incubator-vcl-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Xianqing Yu" <yu267155...@hotmail.com>
Subject Re: The New KVM Provisioning module
Date Tue, 16 Aug 2011 19:13:00 GMT

OK. Let's use both APIs.

I have to clarify one thing. Although qemu-kvm use -snapshot option. But it
is not the snapshot you are thinking about. You don't have to create any
snapshot for the images. I only need to add this option to command line. For
example, if I want one VM boot on the image A and don't modify anything on
the image A, I just simply send this command "qemu-kvm -m 512
imagefile -snapshot -net, network configuartion". So it is actually the same
as non-persistent disk option in VMware, and I think that the name
"snapshot" is very easy to misunderstand here.

About the vmdk on KVM, I think it is an option. The users need to define
installtype in VCL database. If it is vmware, KVM module would consider the
image file is vmdk type 4. If it is kvm, KVM module would use img instead.
So users can decide what they want to use. And by supporting vmdk file, all
current vmdk image running on the VMware ESX can be used directly on KVM
host without any effort. I think it could be very nice feature in future.


Thanks,

Xianqing Yu

------
Graduate Research Assistant, Cyber Defense Lab
Department of Computer Science
North Carolina State University, Raleigh, NC
E-mail: xyu6@ncsu.edu
-----Original Message----- 
From: Aaron Peeler
Sent: Tuesday, August 16, 2011 01:39 PM
To: vcl-dev@incubator.apache.org
Subject: Re: The New KVM Provisioning module

I suggest we use both for now, until it's fully supported in libvirt API.

Use qemu-img to create a qcow2 snapshot for non-imaging reservations.
Then use virsh for the rest of the process.

Once libvirt supports it, then we upgrade that portion.

We'll have to decide if we want to try to also use vmdk files formats
on kvm. It's a nice feature but not sure if it is a requirement for
using vmdk's on kvm hosts.

Also there is a virsh snapshot-create which appears to work only on
qcow2 based files. I've not successfully created a snapshot that I can
boot yet, but it might be another option.

Aaron

On Tue, Aug 16, 2011 at 12:20 PM, Xianqing Yu <yu267155364@hotmail.com> 
wrote:
> Aaron,
>
> Right, but this is only for KVM API, in qemu-kvm description, it said
> "-snapshot       write to temporary files instead of disk image files".
> Right now, Libvirt don't support that. If we using libvirt API in kvm
> provisioning module, we couldn't make non-persistent disk work until 
> libvirt
> getting updated in next version to include this.
>
> So for now, if we want to use non-persistent disk, we have to use KVM API
> directly (without libvirt) in our provisioning module.
>
> So do you means that we should use KVM API instead of libvirt API for now?
>
> Following is email message between Eric and me, FYI.
>
>
> On 08/15/2011 02:57 PM, Xianqing Yu wrote:
>>
>> So, if I understand correctly, non-persistent disk is not there yet. But
>> it may release with next version of libvirt, right?
>
> Correct.
>
> --
> Eric Blake   eblake@redhat.com    +1-801-349-2682
> Libvirt virtualization library http://libvirt.org
>
>
> On 08/15/2011 02:01 PM, Xianqing Yu wrote:
>>
>> Hi Libvirt team,
>>
>> I am trying to create a new domain from XML file by using libvirt. I
>> want this new domain using non-persistent disk as its disk.
>> Non-persistent disk means that when the domain is running, all the
>> modification would not be written to disk file. So next time, when you
>> boot VM from the same disk file, you would have a fresh disk. I check
>> the website but I could not find anything about this. I wonder if
>> libvirt support this function.
>
> Yep, I'm working on it, and hope to have it in 0.9.5.  See my RFC here:
>
> https://www.redhat.com/archives/libvir-list/2011-August/msg00552.html
>
> You will basically call out:
>
> <devices>
>  <disk ...>
>    ...
>    <transient/>
>  </disk>
>  ...
>
>> FYI, if I use KVM, I can use command "qemu-kvm -snapshot" to boot a VM
>> on non-persistent disk.
>
> Unfortunately, that makes for a non-migrateable VM - so I'm working on
> something a bit more robust that stands a chance of creating a transient
> yet migrateable domain.
>
> --
> Eric Blake   eblake@redhat.com    +1-801-349-2682
> Libvirt virtualization library http://libvirt.org
>
>
> Thanks,
>
> Xianqing Yu
>
> ------
> Graduate Research Assistant, Cyber Defense Lab
> Department of Computer Science
> North Carolina State University, Raleigh, NC
> E-mail: xyu6@ncsu.edu
>
> -----Original Message----- From: Aaron Peeler
> Sent: Tuesday, August 16, 2011 10:31 AM
> To: Xianqing Yu
> Cc: vcl-dev@incubator.apache.org
> Subject: Re: The New KVM Provisioning module
>
> Hi Xianqing,
>
> Right I think the snapshot method might be the best way to go. It
> would allow a quick way to clone with doing a full copy.
> Aaron
>
> On Mon, Aug 15, 2011 at 5:02 PM, Xianqing Yu <yu267155364@hotmail.com>
> wrote:
>>
>> Aaron,
>>
>> Libvirt currently don't have support for non-persistent disk. I sent 
>> email
>> to libvirt mail list to ask, and according to Eric Blake said, he is
>> working
>> on that now, and it may be included in libvirt 0.9.5.
>>
>> I also take a look at qemu-kvm API. It support non-persistent disk. I can
>> use command "qemu-kvm -snapshot......" to make it work.
>>
>> For now, I think we have to choose, either use libvirt without
>> non-persistent disk support, or use qemu-kvm API.
>>
>> Thanks,
>>
>> Xianqing Yu
>>
>> ------
>> Graduate Research Assistant, Cyber Defense Lab
>> Department of Computer Science
>> North Carolina State University, Raleigh, NC
>> E-mail: xyu6@ncsu.edu
>>
>> -----Original Message----- From: Aaron Peeler
>> Sent: Wednesday, August 10, 2011 04:01 PM
>> To: vcl-dev@incubator.apache.org
>> Subject: Re: The New KVM Provisioning module
>>
>> Xianqing,
>>
>> I've started reviewing the code and have a couple of questions. Please
>> take this as constructive, I'm excited to get the kvm development
>> started, but would like take a lot of the lessons we've learned from
>> some of the previous provisioning modules to get libvirt prov module
>> on the right path.
>>
>> In your load routine it looks like your having to copy the img and
>> vmdk files before starting the vm, was this intended or could it have
>> been something leftover from another template. This will be very slow
>> before vcld would get to the point of starting a vm, some those files
>> can be very large.
>>
>> With the vmdk's under VMware/VMware.pm we run in non-persistent mode
>> and for image captures started using snapshots in the development
>> code. I'm pretty sure you can do something similar with kvm.
>>
>> Do you know of or have you tested another method in running vms a sort
>> of non-persistent mode under kvm? Copying a large .vmdk or .img file
>> on each provisioning request will not work very well in the long-term.
>>
>> By chance did you have time to look at the
>> lib/VCL/Module/Provisioning/VMware/VMware.pm framework?
>> Ideally we'd like to build a similar framework for libvirt.
>>
>> Thanks,
>> Aaron
>>
>>
>>
>> On Thu, Aug 4, 2011 at 1:34 PM, Aaron Peeler <aaron_peeler@ncsu.edu>
>> wrote:
>>>
>>> This is great. I'll try to have some more useful comments soon. :)
>>> Thanks Xianqing.
>>>
>>> Aaron
>>>
>>> On Thu, Aug 4, 2011 at 11:11 AM, Xianqing Yu <yu267155364@hotmail.com>
>>> wrote:
>>>>
>>>> Hi VCL community,
>>>>
>>>> Sorry for the typo in previous email.
>>>>
>>>> I finish the new version of KVM provision module for VCL 2.2.1. I 
>>>> upload
>>>> the
>>>> kvm.pm code to VCL JIRE. Please take a look at my code and let me know
>>>> what
>>>> do you feel about it. The link is,
>>>>
>>>> https://issues.apache.org/jira/browse/VCL-339
>>>>
>>>> Generally, the new version of KVM module contain these features,
>>>>
>>>> 1. The module will use libvirt API to manipulate the VM on KVM hosts.
>>>> The
>>>> main reason behind this is that libvirt is widely used API, many people
>>>> in
>>>> the community talked about this before, and libvirt support for
>>>> different
>>>> hypervisors, such as KVM, Xen, Qemu, etc.. It would be easier to port
>>>> this
>>>> module to support other hypervisor in future.
>>>>
>>>> 2. An setup script can help the users to setup KVM host much easier. 
>>>> The
>>>> attachment named kvmsetup is the script. I tested it on Fedora 9.
>>>> Basically,
>>>> user need to install Fedora 9 (or use XCAT to install Fedora 9 on blade
>>>> automatically) and run the script as the root user.
>>>>
>>>> 3. Users don't have to use qemu-ifup and mac script. "qemu-ifup" is not
>>>> needed because of using lib-virt and "mac" is now included in kvm.pm. 
>>>> So
>>>> kvm.pm is the only file you need to add to VCL Management Machine.
>>>>
>>>> 4. Both "-flat.vmdk" and "img" image format are supported. So now both
>>>> KVM
>>>> image and VMware ESXi image (flat.vmdk) can be provisioned on KVM host.
>>>>
>>>> 5. Misc code updated.
>>>>
>>>>
>>>> Thanks,
>>>>
>>>> Xianqing
>>>> -----Original Message----- From: Andy Kurth
>>>> Sent: Wednesday, August 03, 2011 02:14 PM
>>>> To: vcl-dev@incubator.apache.org
>>>> Subject: Re: The New KVM Provisioning module
>>>>
>>>> $self->data->get_image_os_install_type() should do it.
>>>>
>>>> On Wed, Aug 3, 2011 at 12:47 PM, Xianqing Yu <yu267155364@hotmail.com>
>>>> wrote:
>>>>>
>>>>> A quick question, do you know any function call can get OSinstalltype
>>>>> content for the target image?
>>>>>
>>>>> Thanks,
>>>>>
>>>>> Xianqing Yu
>>>>>
>>>>> ------
>>>>> Graduate Research Assistant, Cyber Defense Lab
>>>>> Department of Computer Science
>>>>> North Carolina State University, Raleigh, NC
>>>>> E-mail: xyu6@ncsu.edu
>>>>>
>>>>> -----Original Message----- From: Andy Kurth
>>>>> Sent: Wednesday, August 03, 2011 09:38 AM
>>>>> To: vcl-dev@incubator.apache.org
>>>>> Subject: Re: The New KVM Provisioning module
>>>>>
>>>>> I agree that entries should be added to OSinstalltype.  I wouldn't
>>>>> include the 'kvm' hypervisor name in the name since the same type may
>>>>> be used by multiple hypervisors.  The 'vmware' entry can probably be
>>>>> renamed to 'vmdk' and then used by both VMware and KVM.  An entry can
>>>>> be added named something like 'img' or 'vm-img'.
>>>>>
>>>>> On Tue, Aug 2, 2011 at 5:00 PM, Xianqing Yu <yu267155364@hotmail.com>
>>>>> wrote:
>>>>>>
>>>>>> I think it is possible to handle both types: vmdk and img. I believe
>>>>>> that
>>>>>> in
>>>>>> my kvm provisioning code, the code need some conditions to decide
if
>>>>>> image
>>>>>> file is vmdk file or img file. For example, We can add "kvm-img"
and
>>>>>> "kvm-vmware" into OSinstalltype table. If one image's OS module point
>>>>>> to
>>>>>> "kvm-img" as OSinstalltype, then the code should consider the image

>>>>>> as
>>>>>> "img"
>>>>>> image file, and vice verse.
>>>>>>
>>>>>> Thanks,
>>>>>>
>>>>>> Xianqing Yu
>>>>>>
>>>>>> ------
>>>>>> Graduate Research Assistant, Cyber Defense Lab
>>>>>> Department of Computer Science
>>>>>> North Carolina State University, Raleigh, NC
>>>>>> E-mail: xyu6@ncsu.edu
>>>>>> -----Original Message----- From: Aaron Peeler
>>>>>> Sent: Tuesday, August 02, 2011 02:39 PM
>>>>>> To: vcl-dev@incubator.apache.org
>>>>>> Subject: Re: The New KVM Provisioning module
>>>>>>
>>>>>> The imagename is related to the OS table. It's also dependent on
the
>>>>>> the OS.moduleid, OS.installtype and the computer.provisioningid.
We
>>>>>> might need to add new entries for this module, which isn't a problem.
>>>>>> Just need to decide on how to name it.
>>>>>>
>>>>>> Ideally I would like to see the libvirt/kvm module handle both 
>>>>>> types -
>>>>>> vmdk and img.
>>>>>>
>>>>>> Aaron
>>>>>>
>>>>>>
>>>>>> On Mon, Aug 1, 2011 at 11:13 AM, Xianqing Yu 
>>>>>> <yu267155364@hotmail.com>
>>>>>> wrote:
>>>>>>>
>>>>>>> Thanks for your suggestion.
>>>>>>>
>>>>>>> I have a question about define the image name and capture function.
>>>>>>> Currently my kvm module can support vmware esxi image format
(.vmdk
>>>>>>> and
>>>>>>> flat.vmdk, so I can easily support existing flat.vmdk images
by 
>>>>>>> KVM).
>>>>>>> But
>>>>>>> I
>>>>>>> also want it to support other image formats, such as .img.
>>>>>>>
>>>>>>> I think we may need to define the name format both in VCL database
>>>>>>> and
>>>>>>> KVM
>>>>>>> module. Let's said there is an image named CentOS5,
>>>>>>>
>>>>>>> In VMDK format,
>>>>>>> In VCL database, it should be esx-CentOS5-v0,
>>>>>>> In image library, it should be files, esx-CentOS5-v0.vmdk and
>>>>>>> esx-CentOS5-v0-flat.vmdk.
>>>>>>>
>>>>>>> My question is that what its name should be when the image file
is
>>>>>>> img
>>>>>>> format? My final goal is that make KVM module can support flat.vmdk
>>>>>>> and
>>>>>>> img
>>>>>>> file at the same time. Is that possible to do that? Or I should
give
>>>>>>> one
>>>>>>> of
>>>>>>> them up?
>>>>>>>
>>>>>>> Thanks,
>>>>>>>
>>>>>>> Xianqing Yu
>>>>>>>
>>>>>>> ------
>>>>>>> Graduate Research Assistant, Cyber Defense Lab
>>>>>>> Department of Computer Science
>>>>>>> North Carolina State University, Raleigh, NC
>>>>>>> E-mail: xyu6@ncsu.edu
>>>>>>> -----Original Message----- From: Andy Kurth
>>>>>>> Sent: Monday, August 01, 2011 09:02 AM
>>>>>>> To: vcl-dev@incubator.apache.org
>>>>>>> Subject: Re: The New KVM Provisioning module
>>>>>>>
>>>>>>> Hi Xianqing,
>>>>>>> Sounds great.  I'd like to help with this too.  I like the idea

>>>>>>> about
>>>>>>> making it generic.  Since libvirt supports a few different things,
I
>>>>>>> think a very generic libvirt.pm module inheriting from
>>>>>>> Provisioning.pm
>>>>>>> makes sense:
>>>>>>> Package: VCL::Module::Provisioning::libvirt
>>>>>>> File: /usr/local/vcl/lib/VCL/Module/Provisioning/libvirt.pm
>>>>>>>
>>>>>>> I'd create a Provisioning/libvirt subdirectory, and create .pm
files
>>>>>>> for each libvirt/VCL supported product:
>>>>>>> Package: VCL::Module::Provisioning::libvirt::KVM
>>>>>>> File: /usr/local/vcl/lib/VCL/Module/Provisioning/libvirt/KVM.pm
>>>>>>>
>>>>>>> I agree to use the VMware module as an example though there will
be 
>>>>>>> a
>>>>>>> difference in how your Perl package path is constructed and where

>>>>>>> the
>>>>>>> files reside.  The main VMware.pm file should reside 1 directory
>>>>>>> higher, directly under Provisioning.  When the new VMware.pm
module
>>>>>>> was created, the older vmware.pm module already resided in this
>>>>>>> location.  We couldn't remove it at that time so I put VMware.pm
a
>>>>>>> directory lower to keep them straight.
>>>>>>>
>>>>>>> There is a lot of code in the VMware.pm module that could be
reused
>>>>>>> by
>>>>>>> other modules.  I tried to break up the subroutines as much as
>>>>>>> possible so that each one performs a single function.  Many of
them
>>>>>>> are not VMware-specific and could be moved up into Provisioning.pm

>>>>>>> so
>>>>>>> that the code can be reused by libvirt.pm and other provisioning
>>>>>>> modules.
>>>>>>>
>>>>>>> Submitting your code via Jira is the best way to get started.
 We
>>>>>>> already have issue VCL-339 for this.  Please be sure to submit
your
>>>>>>> code as an attached file to the issue, not copied/pasted to the
>>>>>>> description.
>>>>>>>
>>>>>>> Also, 1 suggestion for anyone working on code.  Make sure you
are
>>>>>>> developing using a checked-out copy of trunk from Subversion
and
>>>>>>> perform a 'svn update' regularly.  Don't work off of the code
in the
>>>>>>> VCL 2.2.1 release because it is changing frequently.  It will
also 
>>>>>>> be
>>>>>>> very easy to commit your code to trunk this way.
>>>>>>>
>>>>>>> Thanks,
>>>>>>> Andy
>>>>>>>
>>>>>>> On Fri, Jul 29, 2011 at 4:30 PM, Xianqing Yu <xianyu@us.ibm.com>
>>>>>>> wrote:
>>>>>>>>
>>>>>>>> Aaron,
>>>>>>>>
>>>>>>>> Thank your interesting.
>>>>>>>>
>>>>>>>> I looked at the vmware Provisioning module code, and try
to learn
>>>>>>>> and
>>>>>>>> reuse some code from that. Because there are some many functions
in
>>>>>>>> VMware.pm, I probably start with some basic functions, and
make it
>>>>>>>> work
>>>>>>>> first.
>>>>>>>>
>>>>>>>> About structure, I has several functions inside my kvm.pm,
>>>>>>>> including,
>>>>>>>> initialize, load, capture, node_status and get_image_size.
I 
>>>>>>>> already
>>>>>>>> finish
>>>>>>>> most of them except for capture function. So basically, kvm
module
>>>>>>>> will
>>>>>>>> do
>>>>>>>> very similar thing as VMware.pm. And it generates a XML file
to
>>>>>>>> define
>>>>>>>> the
>>>>>>>> VM (like VMX file in VMware) and creates VM on KVM by sending
ssh
>>>>>>>> command
>>>>>>>> "virsh create vm.xml". I have a script which can install
necessary
>>>>>>>> packages
>>>>>>>> and set up proper network configuration on the host. So I
can use
>>>>>>>> that
>>>>>>>> with
>>>>>>>> xCAT to create kickstart image for KVM host. I also can use
the
>>>>>>>> script
>>>>>>>> to
>>>>>>>> configure existing Fedora machine.
>>>>>>>>
>>>>>>>> I can have my script and kvm provisioning code ready within
three
>>>>>>>> days,
>>>>>>>> so
>>>>>>>> you guys can take a look. So should I publish it on the JIRE?
>>>>>>>>
>>>>>>>> Thanks,
>>>>>>>>
>>>>>>>> Xianqing Yu
>>>>>>>>
>>>>>>>> WSTI Intern
>>>>>>>> xianyu@us.ibm.com
>>>>>>>>
>>>>>>>> -----Aaron Peeler <fapeeler@ncsu.edu> wrote: -----
>>>>>>>>
>>>>>>>>  To: vcl-dev@incubator.apache.org
>>>>>>>>  From: Aaron Peeler <fapeeler@ncsu.edu>
>>>>>>>>  Date: 07/29/2011 08:58AM
>>>>>>>>  Subject: Re: The New KVM Provisioning module
>>>>>>>>
>>>>>>>>  Hi Xianqing,
>>>>>>>>
>>>>>>>>  This is great. I'd like to work with you this module.
>>>>>>>>
>>>>>>>>  What are your thoughts on the structure?
>>>>>>>>
>>>>>>>>  Have you looked at the work Andy did on the VMware Prov
module:
>>>>>>>>  /..../lib/VCL/Module/Provisioning/VMware
>>>>>>>>
>>>>>>>>
>>>>>>>>  Aaron
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>  On Tue, Jul 26, 2011 at 11:18 PM, Xianqing Yu <xianyu@us.ibm.com>
>>>>>>>> wrote:
>>>>>>>>  >
>>>>>>>>  >
>>>>>>>>  >
>>>>>>>>  > Hi VCL community,
>>>>>>>>  >
>>>>>>>>  > I am developing KVM Provisioning Module for VCL system,
which is
>>>>>>>> based
>>>>>>>> on
>>>>>>>>  > my previous version of KVM Provisioning Module. You
can find out
>>>>>>>> my
>>>>>>>>  > previous KVM module's information from here,
>>>>>>>>  > https://issues.apache.org/jira/browse/VCL-339
>>>>>>>>  >
>>>>>>>>  > Currently, I am trying to make this module's code as
generic as
>>>>>>>> possible.
>>>>>>>>  > And I will including these features inside this new
version of
>>>>>>>> module.
>>>>>>>>  > 1. The module will use libvirt API to manipulate the
VM on KVM
>>>>>>>> hosts.
>>>>>>>> The
>>>>>>>>  > main reason behind this is that libvirt is widely used
API, many
>>>>>>>> people in
>>>>>>>>  > the community talked about this before, and libvirt
support for
>>>>>>>> different
>>>>>>>>  > hypervisors, such as KVM, Xen, Qemu, etc.. It would
be easier to
>>>>>>>> port
>>>>>>>> this
>>>>>>>>  > module to support other hypervisor in future.
>>>>>>>>  > 2. A setup scripted will be developed to help the users
to setup
>>>>>>>> host
>>>>>>>> much
>>>>>>>>  > easier.
>>>>>>>>  > 3. A document descripts how to install this module.
>>>>>>>>  >
>>>>>>>>  > Welcome everyone to join the discussion with me if
you are
>>>>>>>> interesting
>>>>>>>>  > about. Such as, is there any new features you expect
to see in
>>>>>>>> this
>>>>>>>> module?
>>>>>>>>  > Do you have any suggestion about this module? Or you
have any
>>>>>>>> questions
>>>>>>>>  > about this module? Please let me know.
>>>>>>>>  >
>>>>>>>>  > Thanks,
>>>>>>>>  >
>>>>>>>>  > Xianqing Yu
>>>>>>>>  > WSTI Intern
>>>>>>>>  > xianyu@us.ibm.com
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>  --
>>>>>>>>  Aaron Peeler
>>>>>>>>  Program Manager
>>>>>>>>  Virtual Computing Lab
>>>>>>>>  NC State University
>>>>>>>>
>>>>>>>>  All electronic mail messages in connection with State business
>>>>>>>> which
>>>>>>>>  are sent to or received by this account are subject to the
NC
>>>>>>>> Public
>>>>>>>>  Records Law and may be disclosed to third parties.
>>>>>>>
>>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>> --
>>>>>> Aaron Peeler
>>>>>> Program Manager
>>>>>> Virtual Computing Lab
>>>>>> NC State University
>>>>>>
>>>>>> All electronic mail messages in connection with State business which
>>>>>> are sent to or received by this account are subject to the NC Public
>>>>>> Records Law and may be disclosed to third parties.
>>>>>>
>>>>>
>>>>>
>>>>
>>>>
>>>
>>>
>>>
>>> --
>>> Aaron Peeler
>>> Program Manager
>>> Virtual Computing Lab
>>> NC State University
>>>
>>> All electronic mail messages in connection with State business which
>>> are sent to or received by this account are subject to the NC Public
>>> Records Law and may be disclosed to third parties.
>>>
>>
>>
>>
>> --
>> Aaron Peeler
>> Program Manager
>> Virtual Computing Lab
>> NC State University
>>
>> All electronic mail messages in connection with State business which
>> are sent to or received by this account are subject to the NC Public
>> Records Law and may be disclosed to third parties.
>>
>
>
>
> --
> Aaron Peeler
> Program Manager
> Virtual Computing Lab
> NC State University
>
> All electronic mail messages in connection with State business which
> are sent to or received by this account are subject to the NC Public
> Records Law and may be disclosed to third parties.
>



-- 
Aaron Peeler
Program Manager
Virtual Computing Lab
NC State University

All electronic mail messages in connection with State business which
are sent to or received by this account are subject to the NC Public
Records Law and may be disclosed to third parties. 


Mime
View raw message