cloudstack-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Alena Prokharchyk <>
Subject Re: [PROPOSAL] ROOT volume detach - feature for CS 4.5
Date Fri, 28 Mar 2014 18:48:17 GMT
I will look into it more, Mike. vmWare indeed can be different.


From: Mike Tutkowski <<>>
Date: Friday, March 28, 2014 at 11:39 AM
To: Alena Prokharchyk <<>>
Cc: "<>" <<>>
Subject: Re: [PROPOSAL] ROOT volume detach - feature for CS 4.5

VMware is also different because when you shut a VMware VM down from CloudStack, the VM still
exists in vCenter Server (whereas for XenServer and KVM, the VM is gone).

Since the life of a datastore that was created for managed storage is tied to the life of
the CloudStack volume it stores, when the CloudStack volume is deleted, the datastore goes
away, as well.

If the datastore in question was automatically created to store a root disk (alongside VM
config files) and you switch the VM to another root disk (which has to necessarily be in another
datastore), you won't see a problem until the original root volume is expunged by CloudStack.
At this point, its datastore will go away along with your VM config files.

On Fri, Mar 28, 2014 at 12:31 PM, Mike Tutkowski <<>>
Well, the reason I brought it up was mainly due to VMware.

Let's use that as an example:

I initiate the process of spinning up a VM based on managed storage.
A volume is dynamically created on a SAN.
VmwareStorageProcessor dynamically creates a datastore to consume the newly created SAN volume.
All VMware VM files (ex. VMX, NVRAM) are placed in the datastore alongside the VMDK file that
represents the root volume.

Now, let's say we want to detach this root volume and give the VM a new root volume.

The new root volume will necessarily be on a different datastore than the datastore of the
previous root volume (because a datastore created to consume managed storage will have at
most one VMDK file*).

Is it going to be a problem that the VM's files (ex. VMX, NVRAM) are on one datastore, but
its root disk is on another?

I don't think it's really a problem until you go to delete the original root volume from CloudStack.
At that point, its datastore will be removed (including, of course, your VM's VMX, NVRAM,
etc. files).

This is not really a problem on XenServer because XenServer does not store VM config files
in the SR, so I think we're OK there.

We should also be OK for KVM.

* Technically it can have many if those other VMDK files are delta snapshots, but they still
- together - represent a single disk.

On Fri, Mar 28, 2014 at 10:36 AM, Alena Prokharchyk <<>>
Mike, thank you for the explanation on managed storage.. As far as I understand from your
email, the main difference is instead of creating an SR on the PS, CloudStack will recognize
pre-existing volume created outside of the CS. Am I correct?

If so, I don’t think there would be any difference. When root volume detach happens, no
storage attributes – path, clusterId – are being changed. And we would apply the same
set of checks to the root volume attach, as for a dataDisk attach.


From: Mike Tutkowski <<>>
Date: Thursday, March 27, 2014 at 9:40 PM
To: Alena Prokharchyk <<>>
Cc: "<>" <<>>
Subject: Re: [PROPOSAL] ROOT volume detach - feature for CS 4.5

Hi Alena,

I was wondering if you've taken "managed" storage into consideration for this?

If you're unfamiliar with it, managed storage is named as such because CloudStack manages
it on behalf of the admin (ex. dynamically creating SRs as needed).

For example, when I add primary storage to CloudStack that is based on the SolidFire SAN,
I use the SolidFire plug-in, which is an example of managed storage.

In this case, the primary storage represents a SAN as opposed to a preallocated volume.

When the time comes to, say, attach a data disk to a VM for the first time, the SolidFire
plug-in goes off to its SAN and dynamically creates a new volume on it (with the appropriate
size and IOPS requirements).

CloudStack has logic that recognizes managed storage.

For example, for XenServer, its logic has been augmented to automatically create an SR based
on the iSCSI target that was created on the SAN and to create a VDI within it that is attached
to the VM in question.

The big takeaway is that each CloudStack volume here will be associated with a unique volume
on a SAN and consumed as an SR (XenServer) or datastore (ESX) (KVM handles this differently).
In this situation, there is a 1:1 mapping between a SAN volume and an SR. No other VDIs are
stored on the SR except for the one representing this one CloudStack volume.

That being the case, I was wondering what you thought of this with regards to your root-volume-detach

If we don't want to look into this for 4.5, it might be best to simply fail to detach a root
volume from a VM if the volume is based on managed storage or to fail to attach a bootable
volume to a VM if it is based on managed storage.

Talk to you later,

On Tue, Mar 25, 2014 at 1:24 PM, Alena Prokharchyk <<>>

Volume has a template_id referencing vm_template table. Vm_template has
bootable flag, so we will derive information from there.
And sure, this information will not change if the root disk is detached.

On 3/25/14, 12:18 PM, "Mike Tutkowski" <<>>

>Hi Alena,
>I was wondering how we plan to keep track of the new "bootable" property?
>When we create a VM, would we just mark its root disk as bootable and then
>that property becomes immutable (for the upgrade case, all root disks
>be marked as bootable)?
>I'm thinking we'd want to keep track of bootable disks even when there are
>detached and turned into data disks. Is that what you had in mind?
>On Tue, Mar 25, 2014 at 12:20 PM, Alena Prokharchyk <
><>> wrote:
>> Here is the link to the corresponding FS (placed in "4.5 Design
>> section)
>> -Alena.
>> From: Alena Prokharchyk <<><mailto:
>> Date: Monday, March 24, 2014 at 11:37 AM
>> To: "<><<>>"
>> Subject: [PROPOSAL] ROOT volume detach - feature for CS 4.5
>> I would like to propose a new feature for CS 4.5 - "ROOT volume detach"
>> that enables support for following use cases:
>> 1) Replace current ROOT volume with the new one for  existing vm.
>> 2) Case when ROOT volume of vm1 gets corrupted, and you want to attach
>> to vm2 to run the recovery utils on it. With current CS implemntation,
>> have to perform several steps - create snapshot of vm1's volume, create
>> volume from snapshot, attach volume to the vm2. New implementation will
>> merge it all to one step.
>> With the planned implementation, once the ROOT volume is detached, it
>> be attached to any existing vm (with respect to Admin/Domain/Physical
>> resources limitations), either as a DataDisk or a Root disk.
>> Amazon EC2 already has this functionality in place, so I think CS would
>> only benefit from having it. Storage experts (Edison, others) please
>> your concerns if you have any, or if you see any potential problems with
>> the planned implementation. And if anyone can think of other use cases
>> feature can possible solve, I would appreciate this input as well.
>> Feature limitations:
>> * ROOT volume can be detached only when vm is in Stopped state
>> * CS will fail to start the vm not having a ROOT volume
>> I will send out the link to the FS once I start getting feedback on the
>> proposal.
>> -Alena.
>*Mike Tutkowski*
>*Senior CloudStack Developer, SolidFire Inc.*
>o: 303.746.7302<tel:303.746.7302>
>Advancing the way the world uses the

Mike Tutkowski
Senior CloudStack Developer, SolidFire Inc.
o: 303.746.7302<tel:303.746.7302>
Advancing the way the world uses the cloud<>™

Mike Tutkowski
Senior CloudStack Developer, SolidFire Inc.
o: 303.746.7302<tel:303.746.7302>
Advancing the way the world uses the cloud<>™

Mike Tutkowski
Senior CloudStack Developer, SolidFire Inc.
o: 303.746.7302
Advancing the way the world uses the cloud<>™

  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message