进程挂起并且PIPE被阻止

Question

I seem to have a deadlock. I have perl scripts forking and calling another perl scripts. And the process hangs somewhere.

I am running the program on : Darwin Kernel Version 12.3.0: Sun Jan 6 22:37:10 PST 2013; root:xnu-2050.22.13~1/RELEASE_X86_64 x86_64

"lsof" has 4 entries referring to the same PIPE:

perl5.12 1414 root 1 PIPE 0x48937dc1254fe937 16384 ->0x48937dc1254fe727

perl5.12 1768 root 1 PIPE 0x48937dc1254fe937 16384 ->0x48937dc1254fe727

perl5.12 1759 root 1 PIPE 0x48937dc1254fe937 16384 ->0x48937dc1254fe727

perl5.12 1760 root 1 PIPE 0x48937dc1254fe937 16384 ->0x48937dc1254fe727

I am suspecting that is the cause of the hang. Do we have any commands that could tell me which process reading/writing into this PIPE? Or any further information would be appreciated. Thanks in advance!

Answer 1

There's two likely possibilities I can think of:

1. There's deadlock due to output buffering. Try enabling autoflush on all the output pipes. This is likely if two processes are communicating bidirectionally using pipes: they each write something and wait to read a response, but because the output is buffered the response is never sent to the pipe.

2. A process is waiting for EOF on the pipe, but it never comes. If the pipe is created in a parent process and then inherited by a child process, you need to ensure that all the processes close the write end of the pipe so that the reader will read EOF.

Answer 2

As Barmar also explained the Child Processes can get stuck when the Output Buffer fills up. In this case you will find the Child Process stuck in the write() Function Call.

You will have to use Perl IO::Select Module in the Parent Process to continuesly read the Output Buffer from the Child and thereby empty it if the Output from the Child Process is bigger than the Buffer.

The official Perl Documentation at http://perldoc.perl.org/functions/sysread.html explains:

sysread FILEHANDLE,SCALAR,LENGTH
Attempts to read LENGTH bytes of data into variable SCALAR from the specified FILEHANDLE, using read(2). It bypasses buffered IO, so mixing this with other kinds of reads, print, write, seek, tell, or eof can cause confusion because the perlio or stdio layers usually buffer data.

Further to see the System Activity I usually use the strace Command which nicely shows how each process reads and writes data.