Re: [PATCH] migration/postcopy: Fix high frequency sync

[Peter Xu]

On Wed, Mar 20, 2024 at 05:44:53PM -0400, peterx@redhat.com wrote:
> From: Peter Xu <peterx@redhat.com>
> 
> On current code base I can observe extremely high sync count during
> precopy, as long as one enables postcopy-ram=on before switchover to
> postcopy.
> 
> To provide some context of when we decide to do a full sync: we check
> must_precopy (which implies "data must be sent during precopy phase"), and
> as long as it is lower than the threshold size we calculated (out of
> bandwidth and expected downtime) we will kick off the slow sync.
> 
> However, when postcopy is enabled (even if still during precopy phase), RAM
> only reports all pages as can_postcopy, and report must_precopy==0.  Then
> "must_precopy <= threshold_size" mostly always triggers and enforces a slow
> sync for every call to migration_iteration_run() when postcopy is enabled
> even if not used.  That is insane.
> 
> It turns out it was a regress bug introduced in the previous refactoring in
> QEMU 8.0 in late 2022. Fix this by checking the whole RAM size rather than
> must_precopy, like before.  Not copy stable yet as many things changed, and
> even if this should be a major performance regression, no functional change
> has observed (and that's also probably why nobody found it).  I only notice
> this when looking for another bug reported by Nina.
> 
> When at it, cleanup a little bit on the lines around.
> 
> Cc: Nina Schoetterl-Glausch <nsg@linux.ibm.com>
> Fixes: c8df4a7aef ("migration: Split save_live_pending() into 
> state_pending_*")
> Signed-off-by: Peter Xu <peterx@redhat.com>

queued for 9.0-rc1.

-- 
Peter Xu