singleinstancestore obsolete?

[Ramprasad]

On Wed, 2011-10-19 at 13:38 +0200, Wolfgang Breyha wrote:
> Bron Gondwana wrote, on 19.10.2011 12:27:
> > On Wednesday, October 19, 2011 12:18 PM, "Wolfgang Breyha" <wbreyha at gmx.net> wrote:
> >> Hi!
> >> 
> >> Is singleinstancestore obsolete/useless in 2.4.(12)?
> >> 
> >> This option is read into "singleinstance" in lmtpd.c and nntpd.c, but never
> >> used afterwards? Or did I miss something searching the source?
> > 
> > Hmm...
> > 
> >         r = append_fromstage(&as, &content->body, stage, 0,
> >                              (const char **) flag, nflags, !singleinstance);
> > 
> > Looks to me like you missed something.
> 
> Obviously;-) Sorry for that.
> 
> The main reason I was looking into these things was, that I'm searching for
> the best way to migrate a 2.3.16 backend with ~4TB spools and ~80k users to
> 2.4. Unlike my other backends this one profits pretty much from
> singleinstancestore (many massmailings) and I don't want to loose these
> savings if possible.
> 
> Simply upgrading the backend is impossible. The index upgrades will kill this
> machine.
> 
> Since replication is not a possible way from 2.3.16 to 2.4 I tried to figure
> out if a simple XFER takes care about that. But it doesn't as far as I can tell.
> 
> Do you know of a possible way to do that? Even if it's not trivial in the
> first place.
> 
> Greetings, Wolfgang

I think , writing a standalone index upgrade utility , like the ipurge ,
seems to be a reasonable thing to do 


If there was a light enough index upgrade possible ( only for inboxes ..
not subfolders ) Then I could stop cyrus , fork probably around 100
parallel upgrades take a 2-3 hour downtime and then start services again