Delivery-Date: Sun Oct 12 20:33:39 2008
Date: Sun, 12 Oct 2008 20:47:30 -0700
From: Leslie Cheung <lccheung@usc.edu>
Subject: broken pipes problems
To: cs551@merlot.usc.edu
Message-id: <00c101c92ce6$6d5be910$4813bb30$@edu>
MIME-version: 1.0
Content-type: text/plain; charset=us-ascii
Content-language: en-us
Content-transfer-encoding: 7bit
Thread-index: Acks5mrrC3aVs7mQT0Oj1DWv50k12g==

Hi class,

I know a few of you are encountering broken pipes problems. Here is a note I
wrote a while ago, and I hope this is useful to all of you.

If what I suggest below cannot fix your problem, it's best to schedule an
appointment with Bill or myself so we can look at your code. It's really
hard to fix a bug like this without going into your code.


--Leslie

---------------------------------------------------------
Hi all,

I debugged a couple SIGPIPE problems, and I hope this information is useful.

So I assume if we are unable to connect to the other node (e.g., it has not
been started), we sleep for a while, and try again later. One problem I saw
is that when you try "reusing" a socket, it is able to complete "connect",
but whenever you send something using that reused socket, you get SIGPIPE.
Why? I have no idea either, but this is how things work.

Instead of saying...

-------------------------------------
int sockfd = socket(...);

while (!done){
     if (connect(sockfd, ...) < 0){
          //cannot connect
     } else {
          //it won't give you an error for connect, so it comes here
          //but if you try to send something, it gives you SIGPIPE
          write(sockfd, ...); //this line give you SIGPIPE
     }
     sleep(10);
}
-------------------------------------

You should do

-------------------------------------

while (!done){
     int sockfd = socket(...);
     if (connect(sockfd, ...) < 0){
          //cannot connect
          //you should now close the socket
          close(sockfd);
     } else {
          //connected
          //now if you send something, it should work
          write(sockfd, ...); //this should work
     }
     sleep(10);
}
-------------------------------------

  In other words, you should create a new socket "inside the loop".
  Again, this is just one possible scenario that may give you SIGPIPE. Yours
might be some other problems.