a895717c   
   From: rick.jones2@hp.com   
      
   David Goodenough  wrote:   
   > What's killing me is that the previous server may have left   
   > connections in any of FIN_WAIT_[12], LAST_ACK or TIME_WAIT.  Even with   
   > SO_REUSEADDR, these "zombie" connections appear to make it so that a   
   > new server starting can't successfully bind / listen for incoming   
   > connections as soon as it starts up.   
      
   Shouldn't be that way - if you have successfully set SO_REUSEADDR the   
   only endpoint that is supposed to preclude your doing a bind/listen on   
   the well-known port number is another listen endpoint bound to that   
   well-known port.   
      
   Are you *quite* certain (IE have seen it in an strace) that your   
   setsockopt() calls to set SO_REUSEADDR are successful?   
      
   > I'm well aware of the theory that says that you leave a socket in   
   > TIME_WAIT for 2*MSL time to ensure you won't have problems if there   
   > happen to be rogue packets floating around the net.  However, keeping   
   > a whole server from restarting, thus denying access to hundreds of   
   > users, to protect against a problem for a single user who might have a   
   > packet stuck temporarily on some router somewhere does not match our   
   > priority model.   
      
   > "The needs of the many outweigh the needs of the few or the one."   
      
   It is probably arguing over the number of angels which can dance on   
   the head of the pin, but I'll ask - what are the consequences of data   
   corruption when it happens to the few or the one?   
      
   > So, what settings can I change in /proc/tcp/* or elsewhere to ensure   
   > that a socket can't remain in any of those states for more than 10   
   > seconds.   
      
   Are you also going to limit TCP retransmissions to no more than 10   
   seconds to deal with connections in ESTABLISHED?  How about SYN_RECVD?   
      
   rick jones   
   --   
   The glass is neither half-empty nor half-full. The glass has a leak.   
   The real question is "Can it be patched?"   
   these opinions are mine, all mine; HP might not want them anyway... :)   
   feel free to post, OR email to rick.jones2 in hp.com but NOT BOTH...   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)