On Mon, Feb 27, 2012 at 11:37 AM, Paolo Bonzini <pbonz...@redhat.com> wrote: > On 02/27/2012 10:23 AM, Stefan Hajnoczi wrote: >> > I think the right approach is to create a single blkmirror driver that >> > also includes blkverify functionality. The code is basically the same >> > except blkverify also compares reads - just use a flag to >> > enable/disable that behavior. >> > >> > Feel free to rename the blkverify driver to blkmirror if you wish. >> >> Federico: Any response to this? I still think blkmirror and blkverify >> do essentially the same thing and should be unified. > > Once non-incremental mode is added, I suspect blkmirror will diverge > from blkverify significantly. In particular, we would need to track > where have writes been done in the destination. We also would need to > hooks for block/stream.c, or perhaps a completely separate > implementation of streaming.
I'm not familiar with non-incremental mode, could you please explain it or link to it? > Also, blkverify doesn't support cancellation. I know we do quite poorly > in this area, but I'd rather not make it worse... That's why I suggested unifying them - take the best of both approaches. Stefan