Re: [Qemu-devel] [PATCH v2 0/3] migration capability to discard the migrated ram pages

["Dr. David Alan Gilbert" <dgilbert@redhat.com>]

* Pavel Butsykin (pbutsykin@virtuozzo.com) wrote:
> On 07.03.2017 17:46, Dr. David Alan Gilbert wrote:
> > * Pavel Butsykin (pbutsykin@virtuozzo.com) wrote:
> > > On 07.03.2017 16:56, Dr. David Alan Gilbert wrote:
> > > > * Pavel Butsykin (pbutsykin@virtuozzo.com) wrote:
> > > > > On 14.02.2017 17:02, Dr. David Alan Gilbert wrote:
> > > > > > Hi Pavel,
> > > > > >      I was curious, having merged this, how you're using postcopy; do you switch
> > > > > > into postcopy mode immediately or wait until the first sync or what?
> > > > > > Do you find yourself in postcopy mode long enough that it's worth
> > > > > > doing the release?  If so on what size VMs are you working with?
> > > > > > 
> > > > > > Dave
> > > > > 
> > > > > Hi Dave,
> > > > > I'm sorry it took so long, I was on vacation. And thanks for the quick
> > > > > review and merge!
> > > > > 
> > > > > We want to use this function to update Qemu without rebooting VMs. It
> > > > > looks like a local migration on the same host. Now, switching into
> > > > > postcopy mode is done immediately, the size of VMs can be very
> > > > > different.
> > > > 
> > > > Thanks! Do you have libvirt magic to do that update?
> > > 
> > > Yes, there are patches that add a flag "--local" for migration. The
> > > flag adds start incoming VM and other magic about replacement source on
> > > destination.
> > 
> > Nice; are those posted to libvirt's list somewhere?
> 
> Not yet, but we're going to do it.

Great, I look forward to that.

Dave

> 
> > Dave
> > 
> > > > Dave
> > > > 
> > > > > > * Pavel Butsykin (pbutsykin@virtuozzo.com) wrote:
> > > > > > > This feature frees the migrated memory on the source during postcopy-ram
> > > > > > > migration. In the second step of postcopy-ram migration when the source vm
> > > > > > > is put on pause we can free unnecessary memory. It will allow, in particular,
> > > > > > > to start relaxing the memory stress on the source host in a load-balancing
> > > > > > > scenario.
> > > > > > > 
> > > > > > > Changes from v1:
> > > > > > > - changed name of the interfaces (discard to release)
> > > > > > > - fix make check error
> > > > > > > - add more comments to qemu_iovec_release_ram()
> > > > > > > - rebase on "Postcopy: Hugepage support" (David's patch series)
> > > > > > > - removed ram_discard_page for xbzrle
> > > > > > > - fix erroneous release memory in complete precopy (tie release-ram to postcopy)
> > > > > > > 
> > > > > > > Pavel Butsykin (3):
> > > > > > >      migration: add MigrationState arg for ram_save_/compressed_/page()
> > > > > > >      add 'release-ram' migrate capability
> > > > > > >      migration: discard non-dirty ram pages after the start of postcopy
> > > > > > > 
> > > > > > >     include/migration/migration.h |  2 ++
> > > > > > >     include/migration/qemu-file.h |  3 ++-
> > > > > > >     migration/migration.c         | 13 ++++++++++
> > > > > > >     migration/qemu-file.c         | 59 ++++++++++++++++++++++++++++++++++++++-----
> > > > > > >     migration/ram.c               | 56 ++++++++++++++++++++++++++++++++++------
> > > > > > >     qapi-schema.json              |  5 +++-
> > > > > > >     6 files changed, 121 insertions(+), 17 deletions(-)
> > > > > > > 
> > > > > > > --
> > > > > > > 2.11.0
> > > > > > > 
> > > > > > > 
> > > > > > --
> > > > > > Dr. David Alan Gilbert / dgilbert@redhat.com / Manchester, UK
> > > > > > 
> > > > --
> > > > Dr. David Alan Gilbert / dgilbert@redhat.com / Manchester, UK
> > > > 
> > --
> > Dr. David Alan Gilbert / dgilbert@redhat.com / Manchester, UK
> > 
--
Dr. David Alan Gilbert / dgilbert@redhat.com / Manchester, UK