nova-feature-parity team mailing list archive

[Vishvananda Ishaya <vishvananda@xxxxxxxxx>]

Perhaps raw was a bad word: non-copy-on-write.  It could still be a qcow or a vhd etc.  Right now we cache all images.  There is a blueprint or two regarding exposing cache mgmt to the admin api iirc.

Vish

On Nov 8, 2011, at 9:46 PM, Paul Voccio wrote:

> Notes below:
> 
> From: Vishvananda Ishaya <vishvananda@xxxxxxxxx>
> Date: Tue, 8 Nov 2011 21:32:00 -0800
> To: Paul Voccio <paul.voccio@xxxxxxxxxxxxx>
> Cc: Ewan Mellor <Ewan.Mellor@xxxxxxxxxxxxx>, "nova-feature-parity@xxxxxxxxxxxxxxxxxxx" <nova-feature-parity@xxxxxxxxxxxxxxxxxxx>, Devdeep Singh <devdeep.singh@xxxxxxxxxx>, Likitha Shetty <likitha.shetty@xxxxxxxxxx>
> Subject: Re: Thin-cloning of disks for XenServer
> 
>> FYI on the kvm side, we use the --use_cow_images flag to toggle on this behavior, so someone who sets --nouse_cow_images will get a raw image.  That said, we don't redownload the image from glance every time.  We keep a local cache of the base image and do a straight file copy instead of a cow snapshot if the flag is false. It would be great to use the same flag for xen.
> 
> 
> Ewan, forgive my ignorance, but does XenServer support raw images? I didn't see anything about it in a cursory google search. Maybe its late and my g-foo is off. 
> 
> Vish – would this only be for base images? Would/Should we cache customer images that are being restored? If we do customer images, we would need to set a configurable cache size and location and figure out purging and rotation. 
> 
> 
>> 
>> Potential future changes:
>> a) provide two flags, one for root drive, and one for local ephemeral storage
> 
> 
>> b) allow this to be set per instance instead of globally (perhaps keyed off of instance type)
> 
> 
> Would this really even affect the user? Why would they get a choice if this is really to help an operator?
> 
>> c) expose it to the user? (allow them to specify whether they want cow images or not)
> 
> Same as above
> 
> 
>> 
>> 
>> Vish
>> 
>> On Nov 8, 2011, at 8:53 PM, Paul Voccio wrote:
>> 
>>> Ewan,
>>> 
>>> I do think this is a needed feature and I would welcome any code that would eliminate bottlenecks within the system. 
>>> 
>>> Would this be optional or a fundamental change to how the images work? There are some assumptions within the code that a vm owns the entire vdi tree. I have a feeling the code could get messy if it is optional, but would create some bugs the short term while we sort this all out. However the approach, I do feel this is the right way to go. 
>>> 
>>> I spoke to some people who had the concern that there may be hotspots on the disk if we get too many vdis pointing to the same parent vdi. I don't yet share concern for this issue, but it is something to be aware of. 
>>> 
>>> I look forward to seeing what has been done so far. 
>>> 
>>> pvo
>>> 
>>> From: Ewan Mellor <Ewan.Mellor@xxxxxxxxxxxxx>
>>> Date: Wed, 9 Nov 2011 04:24:23 +0000
>>> To: "nova-feature-parity@xxxxxxxxxxxxxxxxxxx" <nova-feature-parity@xxxxxxxxxxxxxxxxxxx>, Paul Voccio <paul.voccio@xxxxxxxxxxxxx>, Vishvananda Ishaya <vishvananda@xxxxxxxxx>
>>> Cc: Devdeep Singh <devdeep.singh@xxxxxxxxxx>, Likitha Shetty <likitha.shetty@xxxxxxxxxx>
>>> Subject: Thin-cloning of disks for XenServer
>>> 
>>>> Hi,
>>>>  
>>>> I’d like to introduce Devdeep Singh and Likitha Shetty (both cc’d) from the Citrix OpenStack team in Bengaluru.  They are starting work on a feature for XenServer where the VM disk is streamed fromGlance the first time, and then future requests for that image are thin-cloned from the original, rather than streaming the image a second time from Glance.  This is something that is already implemented for KVM, so I presume that there is interest from the nova-feature-parity team.
>>>>  
>>>> As far as I can see there is no blueprint for this yet, so Devdeep and Likitha will start by publishing what they’ve got so far.  Your feedback is welcome.
>>>>  
>>>> Pvo, in particular I remember at the last summit that you had strong views on how this should work.  Any insight you have would be appreciated.
>>>>  
>>>> If there are any problems with the timezone difference to India then please let me know – I’ll address any communication issues.
>>>>  
>>>> Thanks,
>>>>  
>>>> Ewan.
>>>>  
>>