On Thu, Apr 26, 2018 at 5:58 AM, Bo Berglund <bo.bergl...@gmail.com> wrote: > On Thu, 26 Apr 2018 11:26:38 +0200, OBones <obo...@free.fr> wrote: > >>Well, maybe I read this wrong, but the manual talks about a master and a >>slave, which are tightly related. If I remove the master once the slave >>is fully ready to take over its role, I'm not sure it will work >>smoothly, hence my question. > > I just set up an svnsync system a few weeks back and my read is that > the two repositories are basically mirror images of each other, but > that the mirror should never be committed to. > (Except for svnsync of course, it will shoot the commits performed on > the master to the mirror thus replicating the state)
The mirror should never be committed to, until and unless any planned migration is complete and the original master is disabled. > But if you decide to take the master off-line at a later time and run > with the mirror instead, then what you need to do is modify the hooks > that are present on the mirror to disallow commits except from svnsync > and whatever sync related hooks you have in place. > After that the mirror should act as a regular repository, but you can > no longer sync with the original because of colliding revision > numbers. Bingo. It is, theoretically, possible to individually bring commits done directly on the slave back over to the original master to resolve such a resulting split brain. But it's safer to set up a slave of the slave, rely on that for failover, and shoot the original master server through the head. Nico Kadel-Garcia <nka...@gmail.com> > > -- > Bo Berglund > Developer in Sweden >
