Re: system/3808: select() does not timeout correctly

["Peter Galbavy" <peter_(_dot_)_galbavy_(_at_)_knowtion_(_dot_)_net>]

The following reply was made to PR system/3808; it has been noted by GNATS.

From: "Peter Galbavy" <peter_(_dot_)_galbavy_(_at_)_knowtion_(_dot_)_net>
To: <rogerw_(_at_)_pccorp_(_dot_)_com>, <gnats_(_at_)_openbsd_(_dot_)_org>
Cc:  
Subject: Re: system/3808: select() does not timeout correctly
Date: Fri, 4 Jun 2004 11:09:18 +0100

 rogerw_(_at_)_pccorp_(_dot_)_com wrote:
 >> Description:
 > I am using the postfix MTA (compiled form source and it is timeing out
 >         too quickly when a reverse DNS lookup fails. It timesout in
 >         approx 13 seconds instead of the specified 300 seconds. I
 >         posted this on the postfix list and the author stated
 >         "Postfix implements read/write timeouts with select(). You
 >         may want to capture a system call trace to see why select()
 > returns in less than 300 seconds."
 
 This sounds fishy. I don't know the postfix codebase at all, and so I can't
 say how it is doing reverse lookups, using the system resolver libraries or
 it's own code. select(2) can return "early" for a number of reasons in
 addition to the specified descriptors becoming ready. I would suggest
 grubbing around in the postfix code, extract the code that it having the
 problem and setting up test cases. Then, if the problem is reproducable and
 is NOT the postfix code doing something wrong, post the test code against
 this report and we can then see what might be wrong.
 
 Peter