Bug#399930: logrotation race condition with exim writing to logs

[Marc Haber]

On Fri, Nov 24, 2006 at 11:26:33PM +1100, CaT wrote:
> On Fri, Nov 24, 2006 at 01:15:01PM +0100, Marc Haber wrote:
> > On Fri, Nov 24, 2006 at 10:58:28PM +1100, CaT wrote:
> > > On Fri, Nov 24, 2006 at 10:23:38AM +0100, Marc Haber wrote:
> > > > On Fri, Nov 24, 2006 at 05:11:25PM +1100, CaT wrote:
> > > > > So there is absolutely no need for the create option in logrotate
> > > > 
> > > > Not having the logs created might break monitoring mechanisms.
> > > 
> > > Well it is kind of a choice between maybe breaking poorly written
> > > monitoring schemes (the lack of a log should merely be a hint to
> > > check deeper rather then be the be-all and end-all of ones decision
> > > making) and definately breaking the very system that they are meant
> > > to monitor.
> > 
> > I consider this race condition to be minor, since it obviously only
> > happens on very heavily loaded systems, which should have a competent
> 
> Well not really. It's just more likely to happen on a heavily loaded
> system due to the increased number of msgs it processes. It can happen
> with any system. It only takes one message to come in at the right time.

I know. It happens rarely. This setup has been in place for years,
even with exim 3.

> > system admin around to isolate the issue. You are always free to
> 
> And an on-call monkey who actually gives you a call. :/ As with all
> problems that cause tears they're mostlikely to occur when you're
> asleep.

That's your personal disfortune. How much are you paying for this MTA
software?

> > change the configuration yourself, /etc/logrotate.d/exim4-base is a
> > dpkg-conffile.
> 
> I'm already bulding my own debs for this, which is annoying as I'm not
> doing this to add features but to fix some brokenness.

You are building your own debs to change a single dpkg-conffile?
Excuse me? Are you aware what "foo is a dpkg-conffile" means?

> The thing is, there's no need to precreate the logfile (broken
> monitoring systems aside) and doing so can break things and has
> broken things. I'd say the opposite tact should be taken. If you are
> relying on a monitoring system that relies on the logfile always
> being there then have it pre-created and deal with the race condition of
> the mintor slot where there is no logfile between the move of the old
> and the creation of the new.
> 
> The priority should be to make sure the things the exim package setup
> do not cause exim to die. Other things should take care of themselves.

I do not intend to change logrotate configuration at this time of the
release process.

If you disagree, please take this to debian-devel or the tech ctte.

Greetings
Marc

-- 
-----------------------------------------------------------------------------
Marc Haber         | "I don't trust Computers. They | Mailadresse im Header
Mannheim, Germany  |  lose things."    Winona Ryder | Fon: *49 621 72739834
Nordisch by Nature |  How to make an American Quilt | Fax: *49 621 72739835