On Thu, Jun 13, 2019 at 3:23 PM Eric Mesa <eric@xxxxxxxxxxxx> wrote: > > On Thursday, June 13, 2019 2:26:10 AM EDT Andrei Borzenkov wrote: > > > > All your snapshots on source have the same received_uuid (I have no idea > > how is it possible). If received_uuid exists, it is sent to destination > > instead of subvolume UUID to identify matching snapshot. All your backup > > sbapshots on destination also have the same received_uuid which is > > matched against (received_)UUID of source subvolume. In this case > > receive command takes the first found subvolume (probably the most > > recent, i.e. with the smallest generation number). So you send > > differential stream against one subvolume and this stream is applied to > > another subvolume which explains the error. > > Yup. Any idea of how to fix? Maybe 'cp -a --relink' into a new subvolume on the source. It won't complete immediately like a snapshot. But it'll complete way faster than a normal data copy. From that point, this new subvolume becomes "master" where all changes occur, and all the other snapshots are made from it. Also, you should do incrementals between the most recent two snapshots. i.e. the snapshot that follows -p should be the snapshot most recently successfully received on the destination. That represents the least amount of increment needed to update the destination. It will work the way you're doing it now, with -p being an older snapshot, but you're unnecessarily sending a lot more data every time. -- Chris Murphy
