Re: [Monotone-devel] Database gone wild...

[William Uther]

On 09/03/2007, at 4:58 PM, Nathaniel Smith wrote:

> > Then there was the problem with cvssync where it was moved to a new
> > server that didn't have a key generated before it was run the first
> > time.  Rather than erroring out, it added two revs without any certs
> > and the scripts then synced them into the cloud.  Wheeee.  A db check
> > now gives:
>
> This is really not a big deal.  You can add certs to those revs if you
> want; you can kill_rev_locally them if you want (they probably won't
> even be transmitted over netsync, because they have no branch certs,
> the exception is if they have children that do have branch certs),
> you can ignore them, whatever.  db check is just whining.

netsync (we're mostly using ssh:// here) will send them sometimes.   
This means that they propagate nicely, and *then* start causing  
problems.

> > The problem is that I need to stop people from syncing the old broken
> > stuff into the fixed up db.  I suspect I want to do something with
> > epochs.  If I 'mtn db set_epoch' on each branch to a random number,
> > that should solve the problem, right?
>
> Yes, that's how epoch's work -- mtn won't let data flow between
> databases with mismatched epochs.

Yup - thank you to those on irc who helped me out!

Be well,

Will       :-}