Re: [Qemu-devel] [PATCH v2] migration: Convert 'status' of MigrationInfo to use an enum type

[zhanghailiang]

On 2015/2/28 1:42, Eric Blake wrote:
> On 02/27/2015 10:07 AM, Dr. David Alan Gilbert wrote:
> > > Rather than pollute the user-exposed enum with a state that we will
> > > never report, can we come up with some internal-only method for tracking
> > > cancelling separate from the enum?
> >
> > Well I guess we could just report it; but would that break any external tools?
>
> It might - I seem to recall in the past that when we added a new state
> string, that at least libvirt choked when encountering the unknown
> string (but I don't recall if it was migration or something else).  At
> least libvirt already has an enum tracking:
>
> enum {
>      QEMU_MONITOR_MIGRATION_STATUS_INACTIVE,
>      QEMU_MONITOR_MIGRATION_STATUS_ACTIVE,
>      QEMU_MONITOR_MIGRATION_STATUS_COMPLETED,
>      QEMU_MONITOR_MIGRATION_STATUS_ERROR,
>      QEMU_MONITOR_MIGRATION_STATUS_CANCELLED,
>      QEMU_MONITOR_MIGRATION_STATUS_SETUP,
>
>      QEMU_MONITOR_MIGRATION_STATUS_LAST
> };
>
> and a string mapping of just the following states:
>
> VIR_ENUM_IMPL(qemuMonitorMigrationStatus,
>                QEMU_MONITOR_MIGRATION_STATUS_LAST,
>                "inactive", "active", "completed", "failed", "cancelled",
> "setup")
>
> Furthermore, the function qemuMigrationUpdateJobStatus() is doing
> various things depending on the observed state, where the current trick
> of treating 'cancelling' like 'active' would mean that changing
> 'cancelling' to be an independent state WOULD have observable behavior
> change in libvirt.  But I don't know if the change would break things,

Er, i have tested with returning 'cancelling' to users, and
only when we try to cancel a migration, libvirt sometimes will report :
Migration: [ 69 %]^Cerror: internal error: unexpected migration status in 
cancelling.
But the cancelling process is still completed.

And, yes, it is very rare, depend on the time window. (In my test, i add a 
sleep of 5s
to extend the time between cancelling and cancelled.)

> or if it would still end up resolving nicely (after all, cancelling only
> occurs for a short window before the migration aborts anyway, so it
> might just sort itself out when it finally gets to cancelled).

> On the other hand, we can argue that clients that are unprepared to
> handle new enum states gracefully are broken, and we also have the
> argument that it is okay for a new qemu to require a new libvirt release
> (the other direction is not okay - a new libvirt must not require

Agreed, upgrade libvirt is reasonable.
So, should i send v3 with exposing 'cancelling'to user, and CC libvirt ?

Thanks.

> upgrading to a new qemu).  So exposing 'cancelling' may make this patch
> easier.