Re: [External] Re: [PATCH v3 2/7] migration/multifd: Implement zero page transmission on the multifd thread.

[Markus Armbruster]

Hao Xiang <hao.xiang@bytedance.com> writes:

> On Wed, Feb 28, 2024 at 1:50 AM Markus Armbruster <armbru@redhat.com> wrote:
>>
>> Hao Xiang <hao.xiang@bytedance.com> writes:
>>
>> > 1. Add zero_pages field in MultiFDPacket_t.
>> > 2. Implements the zero page detection and handling on the multifd
>> > threads for non-compression, zlib and zstd compression backends.
>> > 3. Added a new value 'multifd' in ZeroPageDetection enumeration.
>> > 4. Handle migration QEMU9.0 -> QEMU8.2 compatibility.
>> > 5. Adds zero page counters and updates multifd send/receive tracing
>> > format to track the newly added counters.
>> >
>> > Signed-off-by: Hao Xiang <hao.xiang@bytedance.com>
>>
>> [...]
>>
>> > diff --git a/qapi/migration.json b/qapi/migration.json
>> > index 1e66272f8f..5a1bb8ad62 100644
>> > --- a/qapi/migration.json
>> > +++ b/qapi/migration.json
>> > @@ -660,10 +660,13 @@
>> >  #
>> >  # @legacy: Perform zero page checking from main migration thread.
>> >  #
>> > +# @multifd: Perform zero page checking from multifd sender thread.
>> > +#
>> >  # Since: 9.0
>> > +#
>> >  ##
>> >  { 'enum': 'ZeroPageDetection',
>> > -  'data': [ 'none', 'legacy' ] }
>> > +  'data': [ 'none', 'legacy', 'multifd' ] }
>> >
>> >  ##
>> >  # @BitmapMigrationBitmapAliasTransform:
>>
>> What happens when you set "zero-page-detection": "multifd" *without*
>> enabling multifd migration?
>
> Very good question! Right now the behavior is that if "multifd
> migration" is not enabled, it goes through the legacy code path and
> the "multifd zero page" option is ignored. The legacy path has its own
> zero page checking and will run the same way as before. This is for
> backward compatibility.

We need one of two improvements then:

1. Make "zero-page-detection" reject value "multifd" when multifd
   migration is not enabled.  Document this: "Requires migration
   capability @multifd" or similar.

2. Document that "multifd" means multifd only when multifd is enabled,
   else same as "legacy".

I prefer 1., because it's easier to document.  But migration maintainers
may have their own preference.  Peter, Fabiano?