[syslog-ng] patterndb context-timeout problem

[Balazs Scheidler]

On Fri, 2011-03-04 at 10:41 +0100, Zoltán Pallagi wrote:
> Hi,
> 
> I've found a problem in handling context-timeout in a special case.
> 
> If context-timeout=10 (10 sec) and I am using a trigger, where 
> trigger="timeout", it can happen that the timeout won't expire:
> 
> I sent the following two messages to syslog-ng before expiring timeout 
> (but the difference between the two timestamp is bigger than the timeout 
> value):
> 
> "<38>1990-01-01T14:45:25 customhostname program6[1234]: program6 
> testmessage"
> "<38>1990-01-01T14:55:25 customhostname program6[1234]: program6 
> testmessage"
> 
> I receive the timeout message immediately, when the second message 
> arrives (it's ok), but there is no timeout for the second message.

Thanks for the report, it was a really good catch. The unit tests didn't
find it, because it is a key in this bug to have the two timestamps so
much apart (more precisely, it needs to cross an 1024 second boundary to
be triggered).

This patch fixes it:

commit c3b5518d0c76740a43b192a24f5d9778f4e63282
Author: Balazs Scheidler <bazsi at balabit.hu>

    patterndb: fixed timerwheel bug that caused timers not to expire
    
    Reproduction:
    
      - set the current time to 1990-01-01T14:45:25
      - add a timer that expires 10 seconds from the above time
      - before that'd expire, set the current time to 1990-01-01T14:55:25
      - add a timer that expires 10 seconds from the 2nd date
    
    The cause was that base wasn't properly updated when we break out of the
    timer advancement loop in timer_wheel_set_time().
    
    Reported-By: Zoltan Pallagi <pzolee at balabit.hu>
    Signed-off-by: Balazs Scheidler <bazsi at balabit.hu>



-- 
Bazsi