libvirt users - Jun 2016 - Re: virt-sparsify changing the apparent-size of files

Hello, 

I am curious why when using virt-sparsify the apparent-size of the file
gets reduced to the actual file size?  Is there a way to get the
apparent-size left as the full size?

In the man page for virt-sparsify it has a section on
"IMPORTANT NOTE ABOUT SPARSE OUTPUT IMAGES" and its output seems to
indicate what I would expect where "ls -lh" shows the full apparent
size
and "du -sh" shows the actual size.  But my actual results from
testing
are different.

In the following example input.qcow2 is before sparsify and output.qcow2
is after.

Doing ls reports full size for input but a small size for
the output:
root@testingbox:/tmp/test # ls -lh
total 3.1G
-rw-r--r-- 1 root root 512M Jun 14 22:30 file.img
-rw-r--r-- 1 root root  51G Jun 14 22:30 input.qcow2
-rw-r--r-- 1 root root 790M Jun 14 22:36 output.qcow2

Same when asking du for apparent-size:
root@testingbox:/tmp/test # du -h --apparent-size *
512M    file.img
51G     input.qcow2
790M    output.qcow2

A normal du will show the actual size of the files:
root@testingbox:/tmp/test # du -h *
0       file.img
2.4G    input.qcow2
790M    output.qcow2

The only way I have found to check the full virtual size of the file
after virt-sparsify is with qemu-img:
root@testingbox:/tmp/test # qemu-img info input.qcow2
image: input.qcow2
file format: qcow2
virtual size: 50G (53687091200 bytes)
disk size: 2.3G
cluster_size: 65536
Format specific information:
    compat: 1.1
    lazy refcounts: true
root@testingbox:/tmp/test # qemu-img info output.qcow2
image: output.qcow2
file format: qcow2
virtual size: 50G (53687091200 bytes)
disk size: 789M
cluster_size: 65536
Format specific information:
    compat: 1.1
    lazy refcounts: false


Thanks!

On Wed, Jun 15, 2016 at 10:20:54AM -0500, libvirt_users@skagitattic.com
wrote:
> Hello, 
> 
> I am curious why when using virt-sparsify the apparent-size of the file
> gets reduced to the actual file size?  Is there a way to get the
> apparent-size left as the full size?
> 
> In the man page for virt-sparsify it has a section on
> "IMPORTANT NOTE ABOUT SPARSE OUTPUT IMAGES" and its output seems
to
> indicate what I would expect where "ls -lh" shows the full
apparent size
> and "du -sh" shows the actual size.  But my actual results from
testing
> are different.
> 
> In the following example input.qcow2 is before sparsify and output.qcow2
> is after.
> 
> Doing ls reports full size for input but a small size for
> the output:
> root@testingbox:/tmp/test # ls -lh
> total 3.1G
> -rw-r--r-- 1 root root 512M Jun 14 22:30 file.img
> -rw-r--r-- 1 root root  51G Jun 14 22:30 input.qcow2
> -rw-r--r-- 1 root root 790M Jun 14 22:36 output.qcow2
What is file.img?
> Same when asking du for apparent-size:
> root@testingbox:/tmp/test # du -h --apparent-size *
> 512M    file.img
> 51G     input.qcow2
> 790M    output.qcow2
'du --apparent-size' and 'ls -l' are the same thing, so it's
not
really surprising that you see the same numbers here.
> A normal du will show the actual size of the files:
> root@testingbox:/tmp/test # du -h *
> 0       file.img
> 2.4G    input.qcow2
> 790M    output.qcow2
> 
> The only way I have found to check the full virtual size of the file
> after virt-sparsify is with qemu-img:
> root@testingbox:/tmp/test # qemu-img info input.qcow2
> image: input.qcow2
> file format: qcow2
> virtual size: 50G (53687091200 bytes)
> disk size: 2.3G
> cluster_size: 65536
> Format specific information:
>     compat: 1.1
>     lazy refcounts: true
> root@testingbox:/tmp/test # qemu-img info output.qcow2
> image: output.qcow2
> file format: qcow2
> virtual size: 50G (53687091200 bytes)
> disk size: 789M
> cluster_size: 65536
> Format specific information:
>     compat: 1.1
>     lazy refcounts: false
If I understand your question correctly, then everything is working as
it should.

For qcow2 files, the virtual size is not related to the "apparent"
size, as you can easily prove:

$ qemu-img create -f qcow2 huge.qcow2 1T
Formatting 'huge.qcow2', fmt=qcow2 size=1099511627776 encryption=off
cluster_size=65536 lazy_refcounts=off refcount_bits=16
$ ls -lh huge.qcow2
-rw-r--r--. 1 rjones rjones 208K Jun 17 09:09 huge.qcow2

If you use raw files, then the file apparent size and virtual size are
the same.

Rich.

-- 
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
Read my programming and virtualization blog: http://rwmj.wordpress.com
libguestfs lets you edit virtual machines.  Supports shell scripting,
bindings from many languages.  http://libguestfs.org

Hello Rich,

Thanks for the reply and information.
> What is file.img?
Its simply a file I made while testing sparse files "truncate -s 512M
file.img".
> 'du --apparent-size' and 'ls -l' are the same thing, so
it's not
> really surprising that you see the same numbers here.
Indeed, I expected those to be the same, just figured I would show them
both for completeness.
> For qcow2 files, the virtual size is not related to the
"apparent"
> size, as you can easily prove:
So you are saying when I do the virt-sparsify its converting the image
from raw to cow2?

I studied the man page for virt-sparsify and tried again with
the flag "--format raw" (output2.cow2).  This output files looks as I
first expected.

# ls -lrh
total 4.7G
-rw-r--r-- 1 root root 790M Jun 14 22:36 output.qcow2
-rw-r--r-- 1 root root  51G Jun 21 18:34 output2.qcow2
-rw-r--r-- 1 root root  51G Jun 14 22:30 input.qcow2
-rw-r--r-- 1 root root 512M Jun 14 22:30 file.img
# du -sh *
0       file.img
2.4G    input.qcow2
1.6G    output2.qcow2
790M    output.qcow2

So I guess it was detecting the source wrong and doing a conversion?
(As it says "If this is not specified, then the input format is used."
in the man page)

Also seems odd that it is 1.6G rather then the 790M of the output in
qcow2 format.  Does qcow2 do some compression or something?


Is there some way I can see if a file is raw or cow2?  When I did
"qemu-img info" it reported both of them as cow2.

I ran the initial install like the following and it ended up
with the sparse file that reports full size with ls: "virt-install
--name test-install --memory 512 -l
'http://ftp.nl.debian.org/debian/dists/jessie/main/installer-amd64/'
--disk size=50 --nographics -x "console=ttyS0"" 

I tried again and got the same result as the initial input file.  So
why is it naming it qcow2 if its actually raw?  Or is there two
way's of doing formatting inside the qcow2 file, one raw and the other
called something else?

Is there any reason to use raw vs non-raw?  From some reading online it
appears performance is better with raw files.  Is there a downside to
using raw?


Thanks!
Keelan



On Fri, 17 Jun 2016 09:11:04 +0100
"Richard W.M. Jones" <rjones@redhat.com> wrote:
> On Wed, Jun 15, 2016 at 10:20:54AM -0500,
> libvirt_users@skagitattic.com wrote:
> > Hello, 
> > 
> > I am curious why when using virt-sparsify the apparent-size of the
> > file gets reduced to the actual file size?  Is there a way to get
> > the apparent-size left as the full size?
> > 
> > In the man page for virt-sparsify it has a section on
> > "IMPORTANT NOTE ABOUT SPARSE OUTPUT IMAGES" and its output
seems to
> > indicate what I would expect where "ls -lh" shows the full
apparent
> > size and "du -sh" shows the actual size.  But my actual
results
> > from testing are different.
> > 
> > In the following example input.qcow2 is before sparsify and
> > output.qcow2 is after.
> > 
> > Doing ls reports full size for input but a small size for
> > the output:
> > root@testingbox:/tmp/test # ls -lh
> > total 3.1G
> > -rw-r--r-- 1 root root 512M Jun 14 22:30 file.img
> > -rw-r--r-- 1 root root  51G Jun 14 22:30 input.qcow2
> > -rw-r--r-- 1 root root 790M Jun 14 22:36 output.qcow2  
> 
> What is file.img?
> 
> > Same when asking du for apparent-size:
> > root@testingbox:/tmp/test # du -h --apparent-size *
> > 512M    file.img
> > 51G     input.qcow2
> > 790M    output.qcow2  
> 
> 'du --apparent-size' and 'ls -l' are the same thing, so
it's not
> really surprising that you see the same numbers here.
> 
> > A normal du will show the actual size of the files:
> > root@testingbox:/tmp/test # du -h *
> > 0       file.img
> > 2.4G    input.qcow2
> > 790M    output.qcow2
> > 
> > The only way I have found to check the full virtual size of the file
> > after virt-sparsify is with qemu-img:
> > root@testingbox:/tmp/test # qemu-img info input.qcow2
> > image: input.qcow2
> > file format: qcow2
> > virtual size: 50G (53687091200 bytes)
> > disk size: 2.3G
> > cluster_size: 65536
> > Format specific information:
> >     compat: 1.1
> >     lazy refcounts: true
> > root@testingbox:/tmp/test # qemu-img info output.qcow2
> > image: output.qcow2
> > file format: qcow2
> > virtual size: 50G (53687091200 bytes)
> > disk size: 789M
> > cluster_size: 65536
> > Format specific information:
> >     compat: 1.1
> >     lazy refcounts: false  
> 
> If I understand your question correctly, then everything is working as
> it should.
> 
> For qcow2 files, the virtual size is not related to the
"apparent"
> size, as you can easily prove:
> 
> $ qemu-img create -f qcow2 huge.qcow2 1T
> Formatting 'huge.qcow2', fmt=qcow2 size=1099511627776
encryption=off
> cluster_size=65536 lazy_refcounts=off refcount_bits=16 $ ls -lh
> huge.qcow2 -rw-r--r--. 1 rjones rjones 208K Jun 17 09:09 huge.qcow2
> 
> If you use raw files, then the file apparent size and virtual size are
> the same.
> 
> Rich.
>