Re: [Qemu-devel] CoW image commit+shrink(= make_empty) support

[Supriya Kannery <supriyak@linux.vnet.ibm.com>]

On 06/14/2012 11:58 PM, Supriya Kannery wrote:
> On 06/14/2012 07:59 PM, Jeff Cody wrote:
>> On 06/14/2012 10:23 AM, Zhi Hui Li wrote:
>>> On 2012年06月11日 23:37, Jeff Cody wrote:
>>>> On 06/11/2012 10:24 AM, Stefan Hajnoczi wrote:
>>>>> On Mon, Jun 11, 2012 at 1:50 PM, Kevin Wolf<kwolf@redhat.com> wrote:
>>>>>> Am 11.06.2012 14:09, schrieb Stefan Hajnoczi:
>>>>>>> On Fri, Jun 8, 2012 at 6:46 PM, Jeff Cody<jcody@redhat.com> wrote:
>>>>>>>> On 06/08/2012 12:11 PM, Kevin Wolf wrote:
>>>>>>>>> Am 08.06.2012 16:32, schrieb Jeff Cody:
>>>>>>>>>> On 06/08/2012 09:53 AM, Stefan Hajnoczi wrote:
>>>>>>>>>>> On Fri, Jun 8, 2012 at 2:19 PM, Jeff Cody<jcody@redhat.com>
>>>>>>>>>>> wrote:
>>>>>>>>>>>> On 06/08/2012 08:42 AM, Stefan Hajnoczi wrote:
>>>>>>>>>>>>> Let's figure out how to specify block-commit so we're all
>>>>>>>>>>>>> happy, that
>>>>>>>>>>>>> way we can avoid duplicating work. Any comments on my notes
>>>>>>>>>>>>> above?
>>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> I think we are almost completely on the same page - devil is
>>>>>>>>>>>> in the
>>>>>>>>>>>> details, of course (for instance, on how to convert the
>>>>>>>>>>>> destination base
>>>>>>>>>>>> from r/o to r/w).
>>>>>>>>>>>
>>>>>>>>>>> Great. The atomic r/o -> r/w transition and the commit
>>>>>>>>>>> coroutine can
>>>>>>>>>>> be implemented on in parallel. Are you happy to implement the
>>>>>>>>>>> atomic
>>>>>>>>>>> r/o -> r/w transition since you wrote bdrv_append()? Then Zhi
>>>>>>>>>>> Hui can
>>>>>>>>>>> assume that part already works and focus on implementing the
>>>>>>>>>>> commit
>>>>>>>>>>> coroutine in the meantime. I'm just suggesting a way to split
>>>>>>>>>>> up the
>>>>>>>>>>> work, please let me know if you think this is good.
>>>>>>>>>>
>>>>>>>>>> I am happy to do it that way. I'll shift my focus to the
>>>>>>>>>> atomic image
>>>>>>>>>> reopen in r/w mode. I'll go ahead and post my diagrams and
>>>>>>>>>> other info
>>>>>>>>>> for block-commit on the wiki, because I don't think it
>>>>>>>>>> conflicts with we
>>>>>>>>>> discussed above (although I will modify my diagrams to not
>>>>>>>>>> show commit
>>>>>>>>>> from the top-level image). Of course, feel free to change it as
>>>>>>>>>> necessary.
>>>>>>>>>
>>>>>>>>> I may have mentioned it before, but just in case: I think
>>>>>>>>> Supriya's
>>>>>>>>> bdrv_reopen() patches are a good starting point. I don't know
>>>>>>>>> why they
>>>>>>>>> were never completed, but I think we all agreed on the general
>>>>>>>>> design,
>>>>>>>>> so it should be possible to pick that up.
>>>>>>>>>
>>>>>>>>> Though if you have already started with your own work on it,
>>>>>>>>> Jeff, I
>>>>>>>>> expect that it won't be much different because it's basically
>>>>>>>>> the same
>>>>>>>>> transactional approach that you know and that we already used
>>>>>>>>> for group
>>>>>>>>> snapshots.
>>>>>>>>>
>>>>>>>>
>>>>>>>> I will definitely use parts of Supriya's as it makes sense - what I
>>>>>>>> started work on is similar to bdrv_append() and the current
>>>>>>>> transaction
>>>>>>>> approach, so there will be plenty in common to reuse, even with
>>>>>>>> some
>>>>>>>> differences.
>>>>>>>
>>>>>>> I have CCed Supriya who has been working on the reopen patch series.
>>>>>>> She is close to posting a new version.
>>>>>>
>>>>>> It's just a bit disappointing that it takes several months between
>>>>>> each
>>>>>> two versions of the patch series. We'd like to have this in qemu 1.2,
>>>>>> not only in qemu 1.14.
>>>>>>
>>>>>> I can understand if someone can't find the time, but then allow at
>>>>>> least
>>>>>> someone else to pick it up.
>>>>>
>>>>> Hey, don't shoot the messenger :). I just wanted add Supriya to CC so
>>>>> she can join the discussion and see how much overlap there is with
>>>>> what she's doing. We all contribute to QEMU from different angles and
>>>>> with different priorities. If there is a time critical dependency on
>>>>> the reopen code then discuss it here and the result will be that
>>>>> someone officially drives the feature on.
>>>>>
>>>>
>>>> I am more than happy to take the previous reopen() patches, and drive
>>>> those forward, and also do whatever else is needed for live block
>>>> commit.
>>>>
>>>> It sounds like Zhi Hui is working on live block commit patches, and
>>>> Supriya is working on the bdrv_reopen() portion - I don't want to
>>>> duplicate any effort, but if there is anything I can do to help with
>>>> either of those areas, just let me know.
>>>>
>>>> Thanks,
>>>> Jeff
>>>>
>>>>
>>>>
>>> Jeff please go ahead with block-commit, I
>>> am finishing up my fdc async conversion patch series first. I will
>>> reply when I'm ready to work on block-commit.
>>>
>>> Thank you very much!
>>>
>>
>> Hi Zhi,
>>
>> I will do that. Thanks!
>>
>> Jeff
>>
> Jeff,
> Need another day's time for posting bdrv_reopen() patchset. Getting few
> hmp related build errors in hostcache toggling code when built on latest
> git. Will correct those and post the patchset soon.
> - Supriya
>
>
Jeff,
   bdrv_reopen v1 patchset posted. Title: "Dynamic host pagecache change
and image file reopen".
Thanks, Supriya