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 GenerationEden <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
|
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 allcpu=samples|times|old CPU usage off
monitor=y|n monitor contention nformat=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 offdepth=<size> stack trace depth 4interval=<ms> sample interval in ms 10
cutoff=<value> output cutoff point 0.0001lineno=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 nforce=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