Java Runtime exec 运行一段时间后挂起的解决方案

今天写程序遇到一个问题，使用Process process = Runtime.getRuntime().exec(cmd); proc.waitFor(); 创建一个进程后，如果进程产生大量的输出而不去读取的话，当超出系统的缓存时，此进程会被挂起，而正在运行的程序又因 proc.waitFor();在等待进程的消亡，所以产生了一个死锁。

 

从网上找到了一个解决方案：

 

碰到一个项目需要从Java中运行Perl程序，这个Perl程序调用客户的Web service，每次发送一个请求，接受一个响应。Java程序中包含多个请求，需要多次调用Perl程序，并且接受和解析响应（这个烂设计可不是我干 的，我实在不明白强大的Java Web Service为什么要弄成这样，不过客户是老大）。使用Java Runtime的exec()方法，发现运行一段时间后，进程就被挂起了（之前的响应完全正确）。于是分析原因，发现我在运行exec()方法后，立刻执 行了Process的waitFor()方法，这里出了问题。在网上找到一篇文章讲述这个问题：
地址：http://brian.pontarelli.com/2005/11/11/java-runtime-exec-can-hang/

Java Runtime exec can hang

November 11, 2005 on 4:40 pm | In Java |

The next version of Savant is going to focus heavily on the stand-alone runtime and support for dialects and plugins. Supporting all that is largely handled by using a simple executor framework I wrote around Java 1.4 and lower’s Runtime.exec method. A few things to keep in mind when using this:

1. Always read from the streams prior to calling waitFor. Otherwise you could end up waiting forever on Windows and other OS platforms whose I/O buffers can’t store enough from standard out and standard error to ensure the program has finished. These platforms will pause the execution of whatever is running until something reads the buffered content from standard out and standard error. I would imagine all platforms suffer from this, but some platforms have larger buffers than others. Needless to say, always read from the streams first.
2. Always read from standard error first. I ran across a bug where some OS platforms will always open standard out, but never close it. What this means is that if you read from standard out first and the process only writes to standard error, you’ll hang forever waiting to read. If you read from standard error first, you’ll always be okay on these platforms because the OS seems to shutdown standard error. I think however, that the best way to handle all cases is to check both standard error and standard out for readiness and only read from them if they have something to offer. The downside I could see here is that error isn’t ready, but eventually will be.

可以看出：

• 永远要在调用waitFor()方法之前读取数据流
• 永远要先从标准错误流中读取，然后再读取标准输出流

于是将waitFor()方法放在读取数据流后调用，目前没有发现什么问题。

正好解决了我心中的疑问，非常感谢！

我们的程序一开始就是exec完了接着waitFor()，但bat文件执行不完整：

Process proc = Runtime.getRuntime().exec(cmd);
                proc.waitFor();

后面的build中在waitFor()之前读取了数据流，bat文件就可以完整执行了：

Process proc = Runtime.getRuntime().exec(cmd);
     StreamGobbler errorGobbler = new StreamGobbler(proc.getErrorStream(), "Error");           
                 StreamGobbler outputGobbler = new StreamGobbler(proc.getInputStream(), "Output");
                 errorGobbler.start();
                 outputGobbler.start();

     proc.waitFor();

class StreamGobbler extends Thread {
 InputStream is;

 String type;

 StreamGobbler(InputStream is, String type) {
  this.is = is;
  this.type = type;
 }

 public void run() {
  try {
   InputStreamReader isr = new InputStreamReader(is);
   BufferedReader br = new BufferedReader(isr);
   String line = null;
   while ((line = br.readLine()) != null) {
    if (type.equals("Error"))
     LogManager.logError(line);
    else
     LogManager.logDebug(line);
   }
  } catch (IOException ioe) {
   ioe.printStackTrace();
  }
 }
}

TestPrint.bat:

echo P1=%1  >D:/2.1.2env/2.1.2home/CompuSet/output/TestPrint.log
echo P2=%2 >>D:/2.1.2env/2.1.2home/CompuSet/output/TestPrint.log
echo P3=%3 >>D:/2.1.2env/2.1.2home/CompuSet/output/TestPrint.log
echo P4=%4 >>D:/2.1.2env/2.1.2home/CompuSet/output/TestPrint.log
echo P5=%5 >>D:/2.1.2env/2.1.2home/CompuSet/output/TestPrint.log
echo P6=%6 >>D:/2.1.2env/2.1.2home/CompuSet/output/TestPrint.log

Bad_TestPrint.log:

P1=C:/xPression/CompuSet/output/MartyTestOut1.afp 
P2=Literal1
P3="Rick Skinner"
P4=Parameter3

Good_TestPrint.log

P1=C:/xPression/CompuSet/output/MartyTestOut1.afp 
P2=Literal1
P3="Rick Skinner"
P4=Parameter3
P5=Parameter4
P6=Parameter5