序

本文主要研究一下jvm的mapped buffer的统计

示例

private void writeDirectBuffer() {    // 分配一个256MB的直接缓冲区    ByteBuffer buffer = ByteBuffer.allocateDirect(256 * 1024 * 1024);    // 填充数据    Random random = new Random();    while (buffer.remaining() >= 4) {        buffer.putInt(random.nextInt());    }    System.out.println("Allocated direct buffer with capacity " + buffer.capacity());}private void writeMappedBuffer() throws IOException {    RandomAccessFile file = new RandomAccessFile("/tmp/test.txt", "rw");    FileChannel channel = file.getChannel();    MappedByteBuffer buffer = channel.map(FileChannel.MapMode.READ_WRITE, 0, 128 * 1024 * 1024); // 映射128MB的空间    // 随机写入数据    for (int i = 0; i < buffer.capacity(); i++) {        buffer.put((byte) i);    }}

{    "name": "jvm.buffer.memory.used",    "description": "An estimate of the memory that the Java virtual machine is using for this buffer pool",    "baseUnit": "bytes",    "measurements": [        {            "statistic": "VALUE",            "value": 402685952        }    ],    "availableTags": [        {            "tag": "id",            "values": [                "direct",                "mapped"            ]        }    ]}

jvm.buffer.memory.used分了direct和mapped两大类，一共用了384MB

mapped

{    "name": "jvm.buffer.memory.used",    "description": "An estimate of the memory that the Java virtual machine is using for this buffer pool",    "baseUnit": "bytes",    "measurements": [        {            "statistic": "VALUE",            "value": 134217728        }    ],    "availableTags": []}

可以看到这里mapped用了128MB

direct

{    "name": "jvm.buffer.memory.used",    "description": "An estimate of the memory that the Java virtual machine is using for this buffer pool",    "baseUnit": "bytes",    "measurements": [        {            "statistic": "VALUE",            "value": 268500992        }    ],    "availableTags": []}

可以看到这里direct用了256MB

Native Memory Tracking开启

java -XX:+UnlockDiagnosticVMOptions -XX:NativeMemoryTracking=summary -jar target/app.jar

jcmd 2315 VM.native_memory summary scale=MB2315:Native Memory Tracking:Total: reserved=5882MB, committed=1100MB-                 Java Heap (reserved=4096MB, committed=577MB)                            (mmap: reserved=4096MB, committed=577MB)-                     Class (reserved=1066MB, committed=46MB)                            (classes #7035)                            (malloc=10MB #10793)                            (mmap: reserved=1056MB, committed=37MB)-                    Thread (reserved=36MB, committed=36MB)                            (thread #37)                            (stack: reserved=36MB, committed=36MB)-                      Code (reserved=246MB, committed=16MB)                            (malloc=2MB #3971)                            (mmap: reserved=244MB, committed=13MB)-                        GC (reserved=160MB, committed=148MB)                            (malloc=10MB #219)                            (mmap: reserved=150MB, committed=138MB)-                  Internal (reserved=266MB, committed=266MB)                            (malloc=266MB #10067)-                    Symbol (reserved=9MB, committed=9MB)                            (malloc=8MB #74334)                            (arena=2MB #1)-    Native Memory Tracking (reserved=2MB, committed=2MB)                            (tracking overhead=2MB)

NTM的Internal包含了direct buffer的统计，不包含mapped buffer的占用，而BufferPoolMXBean包含了direct和mapped。

另外通过调用System.loadLibrary()加载的共享库分配的内存NTM没办法追踪到，MXBean貌似也没办法统计到。

小结聊聊jvm的direct buffer统计Memory Types in JVMNative Memory Tracking in JVMHow to monitor JVM native memory usage by enabling native memory trackingTroubleshooting Problems With Native (Off-Heap) Memory in Java ApplicationsCan a Java Application Use More Memory Than the Heap Size?java-non-heap-memory-analyzes