On Tue, 6 Jan 2026 14:48:14 -0500
Peter Xu <[email protected]> wrote:

> On Sun, Jan 04, 2026 at 01:44:52PM +0800, Zhang Chen wrote:
> > On Tue, Dec 30, 2025 at 11:02 PM Peter Xu <[email protected]> wrote:  
> > >
> > > On Tue, Dec 30, 2025 at 03:05:43PM +0100, Lukas Straub wrote:  
> > > > Hello everyone,
> > > > This adds COLO multifd support and migration unit tests for COLO 
> > > > migration
> > > > and failover.  
> > >
> > > Hi, Lukas,
> > >
> > > I'll review the series after the new year.
> > >
> > > Could you still introduce some background on how you're deploying COLO?  
> > > Do
> > > you use it in production, or for fun?
> > >
> > > COLO is still a nice and interesting feature, said that, COLO has quite a
> > > lot of code plugged into migration core.  I wished it's like a multifd
> > > compressor which was much more self-contained, but it's not.  I wished we
> > > can simplify the code in QEMU migration.
> > >
> > > We've talked it through before with current COLO maintainers, it looks to
> > > me there aren't really much users using it in production, meanwhile COLO
> > > doesn't look like a feature to benefit individual QEMU users either.
> > >
> > > I want to study the use case of COLO in status quo, and evaluate how much
> > > effort we should put on it in the future.  Note that if it's for fun we 
> > > can
> > > always use a stable branch which will be there forever.  We'll need to
> > > think about QEMU evolving in the future, and what's best for QEMU.
> > >
> > > Thanks,
> > >  
> > 
> > Hi Lukas and Peter,  
> 
> Hi, Chen,
> 
> > 
> > Thanks for this series, I will support for background info if Peter
> > have any questions.  
> 
> Thanks, I believe my major question so far was, whether we should deprecate
> COLO in migration framework. :)
> 
> The netfilters and rest can be discussed separately.
> 
> Now looking back at my initial ask in Zhijian's fix, I still agree with
> Zhijian on these two points mentioned:
> 
> https://lore.kernel.org/all/[email protected]/
> 
> That is:
> 
>         - Active users who depend on it.
>         - A unit test for the COLO framework.
> 
> Meanwhile, I can't see how COLO would win if to be compared with some
> app-level HA infrastructure.. considering the overhead it requires on
> running two VMs and compare every packet.
> 
> Lukas, thanks for trying to fix the 2nd.  I apologize that I still
> requested you to send these patches, without further raising the attention
> that I still want to discuss deprecation.  I don't think anyone yet proved
> we should keep COLO.  I do plan to send one patch adding COLO framework to
> deprecation, if nobody would stop me in a week justifying question 1 above.

Hello Peter,

I am a consultant on open-source high availability and fault tolaerance
solutions. I provide a complete cluster management solution with
automatic failover and failback for Qemu COLO.

Qemu COLOs lockstepping architecture has a big performance advantage
and it outperforms the market leader by 10x-100x in latency.
No one else provides this unique architecture.

I have customers that depend on this.

I occasionally get inquiries about Qemu COLO even without doing
any kind of marketing. So there is a general interest for this.

Also, Canonical considers providing this to one of their customers.

Regards,
Lukas Straub


> 
> We kind of proved almost nobody is actively using COLO anymore in the past
> few releases.  If nobody is using COLO, we should simply drop it.
> 
> > And CC Hailiang Zhang, although he hasn't replied to emails for a long time.
> > If no one objects, I think Lukas can replease Hailiang for COLO Framework.
> > 
> > COLO Framework
> > M: Hailiang Zhang <[email protected]>
> > S: Maintained
> > F: migration/colo*
> > F: include/migration/colo.h
> > F: include/migration/failover.h
> > F: docs/COLO-FT.txt  
> 
> Right, this is also another reason why I think we may want to deprecate
> COLO framework.

I will take over maintainership.

> 
> Since I requested this series (sorry again, Lukas), I'll review it today no
> matter if we decide to merge this series at last, or deprecate COLO
> framework.
> 
> Thanks,
> 
> > 
> > Thanks
> > Chen
> >   
> > > >
> > > > Regards,
> > > > Lukas
> > > >
> > > > Signed-off-by: Lukas Straub <[email protected]>
> > > > ---
> > > > Lukas Straub (3):
> > > >       multifd: Add colo support
> > > >       migration-test: Add -snapshot option for COLO
> > > >       migration-test: Add COLO migration unit test
> > > >
> > > >  migration/meson.build              |   2 +-
> > > >  migration/multifd-colo.c           |  57 ++++++++++++++++++
> > > >  migration/multifd-colo.h           |  26 +++++++++
> > > >  migration/multifd.c                |  14 ++++-
> > > >  tests/qtest/meson.build            |   7 ++-
> > > >  tests/qtest/migration-test.c       |   1 +
> > > >  tests/qtest/migration/colo-tests.c | 115 
> > > > +++++++++++++++++++++++++++++++++++++
> > > >  tests/qtest/migration/framework.c  |  69 +++++++++++++++++++++-
> > > >  tests/qtest/migration/framework.h  |  10 ++++
> > > >  9 files changed, 294 insertions(+), 7 deletions(-)
> > > > ---
> > > > base-commit: 942b0d378a1de9649085ad6db5306d5b8cef3591
> > > > change-id: 20251230-colo_unit_test_multifd-8bf58dcebd46
> > > >
> > > > Best regards,
> > > > --
> > > > Lukas Straub <[email protected]>
> > > >  
> > >
> > > --
> > > Peter Xu
> > >  
> >   
> 

Attachment: pgprL40fNqxpu.pgp
Description: OpenPGP digital signature

Reply via email to