status of non-live migration of HVM with libvirt

status of non-live migration of HVM with libvirt

[Olaf Hering]

[-- Attachment #1.1: Type: text/plain, Size: 426 bytes --]

With 'xl migrate' the migration is always live, while 'virsh migrate' will do an offline migration unless '--live' is used. Such non-live migration will break with HVM because the sending side does not seem to unlock the qcow2 image.

I wonder if its worth keeping (and fixing) the concept of an "offline migration", or if libvirt should be changed to either require '--live' or silently always do live migrations.

Olaf

[-- Attachment #1.2: Digitale Signatur von OpenPGP --]
[-- Type: application/pgp-signature, Size: 195 bytes --]

[-- Attachment #2: Type: text/plain, Size: 157 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel

Re: status of non-live migration of HVM with libvirt

[Olaf Hering]

[-- Attachment #1.1: Type: text/plain, Size: 820 bytes --]

Am Fri, 4 Jan 2019 16:57:55 +0100
schrieb Olaf Hering <olaf@aepfle.de>:

> worth keeping (and fixing) the concept of an "offline migration"

And regarding the fix, it looks like qmp_xen_save_devices_state does not need the concept of "live". Just shutdown the blockdevices and be done with it?

+++ b/migration/savevm.c
@@ -2293,7 +2293,7 @@ void qmp_xen_save_devices_state(const char *filename, bool has_live, bool live,
          * So call bdrv_inactivate_all (release locks) here to let the other
          * side of the migration take controle of the images.
          */
-        if (live && !saved_vm_running) {
+        if (!saved_vm_running) {
             ret = bdrv_inactivate_all();
             if (ret) {
                 error_setg(errp, "%s: bdrv_inactivate_all() failed (%d)",


Olaf

[-- Attachment #1.2: Digitale Signatur von OpenPGP --]
[-- Type: application/pgp-signature, Size: 195 bytes --]

[-- Attachment #2: Type: text/plain, Size: 157 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel

Re: status of non-live migration of HVM with libvirt

[Olaf Hering]

[-- Attachment #1.1: Type: text/plain, Size: 811 bytes --]

Am Fri, 4 Jan 2019 17:48:31 +0100
schrieb Olaf Hering <olaf@aepfle.de>:

> Am Fri, 4 Jan 2019 16:57:55 +0100
> schrieb Olaf Hering <olaf@aepfle.de>:
> 
> > worth keeping (and fixing) the concept of an "offline migration"  
> 
> And regarding the fix, it looks like qmp_xen_save_devices_state does not need the concept of "live". Just shutdown the blockdevices and be done with it?

Anthony, it looks like 5d6c599fe1 ("migration, xen: Fix block image lock issue on live migration") went too far. Why does stopping the disks at the very end of the save/migrate process depend on "live"? During a migration the disks have to be released either way. During "xl save" the domU may continue to run, if '-c' was specified. 

It seems the 'live' parameter for xen-save-devices-state is not needed.

Olaf

[-- Attachment #1.2: Digitale Signatur von OpenPGP --]
[-- Type: application/pgp-signature, Size: 195 bytes --]

[-- Attachment #2: Type: text/plain, Size: 157 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel

Re: status of non-live migration of HVM with libvirt

[Olaf Hering]

[-- Attachment #1.1: Type: text/plain, Size: 1010 bytes --]

Anthony,

looks like this fix missed qemu-4.0.

What is your opinion on this fix? I think there was zero reply...

Olaf

On Mon, Jan 07, Olaf Hering wrote:

> Am Fri, 4 Jan 2019 17:48:31 +0100
> schrieb Olaf Hering <olaf@aepfle.de>:
> 
> > Am Fri, 4 Jan 2019 16:57:55 +0100
> > schrieb Olaf Hering <olaf@aepfle.de>:
> > 
> > > worth keeping (and fixing) the concept of an "offline migration"  
> > 
> > And regarding the fix, it looks like qmp_xen_save_devices_state does not need the concept of "live". Just shutdown the blockdevices and be done with it?
> 
> Anthony, it looks like 5d6c599fe1 ("migration, xen: Fix block image lock issue on live migration") went too far. Why does stopping the disks at the very end of the save/migrate process depend on "live"? During a migration the disks have to be released either way. During "xl save" the domU may continue to run, if '-c' was specified. 
> 
> It seems the 'live' parameter for xen-save-devices-state is not needed.
> 
> Olaf


[-- Attachment #1.2: signature.asc --]
[-- Type: application/pgp-signature, Size: 195 bytes --]

[-- Attachment #2: Type: text/plain, Size: 157 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel

Re: [Xen-devel] status of non-live migration of HVM with libvirt

[Olaf Hering]

[-- Attachment #1.1: Type: text/plain, Size: 1010 bytes --]

Anthony,

looks like this fix missed qemu-4.0.

What is your opinion on this fix? I think there was zero reply...

Olaf

On Mon, Jan 07, Olaf Hering wrote:

> Am Fri, 4 Jan 2019 17:48:31 +0100
> schrieb Olaf Hering <olaf@aepfle.de>:
> 
> > Am Fri, 4 Jan 2019 16:57:55 +0100
> > schrieb Olaf Hering <olaf@aepfle.de>:
> > 
> > > worth keeping (and fixing) the concept of an "offline migration"  
> > 
> > And regarding the fix, it looks like qmp_xen_save_devices_state does not need the concept of "live". Just shutdown the blockdevices and be done with it?
> 
> Anthony, it looks like 5d6c599fe1 ("migration, xen: Fix block image lock issue on live migration") went too far. Why does stopping the disks at the very end of the save/migrate process depend on "live"? During a migration the disks have to be released either way. During "xl save" the domU may continue to run, if '-c' was specified. 
> 
> It seems the 'live' parameter for xen-save-devices-state is not needed.
> 
> Olaf


[-- Attachment #1.2: signature.asc --]
[-- Type: application/pgp-signature, Size: 195 bytes --]

[-- Attachment #2: Type: text/plain, Size: 157 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel

Re: status of non-live migration of HVM with libvirt

[Anthony PERARD]

On Mon, Jan 07, 2019 at 11:07:43AM +0100, Olaf Hering wrote:
> Am Fri, 4 Jan 2019 17:48:31 +0100
> schrieb Olaf Hering <olaf@aepfle.de>:
> 
> > Am Fri, 4 Jan 2019 16:57:55 +0100
> > schrieb Olaf Hering <olaf@aepfle.de>:
> > 
> > > worth keeping (and fixing) the concept of an "offline migration"  
> > 
> > And regarding the fix, it looks like qmp_xen_save_devices_state does not need the concept of "live". Just shutdown the blockdevices and be done with it?
> 
> Anthony, it looks like 5d6c599fe1 ("migration, xen: Fix block image
> lock issue on live migration") went too far. Why does stopping the
> disks at the very end of the save/migrate process depend on "live"?

Because "Snapshots don't need to inactivate images", see:
https://lists.xenproject.org/archives/html/xen-devel/2017-10/msg02782.html

I had to add the `live' parameter to fix the lock issue.

> During a migration the disks have to be released either way. During "xl save" the domU may continue to run, if '-c' was specified. 

xl save -c is a snapshot, right? There isn't a second QEMU process that
is going to read/write the same disk.

> It seems the 'live' parameter for xen-save-devices-state is not needed.
> 
> Olaf



-- 
Anthony PERARD

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel

Re: [Xen-devel] status of non-live migration of HVM with libvirt

[Anthony PERARD]

On Mon, Jan 07, 2019 at 11:07:43AM +0100, Olaf Hering wrote:
> Am Fri, 4 Jan 2019 17:48:31 +0100
> schrieb Olaf Hering <olaf@aepfle.de>:
> 
> > Am Fri, 4 Jan 2019 16:57:55 +0100
> > schrieb Olaf Hering <olaf@aepfle.de>:
> > 
> > > worth keeping (and fixing) the concept of an "offline migration"  
> > 
> > And regarding the fix, it looks like qmp_xen_save_devices_state does not need the concept of "live". Just shutdown the blockdevices and be done with it?
> 
> Anthony, it looks like 5d6c599fe1 ("migration, xen: Fix block image
> lock issue on live migration") went too far. Why does stopping the
> disks at the very end of the save/migrate process depend on "live"?

Because "Snapshots don't need to inactivate images", see:
https://lists.xenproject.org/archives/html/xen-devel/2017-10/msg02782.html

I had to add the `live' parameter to fix the lock issue.

> During a migration the disks have to be released either way. During "xl save" the domU may continue to run, if '-c' was specified. 

xl save -c is a snapshot, right? There isn't a second QEMU process that
is going to read/write the same disk.

> It seems the 'live' parameter for xen-save-devices-state is not needed.
> 
> Olaf



-- 
Anthony PERARD

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel

Re: status of non-live migration of HVM with libvirt

[Olaf Hering]

[-- Attachment #1.1: Type: text/plain, Size: 344 bytes --]

Am Wed, 10 Apr 2019 17:25:04 +0100
schrieb Anthony PERARD <anthony.perard@citrix.com>:

> Because "Snapshots don't need to inactivate images", see:
> https://lists.xenproject.org/archives/html/xen-devel/2017-10/msg02782.html
> 
> I had to add the `live' parameter to fix the lock issue.

Thanks, I will double check this part.

Olaf

[-- Attachment #1.2: Digitale Signatur von OpenPGP --]
[-- Type: application/pgp-signature, Size: 195 bytes --]

[-- Attachment #2: Type: text/plain, Size: 157 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel

Re: [Xen-devel] status of non-live migration of HVM with libvirt

[Olaf Hering]

[-- Attachment #1.1: Type: text/plain, Size: 344 bytes --]

Am Wed, 10 Apr 2019 17:25:04 +0100
schrieb Anthony PERARD <anthony.perard@citrix.com>:

> Because "Snapshots don't need to inactivate images", see:
> https://lists.xenproject.org/archives/html/xen-devel/2017-10/msg02782.html
> 
> I had to add the `live' parameter to fix the lock issue.

Thanks, I will double check this part.

Olaf

[-- Attachment #1.2: Digitale Signatur von OpenPGP --]
[-- Type: application/pgp-signature, Size: 195 bytes --]

[-- Attachment #2: Type: text/plain, Size: 157 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel