Re: [Qemu-devel] [PATCH 2/4] migration/savevm: use migration_is_blocked to validate

[Wei Yang]

On Wed, May 15, 2019 at 10:38:37AM +0100, Dr. David Alan Gilbert wrote:
>* Wei Yang (address@hidden) wrote:
>> On Wed, May 15, 2019 at 02:38:27PM +0800, Wei Yang wrote:
>> >On Tue, May 14, 2019 at 04:18:14PM +0100, Dr. David Alan Gilbert wrote:
>> >>* Wei Yang (address@hidden) wrote:
>> >>> 
>> >>> Well, when you look into the source side of migration:
>> >>> 
>> >>> qmp_migrate
>> >>>   migrate_prepare
>> >>>     migration_is_blocked
>> >>> 
>> >>> This means if migration_is_blocked fails, the source will not start 
>> >>> migration.
>> >>> And it is the same as save_snapshot.
>> >>> 
>> >>> From my understanding, when we load a vm, it should check the same
>> >>> requirement.
>> >>
>> >>I've been thinking about this, and I think I agree with Daniel on this.
>> >>The 'migration_blockers' list tells you that something about the
>> >>*current* state of a device means that it can't be migrated - e.g.
>> >>a 9pfs with a mounted filesystem can't be migrated.
>> >>
>> >>If we're about to reload the state from a snapshot, then the saved
>> >>snapshot's state must have been migratable, so that's OK.
>> >>
>> >
>> >The situation is on a vm with 'migration_blockers' still could reload from a
>> >snapshot.
>> >
>> >This sounds reasonable. Thanks :-)
>> >
>> 
>> Well, this is still a little strange. The means source vm and destination vm
>> could have different configuration. Is this common?
>
>It's not different configuration that I'm worried about here; but it's 
>different runtime state.
>Items can get added/removed from migration_blockers dynamically
>depending on the behaviour of the guest; e.g. a device might only
>migratable in certain states.
>

I am not familiar with the usage of migration_blockers, just found one case
when we add a reason to it. -- vhost_dev_init().

Per my understanding, this is a device. We specify it in command line or use
hot-plug to add it. To me, guest may not alter the add/remove? Looks even we
have one such device, we still could load vm. This looks not bad, but we have
the different devices from source. 

BTW, migration works if source and destination have different devices?

As you mentioned, these is some case where guest could add/remove a reason to
migration_blockers.  This is what you concerned right?

Do we need to limit the usage of migration_blockers? Just use this in the case
you concerned?

>Dave
>
>
>> -- 
>> Wei Yang
>> Help you, Help me
>--
>Dr. David Alan Gilbert / address@hidden / Manchester, UK

-- 
Wei Yang
Help you, Help me