cloudstack-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Kelven Yang <>
Subject RE: [cloudstack-devel] Hyper-V Support
Date Sun, 22 Apr 2012 00:54:03 GMT
>> 1) I'm looking at the commands in Is there some documentation?
Unfortunately, we haven't put a detail document on this. Vmware or XenServer resource is a
good place to look at

>> 2) Commands are synchronous or asynchronous? In case how do you handle running jobs?

There is a design principal that command executors are dumb executors, the command flow business
logic should be implemented in corresponding managers running inside management server, therefore
a resource implementation will be easily hosted either locally within the management server
or remotely at agent side. 

Command is usually executed synchronously(we actually provides mechanism to execute command
asynchronously), managers can schedule asynchronous jobs. 

>> 3) Configuration: how do I tell CS that a new HyperVisor Resource type exists? (both
cases: Manager based / Agent based)
To add a new hypervisor, there are a number of things you need to do together,
	1), add a new type into the enum
	2), add the new type to HypervisorList so that UI may be aware of
	3) Add the resource discover class, refer to,,
	4) Add the resource class(for managed resource), refer to,
	5) Populate guest OS types and system template data for the hypervisor to database, refer
to create-schema.sql and template.sql 
	6) Prepare the default system template for the hypervisor
>> 4) Storage:
This is the part that needs a lot of discussions.
We are talking about secondary storage. Usually only hypervisor host and storage system VM
need to access to it, VMware does one special step that management server needs to access
to it to bootstrap system VMs. So to support hybrid deployment option, we would require secondary
storage to be accessible from all of them, giving the fact that Xen/KVM/VMware only supports
NFS shares, it would be better to make secondary storage dual faces for Hyper-V, in this way,
the same secondary storage will be able to serve XenServer/VMware/KVM/Hyper-V within the same

We do have a Hyper-V prototype in current CloudStack code base, you may also want to check
it out before you start.


-----Original Message-----
From: Alessandro Pilotti [] 
Sent: Saturday, April 21, 2012 1:35 PM
Cc: Development discussions for CloudStack
Subject: Re: [cloudstack-devel] Hyper-V Support

I started looking into the code for VMWare and KVM resources in order to develop the Hyper-V
Resource, the proxy based architecture is clear.

I have some basic questions to start with :-)

1) I'm looking at the commands in Is there some documentation? To start
coding the HyperV adapter, is there a list of mandatory and optional commands? I'm "reverse
engineering" their semantyc looking at the VMWare resources, but it's not so trivial :)

2) Commands are synchronous or asynchronous? In case how do you handle running jobs?

3) Configuration: how do I tell CS that a new HyperVisor Resource type exists? (both cases:
Manager based / Agent based)

4) Storage: the free Hyper-V server does not contain an NFS client. There are of course open
source NFS clients (Cygwin, etc) that can be installed, but supporting SMB might be easier.
What do you think about adding support for SMB shares in CS?  



On Apr 20, 2012, at 02:30 , Frank Zhang wrote:

> > There's a JSON-based protocol to pass commands between Management Server and host.
> >>  That sounds great! Do you maybe have a link for some documentation and samples?
>                 CloudStack has two types of managing host. Agent based and manager based.
Agent based means installing an agent on host and management server directly controls host
through the agent. KVM and XenServer falls into this category,  as KVM agent is written by
CloudStack developer while agent of XenServer is provided by Xapi.  The manager based approach
applies to VMWare, HyperV, OVM3 where CloudStack invokes API of SDK provided by hypervisor
vendor to communicate with *the manager*, for example, VCenter of VMWare, the manager is on
behalf of CloudStack to control the host. in our code paths, these two approaches look like:
> 1.       Agent Based:
> CloudStack business logic -------à Resource -----(XMLRPC or something) ---------à Agent
on host
> 2.       Manager Based:
> CloudStack business logic ------à Resource -----(API in SDK) ----------à hypervisor
manager provided by vendor ------à host
>                 To support a new type of hypervisor, the key is to implement a *Resource*
showed in above. The Resource is a command executor which receives commands from CloudStack
business logic(known as various managers, like networkManager, StorageManager, UserVmManager
. in our code) and performs these commands to hypervisor by means of either XmlRpc or SDK
API. In GoF design patterns, the Resource is a proxy pattern that works as a surrogate between
CloudStack and hypervisors.
>                 If you open one of hypervisor resources source file (for example,,,, you will find all of them implement
the same set of commands that are exhibited by method:
> public Answer executeRequest(Command cmd)
> take a look at that method and implement a similar resource like others, then you add
HyperV support into CloudStackJ
> Thanks,
> Alessandro
> On Apr 19, 2012, at 19:16 , Kevin Kluge wrote:
> Yes, very interesting.   Can you elaborate on the getThumbnail function.   One issue
we have been thinking about with Hyper-V is how to do guest console display (console proxy
functionality, in CloudStack terms).  Since only RDP is available with Hyper-V, and CloudStack
knows only VNC, we've been expecting that RDP is needed in CloudStack to provide console view.
> Did you integrate with Hyper-V in Windows Server 2008 R2?   Or something else?
> The CloudStack has existing code/framework to implement what we call a remote agent (your
scenario 3).   Take a look at how KVM hosts are managed.   There's a JSON-based protocol to
pass commands between Management Server and host.
> -kevin
> -----Original Message-----
> From: Rajesh Battala []
> Sent: Thursday, April 19, 2012 8:59 AM
> To:
> Subject: RE: Hyper-V Support
> Idea is great.
> All these Hyper-V operations are implement to manage the Hyper-V box
> directly  using WMI calls right?
> Or these operations are implemented via SCVMM?
> Thanks
> Rajesh Battala
> -----Original Message-----
> From: Alessandro Pilotti []
> Sent: Thursday, April 19, 2012 9:02 PM
> To:
> Cc:
> Subject: Hyper-V Support
> Hi guys,
> I'm new to this list, so hi everybody :-)
> I'm interested in providing code for integrating Cloudstack with Hyper-V. We
> developend an Hyper-V management framework that we use in our cloud
> products that can be used (at least as as a starting point).
> I'm summing up at the bottom of this email what we already have in terms of
> Hyper-V features handled by our framework (completed and tested). We
> basically cover everything needed for CloudStack and more. :-)
> Beside that we also just released an open source Hyper-V backup library and
> CLI tool: So far it's the only open source
> tool handling VSS backups of VMs on CSV storage :-)
> The assemblies are written in C# with .Net as the only dependency.
> I see 3 options to integrate our work with CloudStack:
> Write a Java adapter on top of the C# assembly (via JNI) Rewrite the C# code
> in Java, considering the quirkness for accessing WMI from java (jWMI, etc)
> Deploy the assembly on the Hyper-V hosts and add a RESTful layer on top to
> be consumed by a Java adapter (locally or remotely). That would be the best
> option in terms of performance and security (and the fastest to release :-) ).
> I prefer the third option, but I'm open to any idea!
> Looking forward for your opinion!
> BTW We plan to setup a CloudStack Hyper-V service in our datacenter on top
> of one of the clusters as soon as we have a working beta.
> Thanks,
> Alessandro Pilotti
> Cloudbase Solutions Srl
> ------------------------------------------------------------
> IT Consultant & Technical Speaker
> RHCE - Red Hat Certified Engineer
> ------------------------------------------------------------
> VM
> Create
> Update
> Delete
> Add / update / remove any type of resource (ethernet emulated/synthetic
> adapther, VHDs, ISO images etc) List Get summary Get thumbnail Get
> integration tools status and KV data Get IP addresses Start Stop Pause Save
> Shutdown Take snapshot List snapshots Revert to snapshot Remove
> snapshots Export Import Network
> Create VirtualSwitch
> Delete VirtualSwitch
> List VirtualSwitches
> Create VirtualSwitch port
> remove VirtualSwitch port
> Bind external ethernet port
> Setup VirtualSwitch (connect to external ethernet port) Terdown switch
> Create internal ethernet port Remove internal ethernet port Connect
> VirtualSwitch port to VM or other ports Disconnect VirtualSwitch port
> Storage
> Create VHD (fixed, dynamic, differencing) Compact VHD Convert VHD type
> Merge VHD with parent Validate VHD Mount / unmount VHD Reconnect
> parent VHD Get VHD info Expand VHD Create Virtual Floppy Disk
> Utility
> Get async job info
> Wait for async job info
> Remote file system management
> Cluster
> Create VM resource
> Remove VM resource
> Live migrate VM
> Create CSV
> Move CSV
> Backup / Restore
> ------------------------------------------------------------------------------
> For Developers, A Lot Can Happen In A Second.
> Boundary is the first to Know...and Tell You.
> Monitor Your Applications in Ultra-Fine Resolution. Try it FREE!
> cloudstack-devel mailing list

View raw message