Re: [Xen-devel] [PATCH v3 COLOPre 11/26] tools/libxl: introduce a new API libxl__domain_restore() to load qemu state

[Ian Campbell <ian.campbell@xxxxxxxxxx>]

On Tue, 2015-06-30 at 18:04 +0800, Yang Hongyang wrote:
> 
> On 06/30/2015 12:38 AM, Ian Campbell wrote:
> > On Thu, 2015-06-25 at 14:25 +0800, Yang Hongyang wrote:
> >> Secondary vm is running in colo mode. So we will do
> >> the following things again and again:
> >> 1. suspend both primay vm and secondary vm
> >> 2. sync the state
> >> 3. resume both primary vm and secondary vm
> >> We will send qemu's state each time in step2, and
> >> slave's qemu should read it each time before resuming
> >> secondary vm. Introduce a new API libxl__domain_restore()
> >> to do it. This API should be called before resuming
> >> secondary vm.
> >
> > I think before this patch the state was passed to qemu as a parameter
> > when it was launched, is that correct? If so then that would be worth
> > mentioning for completeness.
> 
> Inaccurate I think. What you said before is the normal migration, in that
> case, yes, the state was passed to qemu as a parameter.

Which is what I think I said.

> With COLO, the
> first step is live migration, so the state is still passed to qemu as a
> parameter when the live migration ended. The new introduced API only used
> when we need to restore the DM state after a checkpoint, at this point,
> guest QEMU already started, we can not pass the state as a parameter like
> we do on first boot, so we introduce this API to restore the state after
> QEMU has started.

This sort of explanation would be good to have in the commit message,
especially due to the subtlety around first and subsequent checkpoints.

> >> +int libxl__domain_restore(libxl__gc *gc, uint32_t domid)
> >
> > We don't have any libxl__domain_save counterpart, but we do have
> > libxl__domain_save_device_model, so I wonder if the upcoming callers
> > ought to just call that direct? Especially given that this function
> > isn't any kind of generic domain restore, but has rather specific
> > functionality (in particular it fails for PV guests).
> 
> Maybe we just introduce libxl__domain_restore_device_model() and call
> this when needed, discard the new libxl__domain_restore() API, what do
> you think?

Yes, that's what I think I meant to say (but didn't).

Ian.



_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel