Java Interview Questions

Home / All Answers / Memory and Garbage Collection

Question 24 · Section 3

How to get a heap dump?

4. JFR for large Heaps 5. Security — mask PII 6. Be careful with -all=false (Full GC!)

English Memory and Garbage Collection Source Markdown
Language versions: English Russian Ukrainian

Junior Level

Ways to get a Heap Dump:

# 1. Automatically on OOME (mandatory!)
java -XX:+HeapDumpOnOutOfMemoryError -jar app.jar

# 2. Manually via jcmd
jcmd <PID> GC.heap_dump dump.hprof

# 3. Via VisualVM (GUI)
# → Right-click on process → Heap Dump

Where to find PID:

jps  # List of Java processes
# or
ps aux | grep java

Middle Level

# Modern standard
jcmd <PID> GC.heap_dump /tmp/dump.hprof

# With compression (Java 11+)
jcmd <PID> GC.heap_dump -gz=1 /tmp/dump.hprof.gz
# → 32 GB → 6-10 GB

# Live objects only (Full GC!)
jcmd <PID> GC.heap_dump -all=false /tmp/live.hprof

Kubernetes

# 1. Enter pod
kubectl exec -it <pod> -- bash

# 2. Capture dump
# In containers, Java may be PID 1 (without init process) or have
# separate PID (with tini/dumb-init or Java 11+ container awareness).
# Check via `jps`.
jcmd 1 GC.heap_dump -gz=1 /tmp/dump.hprof.gz

# 3. Download
kubectl cp <pod>:/tmp/dump.hprof.gz ./dump.hprof.gz

# 4. Cleanup
rm /tmp/dump.hprof.gz

Programmatic Method

// ⚠️ com.sun.management.* — internal JDK API (HotSpot-specific).
// Not guaranteed on non-HotSpot JVM (OpenJ9, GraalVM Native Image).
import com.sun.management.HotSpotDiagnosticMXBean;

public void dumpHeap(String path, boolean live) {
    HotSpotDiagnosticMXBean bean = ManagementFactory
        .newPlatformMXBeanProxy(
            ManagementFactory.getPlatformMBeanServer(),
            "com.sun.management:type=HotSpotDiagnostic",
            HotSpotDiagnosticMXBean.class);
    bean.dumpHeap(path, live);
}

Senior Level

Production Dump Dangers

STW pause → Liveness Probe fail → pod killed!

Solution:
  1. Increase probe timeouts
  2. Use JFR instead of dump
  3. Capture dump on replica

PII and Security

Dump contains:
  → Passwords, tokens, sessions
  → Personal data

→ Transmit only over secure channels
→ RedactHprof for string masking
→ Restrict file access

Arthas

# Alternative to jcmd
curl -L https://arthas.aliyun.com/arthas-boot.jar -o arthas-boot.jar
java -jar arthas-boot.jar

[arthas]$ heapdump /tmp/dump.hprof
→ Finds PID automatically
→ Convenient CLI

Best Practices

  1. jcmd — official method
  2. -gz=1 (Java 11+) — compression
  3. Kubernetes → external volume
  4. JFR for large Heaps
  5. Security — mask PII
  6. Be careful with -all=false (Full GC!)

Senior Summary

  • jcmd > jmap (modern standard)
  • Gzip (Java 11+) = savings
  • Kubernetes → external volume
  • JFR > Dump for > 64 GB
  • PII = dump contains secrets
  • Arthas = convenient alternative

Interview Cheat Sheet

Must know:

  • jcmd — official method: jcmd <PID> GC.heap_dump /tmp/dump.hprof; compression: -gz=1 (Java 11+)
  • Automatic: -XX:+HeapDumpOnOutOfMemoryError -XX:HeapDumpPath=/var/log/dumps/ — mandatory in production
  • Kubernetes: enter pod → jcmd 1 GC.heap_dump -gz=1 /tmp/dump.hprof.gz → kubectl cp → cleanup
  • -all=false (live): live objects only; requires Full GC before dump → additional STW pause
  • PII and Security: dump contains passwords, tokens, sessions; transmit over secure channels; RedactHprof for masking
  • Production dangers: STW → Liveness Probe fail → pod killed; solutions: increase timeout, use JFR, capture on replica
  • Programmatic method: HotSpotDiagnosticMXBean.dumpHeap() — HotSpot-specific, doesn’t work on OpenJ9/GraalVM Native Image

Common follow-up questions:

  • Why is -gz=1 important? — On-the-fly compression: 32 GB → 6-10 GB; saves disk space and transfer time
  • What if Java is PID 1 in container? — In containers, Java may be PID 1 (without init process); jcmd 1 works; check via jps
  • Why is programmatic method not cross-platform? — com.sun.management.HotSpotDiagnosticMXBean — internal JDK API; HotSpot only
  • When is Arthas better than jcmd? — Arthas finds PID automatically, convenient interactive CLI; good when jcmd is unavailable

Red flags (DO NOT say):

  • “jmap is the best method” — jmap is deprecated; use jcmd
  • “Live dump is faster and safer” — Live dump requires Full GC → additional STW pause
  • “Dump can be safely stored in Kubernetes ephemeral storage” — ephemeral storage fills with dump → pod won’t restart

Related topics:

  • [[23. What is a heap dump]]
  • [[22. What tools help analyze memory]]
  • [[21. What is a memory leak and how to detect it]]
  • [[19. What happens on OutOfMemoryError]]
  • [[18. What are -Xms and -Xmx parameters]]