性能调优案例分享：jvm crash的原因

精选转载

北京茑萝00 2016-02-29 16:54:01

poptest是国内唯一一家培养测试开发工程师的培训机构，以学员能胜任自动化测试，性能测试，测试工具开发等工作为目标。如果对课程感兴趣，请大家咨询qq：908821478，咨询电话010-84505200。poptest培训的课堂上通过实际案例来帮助学员提高性能测试的能力，胜任企业性能测试的工作。

jboss服务器经常莫名地突然停止运行，查看jboss控制台错误日志，发现只有下面一行：

/opt/.../jboss/bin/run.sh: line 181: 26430 段错误 "$JAVA" $JAVA_OPTS -Djava.endorsed.dirs="$JBOSS_ENDORSED_DIRS" -classpath "$JBOSS_CLASSPATH" org.jboss.Main "$@"

一开始对上面的错误日志也没引起重视，一直在找JVM crash的日志hs_err_<pid>，看了很多网上的帖子，包括JVM官网的介绍，为什么没有生成hs_err_pid日志，可能的原因就是没权限或者空间不够，查看了磁盘空间是足够的，于是想到可能没有权限（其实默认目录就是JVM的工作目录，应该是有权限的，但当时唯一能怀疑的就是这个），所以就想到JVM参数（-XX:ErrorFile=./hs_err_<pid>.log）可以配置这个日志文件的保存目录，但这个参数是JDK1.6才引入的，而我们系统杯催的还停留在JDK1.5上，没办法想到另外一个参数（-XX:OnError="<cmd args>;<cmd args>"），里面配置的命令是kill -3 <pid>，这样JVM在crash的时候可以将线程堆栈dump出来，想法是好的，可这个命令一直没配上去，只能作罢。

拿不到hs_err_<pid>日志，唯一的线索就是上面仅有的一行错误日志，于是google“段错误”“segmentation fault”，发现可以通过core dump日志进行分析，解决思路也就自然而然有了。

1.core dump介绍

程序异常退出（crash）时会自动生成一个core文件，包含了程序运行时的内存，寄存器状态，堆栈指针，内存管理等信息，也就是把程序当时工作的状态存储成一个文件。不仅仅是在出错的时候会生成core dump文件，在系统卡住或者cpu使用率很高的时候也可以手动触发生成core dump文件（当然这种情况也可以直接通过jmap和jstack dump出内存和线程堆栈，但JDK1.6以前，这2个命令经常执行不成功），这样就可以很好的分析了。

2.生成core dump文件

core dump文件生成开关其实是通过对生成的文件大小进行控制达到的，默认大小是0，也就是说默认是不生成core dump文件的，可以通过命令ulimit -c进行查看。将此参数修改成unlimited就可以生成core dump文件了，但值得注意的一点是，每个应用进程都会读取自己的一套系统参数，可以查看进程对应的内存文件/proc/<pid> /limits中的信息来判断修改后的参数值是否对此应用进程生效了，limits文件中的信息如下：

Limit                     Soft Limit           Hard Limit           Units
Max cpu time              unlimited            unlimited            seconds
Max file size             unlimited            unlimited            bytes
Max data size             unlimited            unlimited            bytes
Max stack size            10485760             unlimited            bytes
Max core file size        unlimited           unlimited            bytes
Max resident set          unlimited            unlimited            bytes
Max processes             20480                20480                processes
Max open files            204800               204800               files
Max locked memory         32768                32768                bytes
Max address space         unlimited            unlimited            bytes
Max file locks            unlimited            unlimited            locks
Max pending signals       1024                 1024                 signals
Max msgqueue size         819200               819200               bytes

上述信息中的对应的参数值为unlimited，代表此应用进程的core dump开关是开着的。ulimit -c unlimited这个命令可以在很多地方修改，比如用root在命令行执行，在/etc/profile中增加（/etc/profile文件中已经有 defaultulimit -S -c 0 > /dev/null 2>&1，可以直接修改0为unlimited），在用户的.bash_profile中增加..这些方式执行后的影响范围还是挺复杂的，到现在也没弄清楚。当时通过root用户命令行修改后，在shell里查询ulimit -c是生效了，以为应用程序也生效了，觉得不用应用都不需要重启就生效了挺好，等到下一次crash的时候发现还是没有生成core dump文件，于是到文件/proc/<pid>/limits中检查此参数值才发现应用程序所读取的值还是0。后来采用一种比较简单直接的方式，用root用户修改/etc/profile文件中的ulimit -c的值，然后重新登录，重启应用，再检查/proc/<pid>/limits中的值，确保参数已经生效，后来应用再crash时core dump文件也如期而至，心里暗爽，终于可以开始分析堆栈了。

core dump文件生成的默认路径在用户的工作目录（启动脚本目录），可以通过设置参数/proc/sys/kernel/core_pattern的值改变 core dump文件的最终生成路径及文件名格式。core dump文件成功后，在jboss控制台错误日志中的日志也多了core dumped关键字，如下：

/opt/.../jboss/bin/run.sh: line 181: 9176 段错误 (core dumped) "$JAVA" $JAVA_OPTS -Djava.endorsed.dirs="$JBOSS_ENDORSED_DIRS" -classpath "$JBOSS_CLASSPATH" org.jboss.Main "$@"

3.core dump分析

有了core dump文件，接下来要做的就是通过命令去解析此文件，定位具体问题了，主要有以下三个命令：

(1)先执行gdb $JAVA_HOME$/bin/java core-java-16427-1325846515,再执行bt,输出结果如下:

Loaded symbols for /opt/.../oracle/lib/libnnz10.so
Reading symbols from /opt/.../install/oracle/10.2.0.3/lib/libociei.so...(no debugging symbols found)...done.
Loaded symbols for /opt/.../oracle/lib/libociei.so
Reading symbols from /usr/lib/gconv/GB18030.so...(no debugging symbols found)...done.
Loaded symbols for /usr/lib/gconv/GB18030.so
#0 0x4662051e in ttcrd2r ()
   from /opt/.../oracle/lib/libclntsh.so.10.1
(gdb) bt
#0 0x4662051e in ttcrd2r () from /opt/.../oracle/lib/libclntsh.so.10.1
#1 0x46617212 in ttcrbur () from /opt/.../oracle/lib/libclntsh.so.10.1
#2 0x466176e6 in ttcbur () from /opt/.../oracle/lib/libclntsh.so.10.1
#3 0x465ba8c7 in ttcdrv () from /opt/.../oracle/lib/libclntsh.so.10.1
#4 0x464a1bc8 in nioqwa () from /opt/.../oracle/lib/libclntsh.so.10.1
#5 0x4630d618 in upirtrc () from /opt/.../oracle/lib/libclntsh.so.10.1
#6 0x462825f6 in kpurcsc () from /opt/.../oracle/lib/libclntsh.so.10.1
#7 0x46236a0d in kpuexecv8 () from /opt/.../oracle/lib/libclntsh.so.10.1
#8 0x46238ec4 in kpuexec () from /opt/.../oracle/lib/libclntsh.so.10.1
#9 0x463121b6 in OCIStmtExecute () from /opt/.../oracle/lib/libclntsh.so.10.1
#10 0x46e725aa in Java_oracle_jdbc_driver_T2CConnection_t2cClearAllApplicationContext ()
   from /opt/.../install/oracle/10.2.0.3/lib/libocijdbc10.so
#11 0x46e7580f in Java_oracle_jdbc_driver_T2CStatement_t2cDefineExecuteFetch ()
   from /opt/.../install/oracle/10.2.0.3/lib/libocijdbc10.so
#12 0xb125a8b8 in ?? ()
#13 0x47e01bc8 in ?? ()
#14 0x47d46bb0 in ?? ()
#15 0x47d46c40 in ?? ()
#16 0x4babf774 in ?? ()
#17 0x00000000 in ?? ()

通过这个debug命令可以查看到应用crash时正在做什么，这里是在获取jdbc连接，但这个并不代表问题就出在这里，有可能这只是个结果，起因还需要进一步排查。

(2)光有上述信息还不足以定位出问题在哪，可以通过命令jstack $JAVA_HOME$/bin/java core-java-16427-1325846515获取应用crash时的线程堆栈，输出结果如下：

Thread 11586: (state = BLOCKED)
- java.lang.Object.wait(long) @bci=0 (Compiled frame; information may be imprecise)
- java.lang.Object.wait() @bci=2, line=474 (Compiled frame)
- edu.emory.mathcs.backport.java.util.concurrent.LinkedBlockingDeque.takeFirst() @bci=20, line=389 (Compiled frame)
- edu.emory.mathcs.backport.java.util.concurrent.LinkedBlockingDeque.take() @bci=1, line=577 (Compiled frame)
- edu.emory.mathcs.backport.java.util.concurrent.ThreadPoolExecutor$Worker.run() @bci=18, line=674 (Compiled frame)
Error occurred during stack walking:

Thread 10924: (state = IN_NATIVE)
- oracle.jdbc.driver.T2CStatement.t2cDefineExecuteFetch(oracle.jdbc.driver.OracleStatement, long, int, int, int, int, boolean, boolean, byte[], int, byte, int, int, short[], int, byte[], char[], int, int, short[], byte[], int, int, boolean, boolean, oracle.jdbc.driver.Accessor[], byte[][][], long[], byte[], int, char[], int, short[], int) @bci=0 (Interpreted frame)
- oracle.jdbc.driver.T2CPreparedStatement.doDefineExecuteFetch() @bci=190, line=878 (Interpreted frame)
- oracle.jdbc.driver.T2CPreparedStatement.executeForRows(boolean) @bci=34, line=760 (Interpreted frame)
- oracle.jdbc.driver.OracleStatement.executeMaybeDescribe() @bci=154, line=1062 (Interpreted frame)
- oracle.jdbc.driver.OracleStatement.doExecuteWithTimeout() @bci=102, line=1126 (Interpreted frame)
- oracle.jdbc.driver.OraclePreparedStatement.executeInternal() @bci=94, line=3339 (Interpreted frame)
- oracle.jdbc.driver.OraclePreparedStatement.execute() @bci=17, line=3445 (Interpreted frame)
- org.jboss.resource.adapter.jdbc.CachedPreparedStatement.execute() @bci=4, line=216 (Interpreted frame)
- org.jboss.resource.adapter.jdbc.WrappedPreparedStatement.execute() @bci=12, line=209 (Compiled frame)
- com.ibatis.sqlmap.engine.execution.SqlExecutor.executeQuery(com.ibatis.sqlmap.engine.scope.RequestScope, java.sql.Connection, java.lang.String, java.lang.Object[], int, int, com.ibatis.sqlmap.engine.mapping.statement.RowHandlerCallback) @bci=135, line=180 (Compiled frame)
- com.ibatis.sqlmap.engine.mapping.statement.GeneralStatement.sqlExecuteQuery(com.ibatis.sqlmap.engine.scope.RequestScope, java.sql.Connection, java.lang.String, java.lang.Object[], int, int, com.ibatis.sqlmap.engine.mapping.statement.RowHandlerCallback) @bci=15, line=205 (Interpreted frame)
- com.ibatis.sqlmap.engine.mapping.statement.GeneralStatement.executeQueryWithCallback(com.ibatis.sqlmap.engine.scope.RequestScope, java.sql.Connection, java.lang.Object, java.lang.Object, com.ibatis.sqlmap.client.event.RowHandler, int, int) @bci=169, line=173 (Compiled frame)
- com.ibatis.sqlmap.engine.mapping.statement.GeneralStatement.executeQueryForObject(com.ibatis.sqlmap.engine.scope.RequestScope, com.ibatis.sqlmap.engine.transaction.Transaction, java.lang.Object, java.lang.Object) @bci=28, line=104 (Compiled frame)
- com.ibatis.sqlmap.engine.impl.SqlMapExecutorDelegate.queryForObject(com.ibatis.sqlmap.engine.scope.SessionScope, java.lang.String, java.lang.Object, java.lang.Object) @bci=58, line=561 (Compiled frame)
- com.ibatis.sqlmap.engine.impl.SqlMapExecutorDelegate.queryForObject(com.ibatis.sqlmap.engine.scope.SessionScope, java.lang.String, java.lang.Object) @bci=5, line=536 (Interpreted frame)
- com.ibatis.sqlmap.engine.impl.SqlMapSessionImpl.queryForObject(java.lang.String, java.lang.Object) @bci=10, line=93 (Interpreted frame)
- org.springframework.orm.ibatis.SqlMapClientTemplate$1.doInSqlMapClient(com.ibatis.sqlmap.client.SqlMapExecutor) @bci=9, line=273 (Interpreted frame)
- org.springframework.orm.ibatis.SqlMapClientTemplate.execute(org.springframework.orm.ibatis.SqlMapClientCallback) @bci=242, line=209 (Compiled frame)
- org.springframework.orm.ibatis.SqlMapClientTemplate.queryForObject(java.lang.String, java.lang.Object) @bci=11, line=271 (Compiled frame)
......
- org.springframework.web.servlet.DispatcherServlet.doService(javax.servlet.http.HttpServletRequest, javax.servlet.http.HttpServletResponse) @bci=219, line=809 (Interpreted frame)
- org.springframework.web.servlet.FrameworkServlet.processRequest(javax.servlet.http.HttpServletRequest, javax.servlet.http.HttpServletResponse) @bci=10, line=571 (Compiled frame)
- org.springframework.web.servlet.FrameworkServlet.doPost(javax.servlet.http.HttpServletRequest, javax.servlet.http.HttpServletResponse) @bci=3, line=511 (Interpreted frame)
- javax.servlet.http.HttpServlet.service(javax.servlet.http.HttpServletRequest, javax.servlet.http.HttpServletResponse) @bci=139, line=717 (Interpreted frame)
- javax.servlet.http.HttpServlet.service(javax.servlet.ServletRequest, javax.servlet.ServletResponse) @bci=30, line=810 (Compiled frame)
- org.apache.catalina.core.ApplicationFilterChain.doFilter(javax.servlet.ServletRequest, javax.servlet.ServletResponse) @bci=101, line=173 (Interpreted frame)
- org.apache.catalina.core.ApplicationDispatcher.invoke(javax.servlet.ServletRequest, javax.servlet.ServletResponse) @bci=396, line=672 (Interpreted frame)
- org.apache.catalina.core.ApplicationDispatcher.processRequest(javax.servlet.ServletRequest, javax.servlet.ServletResponse) @bci=64, line=463 (Interpreted frame)
- org.apache.catalina.core.ApplicationDispatcher.doForward(javax.servlet.ServletRequest, javax.servlet.ServletResponse) @bci=469, line=398 (Interpreted frame)
- org.apache.catalina.core.ApplicationDispatcher.forward(javax.servlet.ServletRequest, javax.servlet.ServletResponse) @bci=55, line=301 (Interpreted frame)

- org.springframework.web.servlet.view.InternalResourceView.renderMergedOutputModel(java.util.Map, javax.servlet.http.HttpServletRequest, javax.servlet.http.HttpServletResponse) @bci=244, line=240 (Interpreted frame)
- org.springframework.web.servlet.view.AbstractView.render(java.util.Map, javax.servlet.http.HttpServletRequest, javax.servlet.http.HttpServletResponse) @bci=157, line=252 (Interpreted frame)
- org.springframework.web.servlet.DispatcherServlet.render(org.springframework.web.servlet.ModelAndView, javax.servlet.http.HttpServletRequest, javax.servlet.http.HttpServletResponse) @bci=225, line=1173 (Compiled frame)
- org.springframework.web.servlet.DispatcherServlet.doService(javax.servlet.http.HttpServletRequest, javax.servlet.http.HttpServletResponse) @bci=219, line=809 (Interpreted frame)
- org.springframework.web.servlet.FrameworkServlet.processRequest(javax.servlet.http.HttpServletRequest, javax.servlet.http.HttpServletResponse) @bci=10, line=571 (Compiled frame)
- org.springframework.web.servlet.FrameworkServlet.doPost(javax.servlet.http.HttpServletRequest, javax.servlet.http.HttpServletResponse) @bci=3, line=511 (Interpreted frame)
- javax.servlet.http.HttpServlet.service(javax.servlet.http.HttpServletRequest, javax.servlet.http.HttpServletResponse) @bci=139, line=717 (Interpreted frame)
- javax.servlet.http.HttpServlet.service(javax.servlet.ServletRequest, javax.servlet.ServletResponse) @bci=30, line=810 (Compiled frame)
- org.apache.catalina.core.ApplicationFilterChain.doFilter(javax.servlet.ServletRequest, javax.servlet.ServletResponse) @bci=101, line=173 (Interpreted frame)
- org.apache.catalina.core.ApplicationDispatcher.invoke(javax.servlet.ServletRequest, javax.servlet.ServletResponse) @bci=396, line=672 (Interpreted frame)
- org.apache.catalina.core.ApplicationDispatcher.processRequest(javax.servlet.ServletRequest, javax.servlet.ServletResponse) @bci=64, line=463 (Interpreted frame)
- org.apache.catalina.core.ApplicationDispatcher.doForward(javax.servlet.ServletRequest, javax.servlet.ServletResponse) @bci=469, line=398 (Interpreted frame)
- org.apache.catalina.core.ApplicationDispatcher.forward(javax.servlet.ServletRequest, javax.servlet.ServletResponse) @bci=55, line=301 (Interpreted frame)

......

通过这个线程堆栈可以看出问题根本原因是一直在forward，代码出现死循环，总的forward的次数有100多次，直到线程堆栈溢出。这也能说明前面通过gdb命令查看到的获取jdbc连接仅仅是最近的一次循环正在执行的操作而已，不是真正的起因。这个线程的死循环forward同样影响到了其他的线程，导致出现Error occurred during stack walking错误，线程一直处于BLOCKED状态，总之线程堆栈对分析问题原因是非常有帮助的。

(3).可以通过命令jmap $JAVA_HOME$/bin/java core-java-16427-1325846515查看到应用的内存使用，结果如下：

using parallel threads in the new generation.
using thread-local object allocation.
Concurrent Mark-Sweep GC

Heap Configuration:
   MinHeapFreeRatio = 40
   MaxHeapFreeRatio = 70
   MaxHeapSize      = 1342177280 (1280.0MB)
   NewSize          = 134217728 (128.0MB)
   MaxNewSize       = 134217728 (128.0MB)
   OldSize          = 402653184 (384.0MB)
   NewRatio         = 15
   SurvivorRatio    = 1024
   PermSize         = 100663296 (96.0MB)
   MaxPermSize      = 100663296 (96.0MB)

Heap Usage:
New Generation (Eden + 1 Survivor Space):
   capacity = 134152192 (127.9375MB)
   used     = 22525464 (21.481956481933594MB)
   free     = 111626728 (106.4555435180664MB)
   16.790977220856742% used
Eden Space:
   capacity = 134086656 (127.875MB)
   used     = 22525464 (21.481956481933594MB)
   free     = 111561192 (106.3930435180664MB)
   16.799183954591275% used
From Space:
   capacity = 65536 (0.0625MB)
   used     = 0 (0.0MB)
   free     = 65536 (0.0625MB)
   0.0% used
To Space:
   capacity = 65536 (0.0625MB)
   used     = 0 (0.0MB)
   free     = 65536 (0.0625MB)
   0.0% used
concurrent mark-sweep generation:
   capacity = 1207959552 (1152.0MB)
   used     = 764300224 (728.8934936523438MB)
   free     = 443659328 (423.10650634765625MB)

Exception in thread "main" java.lang.OutOfMemoryError: requested 4096 bytes for jbyte in /BUILD_AREA/jdk1.5.0_08/hotspot/src/share/vm/prims/jni.cpp. Out of swap space?

最后一行出现了OOM，这是因为dump出来的文件太大，32位机器支持的最大内存也就2G，所以在解析时出现OOM问题，可以放到64位服务器上进行解析。

4.常见的core dump原因

造成coredump的原因很多，常见的有内存访问越界，非法指针，堆栈溢出等等，也可以通过发送操作系统信号中断应用，比如kill -SIGSEGV <pid>，这是一个无效存储信号。如果想手动生成core dump文件，可以通过gdb -p <pid>生成。