Fwd: Re: KVM management and API

Fwd: Re: KVM management and API

[jd]

[-- Attachment #1.1: Type: text/plain, Size: 208 bytes --]

oops. forgot to put kvm-devel.

/Jd

Note: forwarded message attached.
       
---------------------------------
Ahhh...imagining that irresistible "new car" smell?
 Check outnew cars at Yahoo! Autos.

[-- Attachment #1.2: Type: text/html, Size: 368 bytes --]

[-- Attachment #2: Type: message/rfc822, Size: 16002 bytes --]

[-- Attachment #2.1.1: Type: text/plain, Size: 7067 bytes --]

Thanks Daniel. But I was looking for bit of under the hood answers.

Does libvirt use ps -ef | grep qemu to find out qemu processes ? Or it manages only ones that are started using libvirt. Using some naming scheme.

How do you know if a running process is using kvm kernel module or not ? 

libvirt seems to be connecting to  qemu console through qemud ? right ?  You have done quite a bit of work to send command and parse. !! Isnt this error prone ? (will code for parsing work in different char set ? or qemu console is always guarenteed to be in english ?)

How do you find out what % of allocated memory is actually used by VM ? 

For cpu /proc seems to be used.. which seems fine, but will not work for QEMU on Windows.

Some core qemu quesions still remains 
-- what are the plans to dynamically be able to change cpu , mem, disks and n/w for  a running vm ?

-- proper shutdown/restart of VM ?

Hoping some kvm/qemu experts would thrown in some light on the directions and plans.

Thanks in advance.
/Jd
 



"Daniel P. Berrange" <berrange-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org> wrote: On Fri, Apr 27, 2007 at 02:29:52PM -0700, jd wrote:
> Hi
>    This is a great project with a lots and lots of potential. I have a 
>    bunch of questions related to managing kvm VMs.
> 
>    I would like to know what kind of options are available for doing 
>    effective management ot KVM/QEMU processes (VMs). And know more abut 
>    plans / roadmap /directions to get to a proper API. (similar to Xen-API)

We are supporting QEMU / KVM management in libvirt since version 0.3.0
aiming to have feature parity for our management APIs between Xen, QEMU
and KVM. This gives a C, Python & Perl API. A command line tool virsh.
And a couple of higher level tools virt-install & virt-manager. I'll
illustrate possible answers to your questions in terms of virsh...

> 1. Discovery : How does one find all processes running KVM vms ?
>     ps -ef | grep qemu ? 

libvirt doesn't expose the PIDs to app developers, but you can get a
list using

 # export VIRSH_DEFAULT_CONNECT_URI=qemu:///system
 # virsh list --all
   Id Name                 State
  ----------------------------------
    1 QEMUGuest1           running
    - wizz                 shut off

>     How does one know if qemu is using the kvm or not ?

 # virsh dumpxml QEMUGuest1 | grep domain
 

The 'type' attribute will be one of qemu, kqemu, kvm

> 2. Identity : I am assuming that each VM is a process and pid would 
> be an identifier, is there any other more unique identifier available. 

libvirt provides 3 identifiers with different levels of uniqueness

 1. 'id' - integer uniquely identifying amongst active guests on a host
 2. 'name' - string uniquely identifying amongst active & inactive guests on a host
 3. 'uuid' - hex string uniquely identifying globally

> 3. Managing running processs :
>    a. How does one connect to the manage the running vms ?
>        I have read about screen and socat options.
>       Both of these act as tty client, where success/failure and errors for a given 
>       command are very vague (parsing..). (A way to get return code for the operation 
>       would be great). Parsing ... command response and interpreting... is error prone.
>      Say I choose socat, how do I gracefully detach from the console. ( other than  
>     just killing socat)

virsh provides a set of commands for controlling lifecycle  & configuration
of the guest from the host machine, eg shutdown, start, pause, resume, suspend,
restore, dominfo, dumpxml, etc, etc

The graphical console is typically exposed with VNC / SDL. You can find the
VNC port with:

  # virsh vncdisplay QEMUGuest1
  :0

If you configured a serial port & setup the guest to start a getty on it
you can also connect via the 'serial console'

  # virsh console QEMUGuest1


>    b. How does one get the stats for the running VM ? 
>        1. status : running, stopped, *migrating*..?
>        2. Current use of the memory : example allocated 1 GB, used 512 MB.
>        3. CPU : Effective cpu used ? Are vcpus supported ?
>            /proc can be used... what about windows ? (when and if we have kvm on windows)
>        4. How much I/O and Network I/O
>        5. Wait times to get to I/O or CPU. (contention measures)

Can do the first 3 options

# virsh dominfo QEMUGuest1
Id:             1
Name:           QEMUGuest1
UUID:           c7a5fdbd-edaf-9455-926a-d65c16db1809
OS Type:        hvm
State:          running
CPU(s):         1
CPU time:       98.8s
Max memory:     219200 kB
Used memory:    219200 kB

We don't currently have APIs to show disk & network I/O stats. We also don't
yet provide migration support.

>        6. What is kvm_stats ? How to interpret its output?

No idea :-)

>    c. Can one change memory, cpus, disks and networks on the fly for a running VM ? If so how ? 
>         This is critical for achiving dynamic resource management.

Not AFAIK. To be precise, libvirt provides APIs for that, but it requires better
guest support / paravirt device drivers. Also QEMU can do hot-add/remove of USB
devices.

> 4.  Shutdown : How does one send  command to  VM to shutdown normally. I  see only reset 
> and powerdown  commands only. Also, after shutdown, the image running in the machine 
> shuts down, but the VM keeps running. Shouldnt the process also die ?

A controlled shutdown isn't provided by QEMU/KVM yet. One would need to simulate
some kind of ACPI power management support in the guest so it saw a virtual
software power button press i guess.

> 5. Snapshots : where are snapshots saved ? I tried savevm with /tmp/x, I was expecting a 
>    file containing memory image... do these work differrently ?
>    Also, when one does loadvm, is the snapshot gone ? I mean can one restore the same 
>    snapshot multiple times ?

libvirt has APIs for snapshotting, but we've not implement them for QEMU
or KVM yet. You can save the file to any location you have write permissions
for. Its not a plain memory image because it also contains info about the
state of all the various virtual devices.

The snapshot doesn't deal with changes in state on your disks. So if you restored
a snapshot, ran it for a while doing disk IO and then quit & tried to restore the
snapshot again your disks would be out of sync.

That said if one could take snapshots of the disks (eg with LVM, or QCOW) then
it ought to be possible to use a snapshot multiple times over.

Dan.
-- 
|=- Red Hat, Engineering, Emerging Technologies, Boston.  +1 978 392 2496 -=|
|=-           Perl modules: http://search.cpan.org/~danberr/              -=|
|=-               Projects: http://freshmeat.net/~danielpb/               -=|
|=-  GnuPG: 7D3B9505   F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505  -=| 


       
---------------------------------
Ahhh...imagining that irresistible "new car" smell?
 Check outnew cars at Yahoo! Autos.

[-- Attachment #2.1.2: Type: text/html, Size: 7892 bytes --]

[-- Attachment #3: Type: text/plain, Size: 286 bytes --]

-------------------------------------------------------------------------
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/

[-- Attachment #4: Type: text/plain, Size: 186 bytes --]

_______________________________________________
kvm-devel mailing list
kvm-devel-5NWGOfrQmneRv+LV9MX5uipxlwaOVQ5f@public.gmane.org
https://lists.sourceforge.net/lists/listinfo/kvm-devel

Re: Fwd: Re: KVM management and API

[Daniel P. Berrange]

On Mon, Apr 30, 2007 at 07:01:55PM -0700, jd wrote:
> Does libvirt use ps -ef | grep qemu to find out qemu processes ? Or it 
> manages only ones that are started using libvirt. Using some naming scheme.

We only manage instances started by libvirt. We (briefly) considered 'ps ef'
but this is not at all scalable, since getting a list of active domains is
quite a frequently invoked operation. In latest CVS of QEMU there is a new
option '-daemonize' and '-name string' which will make it leave a UNIX domain
socket & pid file in a predictable location, but since that's not in any
release yet, we can't rely on it. There is also the issue of manage configs
of instances which aren't running & hence have no process.

> How do you know if a running process is using kvm kernel module or not ? 

Since libvirt only manages instances it invokes, it knows whether we 
launched the KVM version. Ideally a monitor command would be able to tell
us what accelerator is in use.

> libvirt seems to be connecting to  qemu console through qemud ? right ? 
> You have done quite a bit of work to send command and parse. !! Isnt 
> this error prone ? (will code for parsing work in different char set ? 
> or qemu console is always guarenteed to be in english ?)

The console is always in English & even if it were translated in the future,
we can control the environment the process is launched it to force it to
be in the C/POSIX locale. Anthony Liguori has done some work in recent 
times to make some of the monitor commands easier / more predictable to
parse, so I'm not really worried about it currently.

The qemu monitor is really the recommended (and only) interface for 
management apps which need to control aspects of QEMU's operation
on-the-fly. We use it for pause/resume, save/restore at this time. I
expect we'll make further use of it in the future, for example to set
the VNC password securely (passsing passwds on the command line or 
via environment vars is not good). 

> How do you find out what % of allocated memory is actually used by VM ? 

Any app on the host can only ever tell how much memory was allocated
to a VM. To determine whether an OS is actually using all of this
allocation requires some form of agent in the guest OS to report back
on info.

> For cpu /proc seems to be used.. which seems fine, but will not work 
> for QEMU on Windows.

Indeed not - it won't even work on non-Linux systems. There isn't any form
of standard API in POSIX world for getting process CPU usage, so we're expecting
to have to implement this for each OS ported to. The code is well isolated
so porting it won't be hard.

Regards,
Dan.
-- 
|=- Red Hat, Engineering, Emerging Technologies, Boston.  +1 978 392 2496 -=|
|=-           Perl modules: http://search.cpan.org/~danberr/              -=|
|=-               Projects: http://freshmeat.net/~danielpb/               -=|
|=-  GnuPG: 7D3B9505   F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505  -=| 

-------------------------------------------------------------------------
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/

Re: Fwd: Re: KVM management and API

[Heidi Eckhart]

Hi Jd,

you brought up a very important discussion and please let me raise another
point to the current low level, implementation specific discussion.

WBEM (Web-Based Enterprise Management) and CIM (Common Information Model)
is the industry defined and accepted Standard for doing different kinds of
systems management. I guess you plan/are writing a systems management
application for KVM, so this technology might be of interest for you -
especially the SMASH profile (www.dmtf.org/standards/mgmt/smash/) and the
"System Virtualization, Partitioning and Clustering" profile
(www.dmtf.org/about/committees/SVPCCharter.pdf). CIM is a conceptional
framework consisting of an object-oriented meta-model and the CIM Schema (
comprehensive set of classes and objects providing effective building
blocks to model system resources).

CIM management application tasks are based on profiles and the CIM classes
part of them. This enables the management application to become independent
of product internal interfaces, which avoids the mentioned error phrone and
not well maintainable code. The resource access, discussed in the mails
before, is done by the so called CIM providers. They handle the
compatibility issues like changing interfaces/commands transparent for the
management application, while product specific features stay accessible.
They are modeled and included to the schema.

Our team has already some years of experience in enabling Linux for CIM and
we plan to CIM-enable KVM by writing the providers based on DMTF's SVPC
profile.

Mit freundlichen Grüßen/Regards

Heidi Eckhart
Software Engineer
Linux Technology Center - Systems Management

heidineu-tA70FqPdS9bQT0dZR+AlfA@public.gmane.org / ++49-(0)35843-22661 / ++49-(0)7031-16-3193
http://sblim.wiki.sourceforge.net

*********************************************************************************

IBM Deutschland Entwicklung GmbH
Vorsitzender des Aufsichtsrats: Johann Weihen
Geschäftsführung: Herbert Kircher
Sitz der Gesellschaft: Böblingen
Registergericht: Amtsgericht Stuttgart, HRB 243294


                                                                           
             "Daniel P.                                                    
             Berrange"                                                     
             <berrange@redhat.                                          To 
             com>                      jd <jdsw2002-/E1597aS9LQAvxtiuMwx3w@public.gmane.org>             
             Sent by:                                                   cc 
             kvm-devel-bounces         kvm-devel                           
             @lists.sourceforg         <kvm-devel-5NWGOfrQmneRv+LV9MX5uipxlwaOVQ5f@public.gmane.org>   
             e.net                                                 Subject 
                                       Re: [kvm-devel] Fwd: Re:  KVM       
                                       management and API                  
             05/01/07 04:18 AM                                             
                                                                           
                                                                           
             Please respond to                                             
                "Daniel P.                                                 
                 Berrange"                                                 
             <berrange@redhat.                                             
                   com>                                                    
                                                                           
                                                                           




On Mon, Apr 30, 2007 at 07:01:55PM -0700, jd wrote:
> Does libvirt use ps -ef | grep qemu to find out qemu processes ? Or it
> manages only ones that are started using libvirt. Using some naming
scheme.

We only manage instances started by libvirt. We (briefly) considered 'ps
ef'
but this is not at all scalable, since getting a list of active domains is
quite a frequently invoked operation. In latest CVS of QEMU there is a new
option '-daemonize' and '-name string' which will make it leave a UNIX
domain
socket & pid file in a predictable location, but since that's not in any
release yet, we can't rely on it. There is also the issue of manage configs
of instances which aren't running & hence have no process.

> How do you know if a running process is using kvm kernel module or not ?

Since libvirt only manages instances it invokes, it knows whether we
launched the KVM version. Ideally a monitor command would be able to tell
us what accelerator is in use.

> libvirt seems to be connecting to  qemu console through qemud ? right ?
> You have done quite a bit of work to send command and parse. !! Isnt
> this error prone ? (will code for parsing work in different char set ?
> or qemu console is always guarenteed to be in english ?)

The console is always in English & even if it were translated in the
future,
we can control the environment the process is launched it to force it to
be in the C/POSIX locale. Anthony Liguori has done some work in recent
times to make some of the monitor commands easier / more predictable to
parse, so I'm not really worried about it currently.

The qemu monitor is really the recommended (and only) interface for
management apps which need to control aspects of QEMU's operation
on-the-fly. We use it for pause/resume, save/restore at this time. I
expect we'll make further use of it in the future, for example to set
the VNC password securely (passsing passwds on the command line or
via environment vars is not good).

> How do you find out what % of allocated memory is actually used by VM ?

Any app on the host can only ever tell how much memory was allocated
to a VM. To determine whether an OS is actually using all of this
allocation requires some form of agent in the guest OS to report back
on info.

> For cpu /proc seems to be used.. which seems fine, but will not work
> for QEMU on Windows.

Indeed not - it won't even work on non-Linux systems. There isn't any form
of standard API in POSIX world for getting process CPU usage, so we're
expecting
to have to implement this for each OS ported to. The code is well isolated
so porting it won't be hard.

Regards,
Dan.
--
|=- Red Hat, Engineering, Emerging Technologies, Boston.  +1 978 392 2496
-=|
|=-           Perl modules: http://search.cpan.org/~danberr/
-=|
|=-               Projects: http://freshmeat.net/~danielpb/
-=|
|=-  GnuPG: 7D3B9505   F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505
-=|

-------------------------------------------------------------------------
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/
_______________________________________________
kvm-devel mailing list
kvm-devel-5NWGOfrQmneRv+LV9MX5uipxlwaOVQ5f@public.gmane.org
https://lists.sourceforge.net/lists/listinfo/kvm-devel



-------------------------------------------------------------------------
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/