转载-JVM性能调优监控工具

时间  2019-11-26
标签 转载 jvm 性能 监控 工具 栏目 Java 繁體版
原文   原文链接

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

 这些问题在平常开发中可能被不少人忽视（好比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），但可以理解并解决这些问题是Java程序员进阶的必备要求。本文将对一些经常使用的JVM性能调优监控工具进行介绍，但愿能起抛砖引玉之用。本文参考了网上不少资料，难以一一列举，在此对这些资料的做者表示感谢！关于JVM性能调优相关的资料，请参考文末。css

 

A、 jps(Java Virtual Machine Process Status Tool)      html

    jps主要用来输出JVM中运行的进程状态信息。语法格式以下：java

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 >

    若是不指定hostid就默认为当前主机或服务器。nginx

    命令行参数选项说明以下：程序员

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 >

   好比下面：算法

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、 jstackshell

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

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以下：ubuntu

1
2
<code class= "hljs haml" >-<span class= "ruby" >l long listings，会打印出额外的锁信息，在发生死锁时能够用jstack -l pid来观察锁持有状况
-<span class= "ruby" >m mixed mode，不只会输出Java堆栈信息，还会输出C /C ++堆栈信息（好比Native方法）< /span >< /span >< /code >

    jstack能够定位到线程堆栈，根据堆栈信息咱们能够定位到具体代码，因此它在JVM性能调优中使用得很是多。下面咱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。数组

    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 >

    获得进程ID为21711，第二步找出该进程内最耗费CPU的线程，可使用ps -Lfp pid或者ps -mp pid -o THREAD, tid, time或者top -Hp pid，我这里用第三个，输出以下：

    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 >

    获得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，而后根据线程ID的十六进制值grep，以下：

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 >

    能够看到CPU消耗在PollIntervalRetrySchedulerThread这个类的Object.wait()，我找了下个人代码，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情况，通常结合jhat使用。

    jmap语法格式以下：

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 >

    若是运行在64位JVM上，可能须要指定-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，以下图：

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

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，若是带上live则只统计活对象，以下：

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，说明以下：

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 其余对象< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /span >< /code >

    还有一个很经常使用的状况是：用jmap把进程内存使用状况dump到文件中，再用jhat分析查看。jmap进行dump命令格式以下：

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 >

    我同样地对上面进程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出来的文件能够用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 >

     注意若是Dump文件太大，可能须要加上-J-Xmx512m这种参数指定最大堆内存，即jhat -J-Xmx512m -port 9998 /tmp/dump.dat。而后就能够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你们能够本身去摸索下，最后一项支持OQL（对象查询语言）。

 

D、jstat（JVM统计监测工具）

    语法格式以下：

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系统上通常就是进程ID。interval是采样时间间隔。count是采样数目。好比下面输出的是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：

    能够看出：

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 >

    如今来解释各列含义：

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可以展示CPU使用率，统计堆内存使用状况。

    语法格式以下：

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 >

    完整的命令选项以下：

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可以得到更加细粒度的CPU消耗信息，可以细到每一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参数能够生成CPU/Heap Profile文件，但对JVM性能影响很是大，不建议在线上服务器环境使用。

 

其余JVM性能调优参考资料：

《Java虚拟机规范》

《Java Performance》

《Trouble Shooting Guide for JavaSE 6 with HotSpot VM》: http://www.oracle.com/technetwork/java/javase/tsg-vm-149989.pdf 

《Effective Java》

VisualVM: http://docs.oracle.com/javase/7/docs/technotes/guides/visualvm/

jConsole: http://docs.oracle.com/javase/1.5.0/docs/guide/management/jconsole.html

Monitoring and Managing JavaSE 6 Applications: http://www.oracle.com/technetwork/articles/javase/monitoring-141801.html

BTrace：https://kenai.com/projects/btrace

相关文章
相关标签/搜索
每日一句
    每一个你不满意的现在，都有一个你没有努力的曾经。
本站公众号
   欢迎关注本站公众号,获取更多信息
联系我们 最近搜索 最新文章 沪ICP备13005482号-10 MyBatis教程 SQL 教程 MySQL教程 Java 教程 Thymeleaf 教程 Hibernate教程 Spring教程 Redis教程