> > > > (Sam)
> > > > > Well, hell, you'll probably lose mail even if you're running a
> > > > > journaling filesystem...
> > > >
> > > (me)
> > > > False. Mail will not be lost if if rename() or link() (depending on
> > >
> > (Sam)
> > > Who said anything about the message already being on the filesystem?
> >
> (me)
> > Then your comment was utterly inane. Any MTA which returns success
> > before writing a message to the filesystem, and syncing it...
>
(Sam)
> Which only syncs the data. close() then updates the metadata, which
> may remain buffered for some time before getting flushed out.
Which brings us back to link() or rename(), of course. Writing data
directly to the target queue/file is a mistake, unless you want
queued, incomplete messages after a failure.
Is it late at night where you are? I'm certain you know all this; you
must be tired.
Len.
--
Look at it this way: sendmail is a whale, and qmail is a shark. Perhaps
you're impressed by the size of the whale; perhaps, if you grew up
surrounded by whales, you find it hard to imagine a big sea creature
without tons of blubber.
-- Dan Bernstein
