What is Java Flight Recorder?

• JFR (Java Flight Recorder) is a profiling and diagnostics framework built into the JVM since Java 11 (free).
• Low overhead: approximately 1-2% performance impact, designed for always-on production monitoring.
• Records: GC pauses, memory allocations, thread contention, CPU hotspots, I/O latency, class loading, and exceptions.
• Start recording: command-line flag -XX:StartFlightRecording=duration=60s,filename=app.jfr or at runtime with jcmd JFR.start.
• Analyze: open the .jfr file in JDK Mission Control (JMC) or use the jfr command-line tool.
• Custom events: extend jdk.jfr.Event to record domain-specific events like settlement batch timing.

What does JFR record?

JFR continuously records JVM events into a compact binary .jfr file. The events include:

• GC events: pause times, heap sizes before and after, GC algorithm details.
• Memory allocation: which threads allocated the most memory, allocation rates.
• Thread contention: lock wait times, thread states, blocked threads.
• CPU profiling: method sampling to identify hot code paths.
• I/O latency: file and socket read/write durations.
• Exceptions: every exception thrown, even if caught.

How do you start a recording?

# Start with the application
java -XX:StartFlightRecording=duration=60s,filename=app.jfr \
     -jar banking-service.jar

# Start on a running process
jcmd <pid> JFR.start duration=60s filename=settlement.jfr

# Stop a running recording
jcmd <pid> JFR.stop name=1

How do you create custom application events?

package com.bank.monitoring;

import jdk.jfr.*;

@Name("com.bank.SettlementBatch")
@Label("Settlement Batch")
@Description("Records settlement batch execution")
public class SettlementEvent extends Event {
    @Label("Batch ID")
    String batchId;

    @Label("Transaction Count")
    int transactionCount;

    @Label("Total Amount")
    long totalAmountPaise;
}

// Usage:
SettlementEvent event = new SettlementEvent();
event.batchId = "BATCH-20240115";
event.transactionCount = 1500;
event.totalAmountPaise = 75000000L;
event.begin();
// ... execute batch ...
event.commit();  // records duration automatically

How do you analyze recordings?

Use JDK Mission Control (JMC) for visual analysis or the jfr command-line tool.

# Print summary of events
jfr summary app.jfr

# Print specific event type
jfr print --events jdk.GarbageCollection app.jfr

# Print custom events
jfr print --events com.bank.SettlementBatch app.jfr

How is JFR different from traditional profiling?

| Feature | JFR | Traditional Profiler |
|---------|-----|---------------------|
| Overhead | ~1-2% | 5-50% |
| Production safe | Yes | Usually not |
| Always-on | Designed for it | Attach-on-demand |
| Built-in | Part of JDK | External tool |
| Custom events | Yes (extend Event) | Varies |

What about JFR Streaming API (Java 14+)?

JFR data can be streamed in real-time for continuous monitoring dashboards.

import jdk.jfr.consumer.RecordingStream;

try (var stream = new RecordingStream()) {
    stream.enable("jdk.GarbageCollection").withThreshold(java.time.Duration.ofMillis(100));
    stream.onEvent("jdk.GarbageCollection", event -> {
        System.out.println("GC pause: " + event.getDuration().toMillis() + "ms");
    });
    stream.startAsync(); // runs in background
}

What are common mistakes?

// WRONG: leaving JFR recordings running indefinitely without a duration limit
java -XX:StartFlightRecording=filename=app.jfr -jar banking-service.jar
// The .jfr file grows unbounded!

// CORRECT: set a duration or maxSize
java -XX:StartFlightRecording=duration=60s,filename=app.jfr -jar banking-service.jar
java -XX:StartFlightRecording=maxsize=500m,filename=app.jfr -jar banking-service.jar

What should a senior engineer know?

• JFR complements metrics and traces: metrics tell you SOMETHING is slow, JFR tells you WHY at the JVM/allocation/lock level.
• The programmatic API (jdk.jfr.Recording) lets you start and stop recordings in integration tests to capture performance baselines.
• JFR data can be streamed in real-time with the JFR Streaming API (Java 14+) for continuous monitoring dashboards.
• In banking, JFR recording of GC pauses during peak trading hours helps tune GC settings for latency-sensitive services.