On Wed, May 05, 2010 at 01:17:45PM -0400, Paul Smith wrote: > On Wed, 2010-05-05 at 01:56 -0700, Karl Fogel wrote: > > Karl Fogel <[email protected]> writes: > > >>I'm personaly not so sure about asking people to re-push and such, but > > >>up to you. > > > > > >Well, I'll have to work out the exact recipe, but if people's > > >already-pushed changes are going to disappear, they need to be warned > > >and informed of how to fix it. I may be able to avoid that by doing the > > >pushes for them, we'll see. > > > > Turns out we shouldn't have to deal with any re-pushes -- the entire bzr > > repositories all together are around 1.3GB. We can just go read-only > > long enough to copy that amount over. > > For 1.3G it's probably not worth it, but for migrating large data sets I > have found rsync to be invaluable. > > What I do is rsync to the new system while everything is still online > and running: for this rsync you don't care whether the result is > coherent or not. You can continue to do this every so often right up > until the time you want to move (plus this gives you a good feel for how > long the eventual downtime will be). Then when you're ready you shut > down write access to the data and do a final rsync, which will be very > quick typically since it's only copying the changes since the last > rsync. > > Then bring up the service on the new system. > > This lets you lock the service from write, and avoid problems with > changes going missing, while still having a quick transition rather than > waiting for all data to copy across. > > > WFM! Thanks for all your work on Savannah, folks...
This is my technique of choice as well, I recommend it :) Also: Karl, you'll want to test the creation of new bzr repositories from the Savane interface in the new environment. -- Sylvain
