Re: [Qemu-devel] [PATCH 17/23] migration: make sure we always have a migration state

[Juan Quintela]

Anthony Liguori <address@hidden> wrote:
> On 10/04/2011 09:49 AM, Juan Quintela wrote:
>> Anthony Liguori<address@hidden>  wrote:
>>> On 09/23/2011 07:57 AM, Juan Quintela wrote:

[ much more stuff ]

>> It avoids s==NULL checks,
>
> In favor of s->state == MIG_STATE_NONE.
>
>> and it also avoids having to have new
>> variables (max_throotle) just because we don't have a migration state
>> handy.
>
> The intention of the global variable is to set a default for new
> sessions.  I can imagine a world where if you had parallel migrations,
> you could either toggle the default (the global variable) or the
> specific parameter within a single migration session.
>
> But it's not about features.  It's a general reluctances to heavily
> modify the code to rely on a global instead of passing the state
> around.  Here's a pretty good short write up of why this is a Bad
> Thing.
>
> http://stackoverflow.com/questions/1392315/problems-with-singleton-pattern
>
> Ultimately, MIG_STATE_NONE shouldn't be needed because we should not
> be relying on a singleton accessor to get the MigrationState.  A
> better cleanup would be to further pass MigrationState between
> functions and eliminate the need for a global at all.

I understand the singleton problem, but the reason to put STATE_NONE is
not for that O:-) (it just happens that we only have a migration now).

Why I want it?

Just now, the only thing that we are "setting" for a migration before it
starts is the "bandwidth".  I see the future when migration becomes
something like:

migration_set_speed ....
migration_set_target ....
migration_set_<whatever else>
migrate

as you can see, we are "preparing" migration, and we don't have a
"STATE" now that describes this state, migration not started, but we
want to prepare it.

Perhaps a better name that STATE_NONE is in order.  I got NONE in the
sense that "migration" has not been attemted yet.

Later, Juan.