Hello,

  I actually meant to add a pid and per-pid sequence to the
setup you put forth, so something like:

  [YYYY][MM][DD][hh][mm][SS][ss][server-id][pid][pid-seq]

All mapped to bits of course, as you only had 32 bits to
put all this in (iirc).  The scenario I was working to avoid
was 2 messages being delivered within the same ms on the same
machine (ie. same "server-id").  As you can have multiple
processes on the same machine delivering simultaneously, using
the pid will in practice likely take care of the problem, but
you could throw a per-pid sequence in to just to be safe.
I went through this same issue when messing with the pop3
unique_id stuff a year or so back (although I probably wouldn't
have noticed the collisions if I had used ms resolution :).

Jn


---- Original Message ----
From: Kevin Baker <[email protected]>
To: "DBMAIL Developers Mailinglist" <[email protected]>
Subject: Re: [Dbmail-dev] GUID vs Auto-incrementing... dev?
Sent: Thu, 2 Sep 2004 09:48:58 -0700 (PDT)

> I'd be concerned about using the pid. The pid on each
> machine could be padded to different lengths causing the
> numbers to be non-ascending between servers.
> 
> With a machine id set in the dbmail conf we'd have control
> over this.
> 
> Example with pid:
> 
> [YYYY][MM][DD][hh][mm][SS][ss][pid]
> 
> if you had two messages arrive at the same time with a pid
> of 123 on machine one and 95432 on machine two  they would
> be out of order and nonsequential. In addition it is very
> unlikely, but the pids could be the same making the id
> non-unique, thus non rfc.
> 
> 
> The other way:
> [YYYY][MM][DD][hh][mm][SS][ss][server-id]
> 
> server one 0001
> server two 0002
> 
> So long as the ids are padded to the same length across
> machines in a cluster the number would always be ascending
> and unique. If messages did happen to come in at the same
> time they would never collide, because of the padded
> server-id suffix.
> 
> [external-ip][YYYY][MM][DD][hh][mm][SS][ss][server-id]
> 
> I like the idea of possibly adding a prefix to the id
> though.. something unique to the email solution such as
> the clusters external ip... this could also be set in the
> conf pulled from the message header. This would actually
> make the ids unique globally, while keeping the ascending
> order. However by RFC they only need to be unique to the
> mailbox(folder).
> 
> 
> - Kevin
> 
> 
> 
> 
> > I don't think so.. I'm pretty sure that the email client
> > will just refresh the mailbox it is displaying.. messages
> > are then sorted by message-id or other if in the view
> > settings.
> >
> > We have had this happen in our current configuration cyrus
> > as messages can sometimes get caught up in amavis running
> > through filters for a second or so... they just appear and
> > everthing is fine...
> >
> >
> >
> >
> > <snip long text>
> >>   I'm not at all familiar with IMAP, would that cause
> >> any
> >> problems with imap clients, having "old" unique-id's
> >> popping
> >   <snip>
> >> "new" messages have what appears to be out of sequence
> >
> > _______________________________________________
> > Dbmail-dev mailing list
> > [email protected]
> > http://twister.fastxs.net/mailman/listinfo/dbmail-dev
> >
> 
> _______________________________________________
> Dbmail-dev mailing list
> [email protected]
> http://twister.fastxs.net/mailman/listinfo/dbmail-dev
> 
-- End Original Message --


--
Jesse Norell

[EMAIL PROTECTED] is not my email address;
change "administrator" to my first name.
--

Reply via email to