[Qemu-devel] vhost-user: migration?

[Qemu-devel] vhost-user: migration?

[Michael S. Tsirkin]

Just a query whether migration works for you
guys with vhost-user and snabbswitch.

vhost user code seems to have enough hooks in
place to make it work, but was it tested
in practice?


-- 
MST

Re: [Qemu-devel] vhost-user: migration?

[Nikolay Nikolaev]

[-- Attachment #1: Type: text/plain, Size: 470 bytes --]

Hello,

On Wed, Jun 18, 2014 at 12:33 PM, Michael S. Tsirkin <mst@redhat.com> wrote:

> Just a query whether migration works for you
> guys with vhost-user and snabbswitch.
>
> I am not aware of anyone ever did such tests.

vhost user code seems to have enough hooks in
> place to make it work, but was it tested
> in practice?
>
>
Can you give some pointers of a simple/quick test scenario. Maybe we'll be
able to set it up.


>
> --
> MST
>

regards,
Nikolay Nikolaev

[-- Attachment #2: Type: text/html, Size: 1272 bytes --]

Re: [Qemu-devel] vhost-user: migration?

[Michael S. Tsirkin]

On Wed, Jun 18, 2014 at 02:24:32PM +0300, Nikolay Nikolaev wrote:
> Hello,
> 
> On Wed, Jun 18, 2014 at 12:33 PM, Michael S. Tsirkin <mst@redhat.com> wrote:
> 
>     Just a query whether migration works for you
>     guys with vhost-user and snabbswitch.
> 
> 
> I am not aware of anyone ever did such tests.
> 
> 
>     vhost user code seems to have enough hooks in
>     place to make it work, but was it tested
>     in practice?
>    
> 
> 
> Can you give some pointers of a simple/quick test scenario. Maybe we'll be able
> to set it up.


Basically migrate while doing networking :)

Does your vhost server have code to set bits in the memory log
when logging is enabled?
The log address is supplied to VHOST_SET_LOG_BASE and log_guest_addr.


> 
>    
>     --
>     MST
> 
> 
> regards,
> Nikolay Nikolaev

Re: [Qemu-devel] vhost-user: migration?

[Nikolay Nikolaev]

[-- Attachment #1: Type: text/plain, Size: 1015 bytes --]

On Wed, Jun 18, 2014 at 2:29 PM, Michael S. Tsirkin <mst@redhat.com> wrote:

> On Wed, Jun 18, 2014 at 02:24:32PM +0300, Nikolay Nikolaev wrote:
> > Hello,
> >
> > On Wed, Jun 18, 2014 at 12:33 PM, Michael S. Tsirkin <mst@redhat.com>
> wrote:
> >
> >     Just a query whether migration works for you
> >     guys with vhost-user and snabbswitch.
> >
> >
> > I am not aware of anyone ever did such tests.
> >
> >
> >     vhost user code seems to have enough hooks in
> >     place to make it work, but was it tested
> >     in practice?
> >
> >
> >
> > Can you give some pointers of a simple/quick test scenario. Maybe we'll
> be able
> > to set it up.
>
>
> Basically migrate while doing networking :)
>
> Does your vhost server have code to set bits in the memory log
> when logging is enabled?
>

No - we didn't implement this feature.

The log address is supplied to VHOST_SET_LOG_BASE and log_guest_addr.
>
>
> >
> >
> >     --
> >     MST
> >
> >
> > regards,
> > Nikolay Nikolaev
>

regards,
Nikolay Nikolaev

[-- Attachment #2: Type: text/html, Size: 1871 bytes --]

Re: [Qemu-devel] vhost-user: migration?

[Michael S. Tsirkin]

On Wed, Jun 18, 2014 at 02:41:15PM +0300, Nikolay Nikolaev wrote:
> 
> 
> 
> On Wed, Jun 18, 2014 at 2:29 PM, Michael S. Tsirkin <mst@redhat.com> wrote:
> 
>     On Wed, Jun 18, 2014 at 02:24:32PM +0300, Nikolay Nikolaev wrote:
>     > Hello,
>     >
>     > On Wed, Jun 18, 2014 at 12:33 PM, Michael S. Tsirkin <mst@redhat.com>
>     wrote:
>     >
>     >     Just a query whether migration works for you
>     >     guys with vhost-user and snabbswitch.
>     >
>     >
>     > I am not aware of anyone ever did such tests.
>     >
>     >
>     >     vhost user code seems to have enough hooks in
>     >     place to make it work, but was it tested
>     >     in practice?
>     >
>     >
>     >
>     > Can you give some pointers of a simple/quick test scenario. Maybe we'll
>     be able
>     > to set it up.
> 
> 
>     Basically migrate while doing networking :)
> 
>     Does your vhost server have code to set bits in the memory log
>     when logging is enabled?
> 
> 
> No - we didn't implement this feature.

Interesting. In that case vhost user should for now
block migration.
You must also check log feature and return failure.


> 
>     The log address is supplied to VHOST_SET_LOG_BASE and log_guest_addr.
> 
> 
>     >
>     >
>     >     --
>     >     MST
>     >
>     >
>     > regards,
>     > Nikolay Nikolaev
> 
> 
> regards,
> Nikolay Nikolaev

Re: [Qemu-devel] vhost-user: migration?

[Nikolay Nikolaev]

[-- Attachment #1: Type: text/plain, Size: 1744 bytes --]

On Wed, Jun 18, 2014 at 3:35 PM, Michael S. Tsirkin <mst@redhat.com> wrote:

> On Wed, Jun 18, 2014 at 02:41:15PM +0300, Nikolay Nikolaev wrote:
> >
> >
> >
> > On Wed, Jun 18, 2014 at 2:29 PM, Michael S. Tsirkin <mst@redhat.com>
> wrote:
> >
> >     On Wed, Jun 18, 2014 at 02:24:32PM +0300, Nikolay Nikolaev wrote:
> >     > Hello,
> >     >
> >     > On Wed, Jun 18, 2014 at 12:33 PM, Michael S. Tsirkin <
> mst@redhat.com>
> >     wrote:
> >     >
> >     >     Just a query whether migration works for you
> >     >     guys with vhost-user and snabbswitch.
> >     >
> >     >
> >     > I am not aware of anyone ever did such tests.
> >     >
> >     >
> >     >     vhost user code seems to have enough hooks in
> >     >     place to make it work, but was it tested
> >     >     in practice?
> >     >
> >     >
> >     >
> >     > Can you give some pointers of a simple/quick test scenario. Maybe
> we'll
> >     be able
> >     > to set it up.
> >
> >
> >     Basically migrate while doing networking :)
> >
> >     Does your vhost server have code to set bits in the memory log
> >     when logging is enabled?
> >
> >
> > No - we didn't implement this feature.
>
> Interesting. In that case vhost user should for now
> block migration.
>
Any place in the source where we can see example of similar behaviour?


> You must also check log feature and return failure.
>
Should this happen on the QEMU side? Is this the log feature of the device?

regards,
Nikolay Nikolaev


>
>
> >
> >     The log address is supplied to VHOST_SET_LOG_BASE and log_guest_addr.
> >
> >
> >     >
> >     >
> >     >     --
> >     >     MST
> >     >
> >     >
> >     > regards,
> >     > Nikolay Nikolaev
> >
> >
> > regards,
> > Nikolay Nikolaev
>

[-- Attachment #2: Type: text/html, Size: 2997 bytes --]

Re: [Qemu-devel] vhost-user: migration?

[Michael S. Tsirkin]

On Wed, Jun 18, 2014 at 04:37:57PM +0300, Nikolay Nikolaev wrote:
> 
> 
> 
> On Wed, Jun 18, 2014 at 3:35 PM, Michael S. Tsirkin <mst@redhat.com> wrote:
> 
>     On Wed, Jun 18, 2014 at 02:41:15PM +0300, Nikolay Nikolaev wrote:
>     >
>     >
>     >
>     > On Wed, Jun 18, 2014 at 2:29 PM, Michael S. Tsirkin <mst@redhat.com>
>     wrote:
>     >
>     >     On Wed, Jun 18, 2014 at 02:24:32PM +0300, Nikolay Nikolaev wrote:
>     >     > Hello,
>     >     >
>     >     > On Wed, Jun 18, 2014 at 12:33 PM, Michael S. Tsirkin <
>     mst@redhat.com>
>     >     wrote:
>     >     >
>     >     >     Just a query whether migration works for you
>     >     >     guys with vhost-user and snabbswitch.
>     >     >
>     >     >
>     >     > I am not aware of anyone ever did such tests.
>     >     >
>     >     >
>     >     >     vhost user code seems to have enough hooks in
>     >     >     place to make it work, but was it tested
>     >     >     in practice?
>     >     >
>     >     >
>     >     >
>     >     > Can you give some pointers of a simple/quick test scenario. Maybe
>     we'll
>     >     be able
>     >     > to set it up.
>     >
>     >
>     >     Basically migrate while doing networking :)
>     >
>     >     Does your vhost server have code to set bits in the memory log
>     >     when logging is enabled?
>     >
>     >
>     > No - we didn't implement this feature.
> 
>     Interesting. In that case vhost user should for now
>     block migration.
> 
> Any place in the source where we can see example of similar behaviour?
>  
> 
>     You must also check log feature and return failure.
> 
> Should this happen on the QEMU side? Is this the log feature of the device?
> 
> regards,
> Nikolay Nikolaev
>  

Well, QEMU is buggy in that it tries to set log
flag without checking.
I will fix that.


However
1.  Your server must not publish the VHOST_F_LOG_ALL feature.

On vhost side it should validate features set
match the supported mask.
Also for flags in the virtqueue address structure.
Please validate as much as possible, e.g.
validate all unused fields are zero.
If things work by mistake now you will have to maintain
them indefinitely.



> 
> 
>     >
>     >     The log address is supplied to VHOST_SET_LOG_BASE and log_guest_addr.
>     >
>     >
>     >     >
>     >     >
>     >     >     --
>     >     >     MST
>     >     >
>     >     >
>     >     > regards,
>     >     > Nikolay Nikolaev
>     >
>     >
>     > regards,
>     > Nikolay Nikolaev
> 
> 

Re: [Qemu-devel] vhost-user: migration?

[Zhang Haoyu]

Hi,
what's the status of migration support for vhost-user?

Thanks,
Zhang Haoyu
                                       
On 2014-06-18 22:07:49, Michael S. Tsirkin wrote:
> On Wed, Jun 18, 2014 at 04:37:57PM +0300, Nikolay Nikolaev wrote:
> > 
> > 
> > 
> > On Wed, Jun 18, 2014 at 3:35 PM, Michael S. Tsirkin <mst@redhat.com> wrote:
> > 
> >     On Wed, Jun 18, 2014 at 02:41:15PM +0300, Nikolay Nikolaev wrote:
> >     >
> >     >
> >     >
> >     > On Wed, Jun 18, 2014 at 2:29 PM, Michael S. Tsirkin <mst@redhat.com>
> >     wrote:
> >     >
> >     >     On Wed, Jun 18, 2014 at 02:24:32PM +0300, Nikolay Nikolaev wrote:
> >     >     > Hello,
> >     >     >
> >     >     > On Wed, Jun 18, 2014 at 12:33 PM, Michael S. Tsirkin <
> >     mst@redhat.com>
> >     >     wrote:
> >     >     >
> >     >     >     Just a query whether migration works for you
> >     >     >     guys with vhost-user and snabbswitch.
> >     >     >
> >     >     >
> >     >     > I am not aware of anyone ever did such tests.
> >     >     >
> >     >     >
> >     >     >     vhost user code seems to have enough hooks in
> >     >     >     place to make it work, but was it tested
> >     >     >     in practice?
> >     >     >
> >     >     >
> >     >     >
> >     >     > Can you give some pointers of a simple/quick test scenario. Maybe
> >     we'll
> >     >     be able
> >     >     > to set it up.
> >     >
> >     >
> >     >     Basically migrate while doing networking :)
> >     >
> >     >     Does your vhost server have code to set bits in the memory log
> >     >     when logging is enabled?
> >     >
> >     >
> >     > No - we didn't implement this feature.
> > 
> >     Interesting. In that case vhost user should for now
> >     block migration.
> > 
> > Any place in the source where we can see example of similar behaviour?
> >  
> > 
> >     You must also check log feature and return failure.
> > 
> > Should this happen on the QEMU side? Is this the log feature of the device?
> > 
> > regards,
> > Nikolay Nikolaev
> >  
> 
> Well, QEMU is buggy in that it tries to set log
> flag without checking.
> I will fix that.
> 
> 
> However
> 1.  Your server must not publish the VHOST_F_LOG_ALL feature.
> 
> On vhost side it should validate features set
> match the supported mask.
> Also for flags in the virtqueue address structure.
> Please validate as much as possible, e.g.
> validate all unused fields are zero.
> If things work by mistake now you will have to maintain
> them indefinitely.
> 
> 
> 
> > 
> > 
> >     >
> >     >     The log address is supplied to VHOST_SET_LOG_BASE and log_guest_addr.
> >     >
> >     >
> >     >     >
> >     >     >
> >     >     >     --
> >     >     >     MST
> >     >     >
> >     >     >
> >     >     > regards,
> >     >     > Nikolay Nikolaev
> >     >
> >     >
> >     > regards,
> >     > Nikolay Nikolaev