Re: [Qemu-devel] [RFC 0/4] qemu-img: add max-size subcommand

[Nir Soffer]

On Sat, Mar 4, 2017 at 12:02 AM, John Snow <address@hidden> wrote:
>
>
> On 03/03/2017 04:38 PM, Nir Soffer wrote:
>> On Fri, Mar 3, 2017 at 3:51 PM, Stefan Hajnoczi <address@hidden> wrote:
>>>
>>> RFCv1:
>>>  * Publishing patch series with just raw support, no qcow2 yet.  Please 
>>> review
>>>    the command-line interface and let me know if you are happy with this
>>>    approach.
>>>
>>> Users and management tools sometimes need to know the size required for a 
>>> new
>>> disk image so that an LVM volume, SAN LUN, etc can be allocated ahead of 
>>> time.
>>> Image formats like qcow2 have non-trivial metadata that makes it hard to
>>> estimate the exact size without knowledge of file format internals.
>>>
>>> This patch series introduces a new qemu-img subcommand that calculates the
>>> required size for both image creation and conversion scenarios.
>>>
>>> The conversion scenario is:
>>>
>>>   $ qemu-img max-size -f raw -O qcow2 input.img
>>>   107374184448
>>
>> Isn't this the minimal size required to convert input.img?
>>
>
> It's an upper bound for the property being measured, which is current
> allocation size, not maximum potential size after growth.

>From my point of view, this is the minimal size you must allocate if you
want to convert the image to logical volume.

>
>>>
>>> Here an existing image file is taken and the output includes the space 
>>> required
>>> for data from the input image file.
>>>
>>> The creation scenario is:
>>>
>>>   $ qemu-img max-size -O qcow2 --size 5G
>>>   196688
>>
>> Again, this is the minimal size.
>>
>> So maybe use min-size?
>>
>> Or:
>>
>>     qemu-img measure -f raw -O qcow2 input.img
>>
>> Works nicely with other verbs like create, convert, check.
>>
>
> Measure what? This is strictly less descriptive even if "max-size" isn't
> a verb.

measure-size?

>> Now about the return value, do we want to return both the minimum size
>> and the maximum size?
>>
>> For ovirt use case, we currently calculate the maximum size by multiplying
>> by 1.1. We use this when doing automatic extending of ovirt thin provisioned
>> disk. We start with 1G lv, and extend it each time it becomes full, stopping
>> when we reach virtual size * 1.1. Using more accurate calculation instead
>> can be nicer.
>>
>> So we can retrun:
>>
>> {
>>     "min-size": 196688,
>>     "max-size": 5905580032
>> }
>>
>> Anyway thanks for working on this!
>>
>
> It sounds like you want something different from what was intuited by
> Maor Lipchuck. There are two things to estimate:
>
> (A) An estimate of the possible size of an image after conversion to a
> different format, and
> (B) An estimate of the possible size after full allocation.
>
> I got the sense that Maor was asking for (A), but perhaps I am wrong
> about that. However, both are "maximums" in different senses.

Both are minimum when you have to allocate the space.

Maor ask about A because he is working on fixing allocation when
converting existing files, but we also have other use cases like B.

Nir

>
> --js
>
>>>
>>> Stefan Hajnoczi (4):
>>>   block: add bdrv_max_size() API
>>>   raw-format: add bdrv_max_size() support
>>>   qemu-img: add max-size subcommand
>>>   iotests: add test 178 for qemu-img max-size
>>>
>>>  include/block/block.h      |   2 +
>>>  include/block/block_int.h  |   2 +
>>>  block.c                    |  37 +++++++++
>>>  block/raw-format.c         |  16 ++++
>>>  qemu-img.c                 | 196 
>>> +++++++++++++++++++++++++++++++++++++++++++++
>>>  qemu-img-cmds.hx           |   6 ++
>>>  tests/qemu-iotests/178     |  75 +++++++++++++++++
>>>  tests/qemu-iotests/178.out |  25 ++++++
>>>  tests/qemu-iotests/group   |   1 +
>>>  9 files changed, 360 insertions(+)
>>>  create mode 100755 tests/qemu-iotests/178
>>>  create mode 100644 tests/qemu-iotests/178.out
>>>
>>> --
>>> 2.9.3
>>>
>