等待进程和读取流之间的并发问题？

Question

I use a ProcessBuilder to run processes. I handle Input/Output streams by submitting the corresponding runnables handling them in a thread pool ( Executors.newCachedThreadPool() ).
I get the result but every now and then I don't get anything.
For instance if I do: cmd \\C dir to the process builder I get the results of dir back but sometimes I don't get anything (despite that the result seems to comeback from the runnable that handles the process.getInputStream ).
How can I debug this? It shows up intermitently. With the same code I did not have any problem when I did new Thread(runnable).start() . It started to happen after I switched to a thread pool.

Update:
I think I found something:
I do the following in the Runnable :

 try {  
    while ( (line = br.readLine()) != null) {  
            pw.println(line);  
                sb.append(line);   
    }  
    System.out.println("Finished reading "+sb.length());  
} catch (IOException e) {             
    e.printStackTrace();  
}  
finally{  
   pw.flush();     
  try{
    isr.close();  
  }catch(Exception e){}  
}

In the cases that does not work it prints Finished reading 521 . But I try to get the result via the pw and not sb .
pw is PrintWriter pw = PrintWriter(outputStream);` which I pass in the runnable

Update 2:
It seems that: status = process.waitFor(); returns earlier before the runnable that handled the inputstream finishes. How can this happen?
I read in the javadoc:
the calling thread will be blocked until the subprocess exits . So does that mean that I could return before consuming the I/O streams?

Update 3:
Seems to be the same issue here in Ruby
Ie there is some race condition between the process ending and consuming the output

Answer 1

Yes. stdio between processes is buffered (usually 4KB buffers). Process A write into the buffer and exist. Process B has two threads; one waits for the end of A and the other reads the output from A. There is no way to be sure which thread executes first.

So it's possible (even likely when there is a lot of output) that process.waitFor(); returns before all buffered output has been read.

Note that flushing doesn't help here because it just makes sure that A has written everything. There is no way to "force" B to read the data in a similar way.

Therefore, you should remember the exit status and consider the process as "fully terminated" only when you read EOF from the input stream.

EDIT One solution would be to move the waitFor() into the stream gobbler and convert the gobbler into a Callable which you could then submit to the executor and use the Future API ( example ) to get the result(s).