NetBSD Problem Report #3117

Received: (qmail-queue invoked from smtpd); 17 Jan 1997 11:08:57 -0000
Message-Id: <199701171105.MAA13657@abro.ericsson.se>
Date: Fri, 17 Jan 1997 12:05:02 +0100 (MET)
From: Michael.Eriksson@era-t.ericsson.se
Reply-To: Michael.Eriksson@era-t.ericsson.se
To: gnats-bugs@gnats.netbsd.org
Cc: eramer@era-t.ericsson.se
Subject: YP doesn't let RPC do any retransmissions
X-Send-Pr-Version: 3.95

>Number:         3117
>Category:       lib
>Synopsis:       YP doesn't let RPC do any retransmissions
>Confidential:   no
>Severity:       non-critical
>Priority:       low
>Responsible:    thorpej
>State:          closed
>Class:          change-request
>Submitter-Id:   net
>Arrival-Date:   Fri Jan 17 03:20:00 +0000 1997
>Closed-Date:    Tue Jan 21 20:36:45 +0000 1997
>Last-Modified:  Tue Jan 21 20:41:42 +0000 1997
>Originator:     Michael Eriksson
>Release:        1.2
>Organization:
Ericsson Radio Systems AB
>Environment:
NetBSD 1.2 (and -current)
>Description:

YP sets the resend interval to the same value as the total RPC
timeout, which means that it won't let RPC do any retries before doing
a rebind. In NetBSD 1.1 the retry interval was half the total timeout,
i.e., RPC was allowed to do one retransmission. This is desirable, as
the most common reason for a time out (at least on the network where
my machine is :-)) is a dropped packet, not a crashed/overloaded YP
server.

>How-To-Repeat:

Code inspection? Also note the need for the recent _yplib_nerrs change
in YP.

>Fix:

This patch is for 1.2 as released (as that is what I run). I've set
RPC_RETRIES to 4, which is nice for our lossy network, but might be
too demanding for other environments.

*** yplib.c.orig	Sun May 26 08:13:58 1996
--- yplib.c	Fri Jan 17 10:22:48 1997
***************
*** 56,62 ****
  struct dom_binding *_ypbindlist;
  char _yp_domain[MAXHOSTNAMELEN];

! struct timeval _yplib_timeout = { 10, 0 };

  void _yp_unbind __P((struct dom_binding *));

--- 56,64 ----
  struct dom_binding *_ypbindlist;
  char _yp_domain[MAXHOSTNAMELEN];

! #define RPC_TIMEOUT	10
! #define RPC_RETRIES	4
! struct timeval _yplib_timeout = { RPC_TIMEOUT, 0 };

  void _yp_unbind __P((struct dom_binding *));

***************
*** 74,79 ****
--- 76,82 ----
  	CLIENT         *client;
  	int             new = 0, r;
  	int             count = 0;
+ 	struct timeval  tv;

  	if (dom == NULL || *dom == 0)
  		return YPERR_BADARGS;
***************
*** 219,227 ****
  	if (ysd->dom_client)
  		clnt_destroy(ysd->dom_client);
  	ysd->dom_socket = RPC_ANYSOCK;
  	ysd->dom_client = clntudp_create(&ysd->dom_server_addr,
! 				      YPPROG, YPVERS, _yplib_timeout,
! 				      &ysd->dom_socket);
  	if (ysd->dom_client == NULL) {
  		clnt_pcreateerror("clntudp_create");
  		ysd->dom_vers = -1;
--- 222,231 ----
  	if (ysd->dom_client)
  		clnt_destroy(ysd->dom_client);
  	ysd->dom_socket = RPC_ANYSOCK;
+ 	tv.tv_sec = RPC_TIMEOUT / RPC_RETRIES;
+ 	tv.tv_usec = 1000000 * (RPC_TIMEOUT % RPC_RETRIES) / RPC_RETRIES;
  	ysd->dom_client = clntudp_create(&ysd->dom_server_addr,
! 				      YPPROG, YPVERS, tv, &ysd->dom_socket);
  	if (ysd->dom_client == NULL) {
  		clnt_pcreateerror("clntudp_create");
  		ysd->dom_vers = -1;

>Release-Note:
>Audit-Trail:
State-Changed-From-To: open->analyzed 
State-Changed-By: thorpej 
State-Changed-When: Tue Jan 21 11:44:15 PST 1997 
State-Changed-Why:  
This is indeed a bug.  This patch won't work "as-is" since the -current 
libc has changed somewhat. 


Responsible-Changed-From-To: lib-bug-people->thorpej 
Responsible-Changed-By: thorpej 
Responsible-Changed-When: Tue Jan 21 11:44:15 PST 1997 
Responsible-Changed-Why:  
I'll take care of this. 
State-Changed-From-To: analyzed->closed 
State-Changed-By: thorpej 
State-Changed-When: Tue Jan 21 12:36:45 PST 1997 
State-Changed-Why:  
Suggested fix applied, slightly modified to compute the rpc timeout 
only once at compile-time, rather than N times at run-time. 

Thanks! 
>Unformatted:

Home	PR Database Search