "$POD" -- jcmd 1 VM.native_memory summary
The top of the output gives total reserved and committed memory:
Native Memory Tracking:
Total: reserved=1677076KB, committed=43616KB
malloc: 3328KB #7179, peak=3201KB #7181
mmap: reserved=1673748KB, committed=40288KB
Then NMT breaks memory into JVM categories:
- Java Heap (reserved=262144KB, committed=16384KB)
- Class (reserved=1048664KB, committed=280KB)
- Thread (reserved=30653KB, committed=1141KB)
- Code (reserved=249735KB, committed=7735KB)
- GC (reserved=860KB, committed=64KB)
Use this breakdown to explain where memory outside the Java heap can come from: thread stacks, class metadata, code cache, GC structures, direct buffers, native libraries, and JVM internals.
Compare Heap and NMT
GC.heap_info explains the Java heap from the collector’s point of view:
kubectl exec "$POD" -- jcmd 1 GC.heap_info
Example:
garbage-first heap total reserved 262144K, committed 18432K, used 2925K
For the Java heap, reserved and committed should line up with the
Java Heap row in NMT. heap_info adds used, which is the live heap occupancy
inside the committed heap space.
Inspect VM and Cgroup Details
VM.info gives a broader diagnostic snapshot, including the JVM’s cgroup view:
kubectl exec "$POD" -- jcmd 1 VM.info
Useful segment:
container (cgroup) information:
container_type: cgroupv2
active_processor_count: 1
cpu_quota: 100000
cpu_period: 100000
memory_limit_in_bytes: 1048576 k
memory_and_swap_limit_in_bytes: 1048576 k
memory_usage_in_bytes: 118316 k
rss_usage_in_bytes: 116428 k
cache_usage_in_bytes: 64 k
current number of tasks: 29
Capture this beside the pod spec, heap settings, and NMT output when you want to explain why a container is close to its limit.
Record Heap Behavior with JFR
Java Flight Recorder is useful during load tests because it keeps the heap story tied to time. Use it to see heap usage, allocation pressure, and GC behavior across the same window where you are collecting NMT and cgroup data.
The manifest starts a recording automatically:
-XX:+FlightRecorder
-XX:StartFlightRecording=filename=/tmp/startup.jfr,dumponexit=true
For a specific test window, start and stop a named recording with jcmd:
kubectl exec "$POD" -- jcmd 1 JFR.start name=load-test filename=/tmp/load-test.jfr settings=profile
# run representative load
kubectl exec "$POD" -- jcmd 1 JFR.dump name=load-test filename=/tmp/load-test.jfr
kubectl exec "$POD" -- jcmd 1 JFR.stop name=load-test
Copy the recording locally:
kubectl cp "${POD}:/tmp/load-test.jfr" ./load-test.jfr
Open load-test.jfr in JDK Mission Control. For heap-focused analysis, start
with heap usage, allocation, GC pause, and GC heap summary views.
Use JDK Mission Control
The manifest opens JMX on container port 9091. Port-forward it to your machine:
kubectl port-forward "$POD" 9091:9091
Then connect JDK Mission Control to:
localhost:9091
This is useful when you want to show the memory story over time rather than only single command snapshots.
Cleanup
Remove the demo resources when you are done:
kubectl delete -f k8s-deployment.yaml
References
java.lang.management.OperatingSystemMXBean: https://docs.oracle.com/en/java/javase/11/docs/api/java.management/java/lang/management/OperatingSystemMXBean.htmlcom.sun.management.OperatingSystemMXBean: https://docs.oracle.com/en/java/javase/25/docs/api/jdk.management/com/sun/management/OperatingSystemMXBean.html- Native Memory Tracking: https://docs.oracle.com/en/java/javase/11/vm/native-memory-tracking.html