[one-users] [DEV] Questions about scale-in implementation in OpenNebula
Carlos Martín Sánchez
cmartin at opennebula.org
Wed Feb 8 03:59:07 PST 2012
Hi Paul,
This is a very interesting feature. You should open a new ecosystem project
[1] as soon as your code is usable, so others can test it. If you would
like to see your code merged upstream once it gets to a mature enough
state, make sure that whoever has to give the thumbs up in your company is
aware of the License Agreement [2].
And now some quick comments to your questions:
On Mon, Feb 6, 2012 at 12:19 PM, Paul Grandperrin <
paul.grandperrin at alterway.fr> wrote:
>
> 1. About VM memory scaling: Currently, AFAIK the vm.memory is used when
> deploying a VM to set it's initial memory and then is regularly updated via
> hypervisor polling.
> ATM, i'm also using this attribute to change memory size. I think it's
> really not the best way thing to do. I'd like to separate theses different
> things in separate variables.
> For exemple:
> -memory: the same as of now.
> -memory_target: the target amount of memory when scaling memory.
>
> I could also use VM history but I'm not very familiar with this class.
>
>
Each history entry represents a host change, so new ones are created only
when the VM is deployed, migrated, or stopped + resumed. That's not the
best place to log the scaling changes.
About storing the target amount of memory: VM::memory is the used memory,
as reported by the polling. The memory definition, set by the user and used
to create the deployment file, is taken from the MEMORY attribute of
VM::obj_template. I think you should overwrite that attribute to store the
target memory.
Before doing this scaling operation, you should check that the host has
enough free memory. After the operation, the host share should be updated,
take a look at Host::host_share, Host::add_capacity and Host::del_capacity.
If you don't update the host share memory, when the VM is shutdown it will
leave the host with a negative memory value.
> 2. When scalling the number of VCPUs, should we also scale the VM's cpu
> share? If so, how to implement it?
>
I'm not sure about the desirable behaviour. Maybe this should be decided by
the user? If you are going to modify the CPU, and not only the VCPU, all
the above comments about the MEMORY apply.
> 3. In the case of a scalling failure (memory or vcpu), what should we do?
> -Consider the VM failed and not usable anymore? (I think it's way too
> strict)
> -Consider the VM still ACTIVE. However, how to inform the user about
> the failure (something else than writing in logs).
> And then what should we do?
> -immediatly throw a monitor request to update to the correct value?
> -Consider the worst case: if scaling down the memory => consider
> the old value/ if scaling up the memory, consider the new value
> -Other ideas?
>
I've seen you are creating new LCM states. This can be very tricky, maybe
you should just apply the action without moving from the RUNNING state,
like the reboot action. And, if you are creating new states, at least try
to keep it simple and merge those two new ones into just one. SCALING, or
even a more generic... HOTPLUG?
I would always return to the RUNNING state, updating MEMORY and CPU (and
Host:: host_share) in case of success.
The user will see that the operation finished, and will see if it succeeded
taking a look at the VM template. You can also include an error message in
the template if the operation failed.
If the scaling command returns success/failure immediately, I would not
force a poll. As I said, the poll updates the used memory, not the amount
set for the VM.
Regards... and good luck!
[1] http://opennebula.org/community:ecosystem
[2] http://opennebula.org/community:contribute
--
Carlos Martín, MSc
Project Engineer
OpenNebula - The Open Source Toolkit for Data Center Virtualization
www.OpenNebula.org | cmartin at opennebula.org |
@OpenNebula<http://twitter.com/opennebula><cmartin at opennebula.org>
On Mon, Feb 6, 2012 at 12:19 PM, Paul Grandperrin <
paul.grandperrin at alterway.fr> wrote:
> Hi all,
>
> I'm implementing scale-in features in OpenNebula like live memory
> growth/shrinking and vcpus hotplugging/hotunplugging.
>
> You can see my git there: http://paulg.fr/gitweb/?p=one.git;a=summary;js=1
> My developement branch is feature-scalein. It's still very much a WIP most
> of the interesting code is there and basic features are functionnal on Xen
> at the moment.
> My dev branch is based on one-3.2 but can easily be rebased on master.
>
> Everything is meant to eventualy hit upstream, that why I'd like to get
> some advices and feedback from you.
>
> Here are my questions:
>
> 1. About VM memory scaling: Currently, AFAIK the vm.memory is used when
> deploying a VM to set it's initial memory and then is regularly updated via
> hypervisor polling.
> ATM, i'm also using this attribute to change memory size. I think it's
> really not the best way thing to do. I'd like to separate theses different
> things in separate variables.
> For exemple:
> -memory: the same as of now.
> -memory_target: the target amount of memory when scaling memory.
>
> I could also use VM history but I'm not very familiar with this class.
>
> 2. When scalling the number of VCPUs, should we also scale the VM's cpu
> share? If so, how to implement it?
>
> 3. In the case of a scalling failure (memory or vcpu), what should we do?
> -Consider the VM failed and not usable anymore? (I think it's way too
> strict)
> -Consider the VM still ACTIVE. However, how to inform the user about
> the failure (something else than writing in logs).
> And then what should we do?
> -immediatly throw a monitor request to update to the correct value?
> -Consider the worst case: if scaling down the memory => consider
> the old value/ if scaling up the memory, consider the new value
> -Other ideas?
>
> Any suggestions about the code structure, writing style, naming
> conventions, whatever... are welcome :D
>
> You can also see my TODO list here:
> http://paulg.fr/gitweb/?p=one.git;a=blob_plain;f=TODO;h=79c65a4a6eba19095a43191a75fc1e5d58d7e01a;hb=refs/heads/feature-scalein;js=1
>
> What changed:
> paulg at debian-pro:~/projects/one$ git diff one-3.2 --stat
> TODO | 12 ++
> include/DispatchManager.h | 20 +++
> include/LifeCycleManager.h | 20 +++-
> include/RequestManagerVirtualMachine.h | 36 +++++
> include/VirtualMachine.h | 43 +++++-
> include/VirtualMachineManager.h | 50 +++++--
> include/VirtualMachineManagerDriver.h | 50 +++++--
> install.sh | 4 +-
> share/man/onevm.1 | 60 ++++++++
> src/cli/one_helper.rb | 2 +-
> src/cli/one_helper/onevm_helper.rb | 24 +++
> src/cli/onevm | 32 ++++
> src/dm/DispatchManagerActions.cc | 90 +++++++++++
> src/lcm/LifeCycleActions.cc | 68 +++++++++-
> src/lcm/LifeCycleManager.cc | 48 ++++++
> src/lcm/LifeCycleStates.cc | 123 +++++++++++++++
> src/mad/ruby/VirtualMachineDriver.rb | 56 +++++--
> src/oca/ruby/OpenNebula/VirtualMachine.rb | 27 +++-
> src/rm/RequestManager.cc | 4 +
> src/rm/RequestManagerVirtualMachine.cc | 105 +++++++++++++-
> src/vm/VirtualMachine.cc | 3 +
> src/vmm/VirtualMachineManager.cc | 231
> +++++++++++++++++++++++++++--
> src/vmm/VirtualMachineManagerDriver.cc | 72 +++++++++-
> src/vmm_mad/dummy/one_vmm_dummy.rb | 8 +
> src/vmm_mad/exec/one_vmm_exec.rb | 42 +++++-
> src/vmm_mad/exec/one_vmm_sh | 2 +-
> src/vmm_mad/remotes/xen/scale_memory | 26 ++++
> src/vmm_mad/remotes/xen/scale_vcpu | 26 ++++
> src/vmm_mad/remotes/xen/xenrc | 3 +-
> 29 files changed, 1204 insertions(+), 83 deletions(-)
>
> Thank for your help,
>
> Paul Grandperrin
>
> _______________________________________________
> Users mailing list
> Users at lists.opennebula.org
> http://lists.opennebula.org/listinfo.cgi/users-opennebula.org
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.opennebula.org/pipermail/users-opennebula.org/attachments/20120208/d3437826/attachment-0003.htm>
More information about the Users
mailing list