JVM性能调优监控工具jps、jstack、jmap、jhat、jstat、hprof使用详解

JDK本身提供了很多方便的JVM性能调优监控工具，除了集成式的VisualVM和jConsole外，还有jps、jstack、jmap、jhat、jstat、hprof等小巧的工具，本博客希望能起抛砖引玉之用，让大家能开始对JVM性能调优的常用工具有所了解。

    现实企业级Java开发中，有时候我们会碰到下面这些问题：

• OutOfMemoryError，内存不足

• 内存泄露

• 线程死锁

• 锁争用（Lock Contention）

• Java进程消耗CPU过高

• ......

    这些问题在日常开发中可能被很多人忽视（比如有的人遇到上面的问题只是重启服务器或者调大内存，而不会深究问题根源），但能够理解并解决这些问题是Java程序员进阶的必备要求。本文将对一些常用的JVM性能调优监控工具进行介绍，希望能起抛砖引玉之用。本文参考了网上很多资料，难以一一列举，在此对这些资料的作者表示感谢！关于JVM性能调优相关的资料，请参考文末。

 

A、 jps(Java Virtual Machine Process Status Tool)      

    jps主要用来输出JVM中运行的进程状态信息。语法格式如下：

1	<code class= "hljs css" ><span class= "hljs-selector-tag" >jps <span class= "hljs-selector-attr" >[options] <span class= "hljs-selector-attr" >[hostid]< /span >< /span >< /span >< /code >

    如果不指定hostid就默认为当前主机或服务器。

    命令行参数选项说明如下：

1 2 3 4

<code class= "hljs haml" >-<span class= "ruby" >q 不输出类名、Jar名和传入main方法的参数 -<span class= "ruby" >m 输出传入main方法的参数 -<span class= "ruby" >l 输出main类或Jar的全限名 -<span class= "ruby" > v  输出传入JVM的参数< /span >< /span >< /span >< /span >< /code >

   比如下面：

1 2 3 4 5 6 7 8

<code class= "hljs less" >root<span class= "hljs-variable" >@ubuntu:/ # jps -m -l <span class= "hljs-number" >2458 org.artifactory.standalone.main.Main  /usr/local/artifactory- <span class= "hljs-number" >2.2.<span class= "hljs-number" >5 /etc/jetty .xml <span class= "hljs-number" >29920 com.sun.tools.hat.Main -port <span class= "hljs-number" >9998  /tmp/dump .dat <span class= "hljs-number" >3149 org.apache.catalina.startup.Bootstrap start <span class= "hljs-number" >30972 sun.tools.jps.Jps -m -l <span class= "hljs-number" >8247 org.apache.catalina.startup.Bootstrap start <span class= "hljs-number" >25687 com.sun.tools.hat.Main -port <span class= "hljs-number" >9999 dump.dat <span class= "hljs-number" >21711 mrf-center.jar< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /code >

 

B、 jstack

    jstack主要用来查看某个Java进程内的线程堆栈信息。语法格式如下：

1 2 3

<code class= "hljs css" ><span class= "hljs-selector-tag" >jstack <span class= "hljs-selector-attr" >[option] <span class= "hljs-selector-tag" >pid <span class= "hljs-selector-tag" >jstack <span class= "hljs-selector-attr" >[option] <span class= "hljs-selector-tag" >executable <span class= "hljs-selector-tag" >core <span class= "hljs-selector-tag" >jstack <span class= "hljs-selector-attr" >[option] <span class= "hljs-selector-attr" >[server- id @]<span class= "hljs-selector-tag" >remote- hostname -or-ip< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /code >

    命令行参数选项说明如下：

1 2	<code class= "hljs haml" >-<span class= "ruby" >l long listings，会打印出额外的锁信息，在发生死锁时可以用jstack -l pid来观察锁持有情况 -<span class= "ruby" >m mixed mode，不仅会输出Java堆栈信息，还会输出C /C ++堆栈信息（比如Native方法）< /span >< /span >< /code >

    jstack可以定位到线程堆栈，根据堆栈信息我们可以定位到具体代码，所以它在JVM性能调优中使用得非常多。下面我们来一个实例找出某个Java进程中最耗费CPU的Java线程并定位堆栈信息，用到的命令有ps、top、printf、jstack、grep。

    第一步先找出Java进程ID，我部署在服务器上的Java应用名称为mrf-center：

1 2

<code class= "hljs less" >root<span class= "hljs-variable" >@ubuntu:/ # ps -ef | grep mrf-center | grep -v grep root     <span class= "hljs-number" >21711     <span class= "hljs-number" >1  <span class= "hljs-number" >1 <span class= "hljs-number" >14:<span class= "hljs-number" >47 pts/<span class= "hljs-number" >3    <span class= "hljs-number" >00:<span class= "hljs-number" >02:<span class= "hljs-number" >10 java -jar mrf-center.jar< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /code >

    得到进程ID为21711，第二步找出该进程内最耗费CPU的线程，可以使用ps -Lfp pid或者ps -mp pid -o THREAD, tid, time或者top -Hp pid，我这里用第三个，输出如下：

    TIME列就是各个Java线程耗费的CPU时间，CPU时间最长的是线程ID为21742的线程，用

1	<code class= "hljs perl" ><span class= "hljs-keyword" > printf  <span class= "hljs-string" > "%x\n"  <span class= "hljs-number" >21742< /span >< /span >< /span >< /code >

    得到21742的十六进制值为54ee，下面会用到。    

    OK，下一步终于轮到jstack上场了，它用来输出进程21711的堆栈信息，然后根据线程ID的十六进制值grep，如下：

1 2

<code class= "hljs less" >root<span class= "hljs-variable" >@ubuntu:/ # jstack <span class="hljs-number">21711 | grep <span class="hljs-number">54ee <span class= "hljs-string" > "PollIntervalRetrySchedulerThread"  prio=<span class= "hljs-number" >10 tid=<span class= "hljs-number" >0x00007f950043e000 nid=<span class= "hljs-number" >0x54ee  in  Object.wait() [<span class= "hljs-number" >0x00007f94c6eda000]< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /code >

    可以看到CPU消耗在PollIntervalRetrySchedulerThread这个类的Object.wait()，我找了下我的代码，定位到下面的代码：

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

<code class = "hljs java" ><span class = "hljs-comment" > // Idle wait getLog().info(<span class = "hljs-string" > "Thread ["  + getName() + <span class = "hljs-string" > "] is idle waiting..." ); schedulerThreadState = PollTaskSchedulerThreadState.IdleWaiting; <span class = "hljs-keyword" > long  now = System.currentTimeMillis(); <span class = "hljs-keyword" > long  waitTime = now + getIdleWaitTime(); <span class = "hljs-keyword" > long  timeUntilContinue = waitTime - now; <span class = "hljs-keyword" > synchronized (sigLock) {      <span class = "hljs-keyword" > try  {          <span class = "hljs-keyword" > if (!halted.get()) {              sigLock.wait(timeUntilContinue);          }      }       <span class = "hljs-keyword" > catch  (InterruptedException ignore) {      } }</span></span></span></span></span></span></span></span></span></span></code>

    它是轮询任务的空闲等待代码，上面的sigLock.wait(timeUntilContinue)就对应了前面的Object.wait()。

 

C、 jmap（Memory Map）和jhat（Java Heap Analysis Tool）

    jmap用来查看堆内存使用状况，一般结合jhat使用。

    jmap语法格式如下：

1 2 3

<code class= "hljs css" ><span class= "hljs-selector-tag" >jmap <span class= "hljs-selector-attr" >[option] <span class= "hljs-selector-tag" >pid <span class= "hljs-selector-tag" >jmap <span class= "hljs-selector-attr" >[option] <span class= "hljs-selector-tag" >executable <span class= "hljs-selector-tag" >core <span class= "hljs-selector-tag" >jmap <span class= "hljs-selector-attr" >[option] <span class= "hljs-selector-attr" >[server- id @]<span class= "hljs-selector-tag" >remote- hostname -or-ip< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /code >

    如果运行在64位JVM上，可能需要指定-J-d64命令选项参数。

1	<code class= "hljs tcl" >jmap -permstat <span class= "hljs-keyword" >pid< /span >< /code >

    打印进程的类加载器和类加载器加载的持久代对象信息，输出：类加载器名称、对象是否存活（不可靠）、对象地址、父类加载器、已加载的类大小等信息，如下图：

   使用jmap -heap pid查看进程堆内存使用情况，包括使用的GC算法、堆配置参数和各代中堆内存使用情况。比如下面的例子：

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49

<code class= "hljs less" >root<span class= "hljs-variable" >@ubuntu:/ # jmap -heap <span class="hljs-number">21711 Attaching to process ID <span class= "hljs-number" >21711, please wait... Debugger attached successfully. Server compiler detected. JVM version is <span class= "hljs-number" >20.10-b01   using thread- local  object allocation. Parallel GC with <span class= "hljs-number" >4 thread(s)   Heap <span class= "hljs-attribute" >Configuration:     MinHeapFreeRatio = <span class= "hljs-number" >40     MaxHeapFreeRatio = <span class= "hljs-number" >70     MaxHeapSize      = <span class= "hljs-number" >2067791872 (<span class= "hljs-number" >1972.0MB)     NewSize          = <span class= "hljs-number" >1310720 (<span class= "hljs-number" >1.25MB)     MaxNewSize       = <span class= "hljs-number" >17592186044415 MB     OldSize          = <span class= "hljs-number" >5439488 (<span class= "hljs-number" >5.1875MB)     NewRatio         = <span class= "hljs-number" >2     SurvivorRatio    = <span class= "hljs-number" >8     PermSize         = <span class= "hljs-number" >21757952 (<span class= "hljs-number" >20.75MB)     MaxPermSize      = <span class= "hljs-number" >85983232 (<span class= "hljs-number" >82.0MB)   Heap <span class= "hljs-attribute" >Usage: PS Young Generation Eden <span class= "hljs-attribute" >Space:     capacity = <span class= "hljs-number" >6422528 (<span class= "hljs-number" >6.125MB)     used     = <span class= "hljs-number" >5445552 (<span class= "hljs-number" >5.1932830810546875MB)     free      = <span class= "hljs-number" >976976 (<span class= "hljs-number" >0.9317169189453125MB)     <span class= "hljs-number" >84.78829520089286% used From <span class= "hljs-attribute" >Space:     capacity = <span class= "hljs-number" >131072 (<span class= "hljs-number" >0.125MB)     used     = <span class= "hljs-number" >98304 (<span class= "hljs-number" >0.09375MB)     free      = <span class= "hljs-number" >32768 (<span class= "hljs-number" >0.03125MB)     <span class= "hljs-number" >75.0% used To <span class= "hljs-attribute" >Space:     capacity = <span class= "hljs-number" >131072 (<span class= "hljs-number" >0.125MB)     used     = <span class= "hljs-number" >0 (<span class= "hljs-number" >0.0MB)     free      = <span class= "hljs-number" >131072 (<span class= "hljs-number" >0.125MB)     <span class= "hljs-number" >0.0% used PS Old Generation     capacity = <span class= "hljs-number" >35258368 (<span class= "hljs-number" >33.625MB)     used     = <span class= "hljs-number" >4119544 (<span class= "hljs-number" >3.9287033081054688MB)     free      = <span class= "hljs-number" >31138824 (<span class= "hljs-number" >29.69629669189453MB)     <span class= "hljs-number" >11.683876009235595% used PS Perm Generation     capacity = <span class= "hljs-number" >52428800 (<span class= "hljs-number" >50.0MB)     used     = <span class= "hljs-number" >26075168 (<span class= "hljs-number" >24.867218017578125MB)     free      = <span class= "hljs-number" >26353632 (<span class= "hljs-number" >25.132781982421875MB)     <span class= "hljs-number" >49.73443603515625% used     ....< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /code >

    使用jmap -histo[:live] pid查看堆内存中的对象数目、大小统计直方图，如果带上live则只统计活对象，如下：

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37

<code class= "hljs scss" >root@ubuntu:/ # jmap -histo:live 21711 | more     num      #instances         <span class="hljs-number">#bytes  class name ----------------------------------------------     1:         38445        5597736  <constMethodKlass>     2:         38445        5237288  <methodKlass>     3:          3500        3749504  <constantPoolKlass>     4:         60858        3242600  <symbolKlass>     5:          3500        2715264  <instanceKlassKlass>     6:          2796        2131424  <constantPoolCacheKlass>     7:          5543        1317400  [I     8:         13714        1010768  [C     9:          4752        1003344  [B    10:          1225         639656  <methodDataKlass>    11:         14194         454208  java.lang.String    12:          3809         396136  java.lang.Class    13:          4979         311952  [S    14:          5598         287064  [[I    15:          3028         266464  java.lang.reflect.Method    16:           280         163520  <objArrayKlassKlass>    17:          4355         139360  java.util.HashMap<span class= "hljs-variable" >$Entry    18:          1869         138568  [Ljava.util.HashMap<span class= "hljs-variable" >$Entry;    19:          <span class= "hljs-number" >2443          <span class= "hljs-number" >97720  java.util.LinkedHashMap<span class= "hljs-variable" >$Entry    <span class= "hljs-number" >20:          <span class= "hljs-number" >2072          <span class= "hljs-number" >82880  java.lang.ref.SoftReference    <span class= "hljs-number" >21:          <span class= "hljs-number" >1807          <span class= "hljs-number" >71528  [Ljava.lang.Object;    22:          <span class= "hljs-number" >2206          <span class= "hljs-number" >70592  java.lang.ref.WeakReference    <span class= "hljs-number" >23:           <span class= "hljs-number" >934          <span class= "hljs-number" >52304  java.util.LinkedHashMap    <span class= "hljs-number" >24:           <span class= "hljs-number" >871          <span class= "hljs-number" >48776  java.beans.MethodDescriptor    <span class= "hljs-number" >25:          <span class= "hljs-number" >1442          <span class= "hljs-number" >46144  java.util.concurrent.ConcurrentHashMap<span class= "hljs-variable" >$HashEntry    <span class= "hljs-number" >26:           <span class= "hljs-number" >804          <span class= "hljs-number" >38592  java.util.HashMap    <span class= "hljs-number" >27:           <span class= "hljs-number" >948          <span class= "hljs-number" >37920  java.util.concurrent.ConcurrentHashMap<span class= "hljs-variable" >$Segment    <span class= "hljs-number" >28:          <span class= "hljs-number" >1621          <span class= "hljs-number" >35696  [Ljava.lang.Class;    29:          <span class= "hljs-number" >1313          <span class= "hljs-number" >34880  [Ljava.lang.String;    30:          <span class= "hljs-number" >1396          <span class= "hljs-number" >33504  java.util.LinkedList<span class= "hljs-variable" >$Entry    <span class= "hljs-number" >31:           <span class= "hljs-number" >462          <span class= "hljs-number" >33264  java.lang.reflect.Field    <span class= "hljs-number" >32:          <span class= "hljs-number" >1024          <span class= "hljs-number" >32768  java.util.Hashtable<span class= "hljs-variable" >$Entry    <span class= "hljs-number" >33:           <span class= "hljs-number" >948          <span class= "hljs-number" >31440  [Ljava.util.concurrent.ConcurrentHashMap<span class= "hljs-variable" >$HashEntry;< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /code >

    class name是对象类型，说明如下：

1 2 3 4 5 6 7 8 9

<code class= "hljs java" >B  <span class= "hljs-keyword" >byte C  <span class= "hljs-keyword" >char D  <span class= "hljs-keyword" >double F  <span class= "hljs-keyword" >float I  <span class= "hljs-keyword" >int J  <span class= "hljs-keyword" >long Z  <span class= "hljs-keyword" >boolean [  数组，如[I表示<span class= "hljs-keyword" >int[] [L+类名 其他对象< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /code >

    还有一个很常用的情况是：用jmap把进程内存使用情况dump到文件中，再用jhat分析查看。jmap进行dump命令格式如下：

1	<code class= "hljs tcl" >jmap -dump:<span class= "hljs-keyword" > format =b,<span class= "hljs-keyword" > file =dumpFileName <span class= "hljs-keyword" >pid< /span >< /span >< /span >< /code >

    我一样地对上面进程ID为21711进行Dump：

1 2 3

<code class= "hljs less" >root<span class= "hljs-variable" >@ubuntu:/ # jmap <span class="hljs-attribute">-dump:format=b,file=/tmp/dump.dat <span class="hljs-number">21711      Dumping heap to  /tmp/dump .dat ... Heap dump  file  created< /span >< /span >< /span >< /code >

   dump出来的文件可以用MAT、VisualVM等工具查看，这里用jhat查看：

1 2 3 4 5 6 7 8 9 10

<code class= "hljs less" >root<span class= "hljs-variable" >@ubuntu:/ # jhat -port <span class="hljs-number">9998 /tmp/dump.dat Reading from  /tmp/dump .dat... Dump  file  created Tue Jan <span class= "hljs-number" >28 <span class= "hljs-number" >17:<span class= "hljs-number" >46:<span class= "hljs-number" >14 CST <span class= "hljs-number" >2014 Snapshot  read , resolving... Resolving <span class= "hljs-number" >132207 objects... Chasing references, expect <span class= "hljs-number" >26 dots.......................... Eliminating duplicate references.......................... Snapshot resolved. Started HTTP server on port <span class= "hljs-number" >9998 Server is ready.< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /code >

     注意如果Dump文件太大，可能需要加上-J-Xmx512m这种参数指定最大堆内存，即jhat -J-Xmx512m -port 9998 /tmp/dump.dat。然后就可以在浏览器中输入主机地址:9998查看了：

    上面红线框出来的部分大家可以自己去摸索下，最后一项支持OQL（对象查询语言）。

 

D、jstat（JVM统计监测工具）

    语法格式如下：

1	<code class= "hljs css" ><span class= "hljs-selector-tag" >jstat <span class= "hljs-selector-attr" >[ generalOption | outputOptions vmid [interval[s|ms] <span class= "hljs-selector-attr" >[count]] ]< /span >< /span >< /span >< /code >

    vmid是Java虚拟机ID，在Linux/Unix系统上一般就是进程ID。interval是采样时间间隔。count是采样数目。比如下面输出的是GC信息，采样时间间隔为250ms，采样数为4：

1 2 3 4 5 6

<code class= "hljs less" >root<span class= "hljs-variable" >@ubuntu:/ # jstat -gc <span class="hljs-number">21711 <span class="hljs-number">250 <span class="hljs-number">4   S0C    S1C    S0U    S1U      EC       EU        OC         OU       PC     PU    YGC     YGCT    FGC    FGCT     GCT    <span class= "hljs-number" >192.0  <span class= "hljs-number" >192.0   <span class= "hljs-number" >64.0   <span class= "hljs-number" >0.0    <span class= "hljs-number" >6144.0   <span class= "hljs-number" >1854.9   <span class= "hljs-number" >32000.0     <span class= "hljs-number" >4111.6   <span class= "hljs-number" >55296.0 <span class= "hljs-number" >25472.7    <span class= "hljs-number" >702    <span class= "hljs-number" >0.431   <span class= "hljs-number" >3      <span class= "hljs-number" >0.218    <span class= "hljs-number" >0.649 <span class= "hljs-number" >192.0  <span class= "hljs-number" >192.0   <span class= "hljs-number" >64.0   <span class= "hljs-number" >0.0    <span class= "hljs-number" >6144.0   <span class= "hljs-number" >1972.2   <span class= "hljs-number" >32000.0     <span class= "hljs-number" >4111.6   <span class= "hljs-number" >55296.0 <span class= "hljs-number" >25472.7    <span class= "hljs-number" >702    <span class= "hljs-number" >0.431   <span class= "hljs-number" >3      <span class= "hljs-number" >0.218    <span class= "hljs-number" >0.649 <span class= "hljs-number" >192.0  <span class= "hljs-number" >192.0   <span class= "hljs-number" >64.0   <span class= "hljs-number" >0.0    <span class= "hljs-number" >6144.0   <span class= "hljs-number" >1972.2   <span class= "hljs-number" >32000.0     <span class= "hljs-number" >4111.6   <span class= "hljs-number" >55296.0 <span class= "hljs-number" >25472.7    <span class= "hljs-number" >702    <span class= "hljs-number" >0.431   <span class= "hljs-number" >3      <span class= "hljs-number" >0.218    <span class= "hljs-number" >0.649 <span class= "hljs-number" >192.0  <span class= "hljs-number" >192.0   <span class= "hljs-number" >64.0   <span class= "hljs-number" >0.0    <span class= "hljs-number" >6144.0   <span class= "hljs-number" >2109.7   <span class= "hljs-number" >32000.0     <span class= "hljs-number" >4111.6   <span class= "hljs-number" >55296.0 <span class= "hljs-number" >25472.7    <span class= "hljs-number" >702    <span class= "hljs-number" >0.431   <span class= "hljs-number" >3      <span class= "hljs-number" >0.218    <span class= "hljs-number" >0.649< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /code >

    要明白上面各列的意义，先看JVM堆内存布局：

    可以看出：

1 2	<code class= "hljs gradle" >堆内存 = 年轻代 + 年老代 + 永久代 年轻代 = Eden区 + 两个Survivor区（<span class= "hljs-keyword" >From和To）< /span >< /code >

    现在来解释各列含义：

1 2 3 4 5 6 7

<code class= "hljs" >S0C、S1C、S0U、S1U：Survivor 0 /1 区容量（Capacity）和使用量（Used） EC、EU：Eden区容量和使用量 OC、OU：年老代容量和使用量 PC、PU：永久代容量和使用量 YGC、YGT：年轻代GC次数和GC耗时 FGC、FGCT：Full GC次数和Full GC耗时 GCT：GC总耗时< /code >

 

E、hprof（Heap/CPU Profiling Tool）

    hprof能够展现CPU使用率，统计堆内存使用情况。

    语法格式如下：

1 2 3

<code class= "hljs css" ><span class= "hljs-selector-tag" >java <span class= "hljs-selector-tag" >-agentlib<span class= "hljs-selector-pseudo" >:hprof<span class= "hljs-selector-attr" >[=options] <span class= "hljs-selector-tag" >ToBeProfiledClass <span class= "hljs-selector-tag" >java <span class= "hljs-selector-tag" >-Xrunprof<span class= "hljs-selector-attr" >[:options] <span class= "hljs-selector-tag" >ToBeProfiledClass <span class= "hljs-selector-tag" >javac <span class= "hljs-selector-tag" >-J-agentlib<span class= "hljs-selector-pseudo" >:hprof<span class= "hljs-selector-attr" >[=options] <span class= "hljs-selector-tag" >ToBeProfiledClass< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /code >

    完整的命令选项如下：

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17

<code class= "hljs makefile" >Option Name and Value  Description                    Default ---------------------  -----------                    ------- heap=dump|sites|all    heap profiling                 all cpu=samples| times |old  CPU usage                      off monitor=y|n            monitor contention             n format =a|b             text(txt) or binary output     a file =< file >            write data to  file              java.hprof[.txt] net=<host>:<port>      send data over a socket        off depth=<size>           stack trace depth              4 interval=<ms>          sample interval  in  ms          10 cutoff=<value>         output cutoff point            0.0001 lineno=y|n             line number  in  traces?         y thread=y|n             thread  in  traces?              n doe=y|n                dump on  exit ?                  y msa=y|n                Solaris micro state accounting n force=y|n              force output to < file >         y verbose=y|n            print messages about dumps     y< /code >

    来几个官方指南上的实例。

    CPU Usage Sampling Profiling(cpu=samples)的例子：

1	<code class= "hljs nginx" ><span class= "hljs-attribute" >java -agentlib:hprof=cpu=samples,interval=<span class= "hljs-number" >20,depth=<span class= "hljs-number" >3 Hello< /span >< /span >< /span >< /code >

    上面每隔20毫秒采样CPU消耗信息，堆栈深度为3，生成的profile文件名称是java.hprof.txt，在当前目录。 

    CPU Usage Times Profiling(cpu=times)的例子，它相对于CPU Usage Sampling Profile能够获得更加细粒度的CPU消耗信息，能够细到每个方法调用的开始和结束，它的实现使用了字节码注入技术（BCI）：

1	<code class= "hljs bash" >javac -J-agentlib:hprof=cpu=<span class= "hljs-built_in" > times  Hello.java< /span >< /code >

    Heap Allocation Profiling(heap=sites)的例子：

1	<code class= "hljs nginx" ><span class= "hljs-attribute" >javac -J-agentlib:hprof=heap=sites Hello.java< /span >< /code >

    Heap Dump(heap=dump)的例子，它比上面的Heap Allocation Profiling能生成更详细的Heap Dump信息：

1	<code class= "hljs nginx" ><span class= "hljs-attribute" >javac -J-agentlib:hprof=heap=dump Hello.java< /span >< /code >

    虽然在JVM启动参数中加入-Xrunprof:heap=sites参数可以生成CPU/Heap Profile文件，但对JVM性能影响非常大，不建议在线上服务器环境使用