Re: [Qemu-block] [PATCH 08/21] backup: skip unallocated clusters for full mode

[Fam Zheng]

On Tue, 01/24 13:13, Vladimir Sementsov-Ogievskiy wrote:
> 24.01.2017 12:36, Fam Zheng wrote:
> > On Tue, 01/24 12:18, Vladimir Sementsov-Ogievskiy wrote:
> > > 24.01.2017 10:59, Fam Zheng wrote:
> > > > On Fri, 12/23 17:28, Vladimir Sementsov-Ogievskiy wrote:
> > > > > diff --git a/tests/qemu-iotests/055 b/tests/qemu-iotests/055
> > > > > index 1d3fd04..388b7b2 100755
> > > > > --- a/tests/qemu-iotests/055
> > > > > +++ b/tests/qemu-iotests/055
> > > > > @@ -30,6 +30,7 @@ target_img = os.path.join(iotests.test_dir, 
> > > > > 'target.img')
> > > > >    blockdev_target_img = os.path.join(iotests.test_dir, 
> > > > > 'blockdev-target.img')
> > > > >    image_len = 64 * 1024 * 1024 # MB
> > > > > +pause_write = '3M'
> > > > >    def setUpModule():
> > > > >        qemu_img('create', '-f', iotests.imgfmt, test_img, 
> > > > > str(image_len))
> > > > > @@ -39,6 +40,7 @@ def setUpModule():
> > > > >        qemu_io('-f', iotests.imgfmt, '-c', 'write -P0xd5 1M 32k', 
> > > > > test_img)
> > > > >        qemu_io('-f', iotests.imgfmt, '-c', 'write -P0xdc 32M 124k', 
> > > > > test_img)
> > > > >        qemu_io('-f', iotests.imgfmt, '-c', 'write -P0x33 67043328 
> > > > > 64k', test_img)
> > > > > +    qemu_io('-f', iotests.imgfmt, '-c', 'write -P0xdc 40M ' + 
> > > > > pause_write, test_img)
> > > > What does this iotest change do?
> > > Without this backup block-job finishes before the next query to it from 
> > > test
> > > and test fails. This is a wide problem I suffer of on the way of backup
> > > improvement and it should have better solution then adjusting amount of 
> > > data
> > > being copied...
> > You can use blkdebug to pause I/O before querying the block job, or simply
> > throttle it down.
> > 
> > Fam
> 
> throttling down is not better I thing then just copying additional 3M, it is
> environment-dependent hack too. And we can't use throttling in all tests -
> throttling should be itself tested separately.

Then maybe opt to blkdebug?

I didn't mean the speed parameter of block jobs, but the block_set_io_throttle.
It is not perfect, but still better than simply enlarging the data size, and
is slightly simpler than blkdebug pause/resume.

Fam