On 08/05/2011 02:12 AM, Alex Jia wrote:
On 08/04/2011 10:27 PM, Eric Blake wrote:
> On 08/03/2011 11:30 PM, Alex Jia wrote:
>> Value stored to 'ret' is never read. If Confirm3 returns -1, there's
>> nothing
>> more we can do, here should remove this dead assignment.
>>
>> * src/qemu/qemu_migration.c: kill dead assignment.
>>
>> Signed-off-by: Alex Jia<ajia(a)redhat.com>
>> ---
>> src/qemu/qemu_migration.c | 6 +++---
>> 1 files changed, 3 insertions(+), 3 deletions(-)
>
> I'm thinking NACK as written. We should probably emit a log message if
> confirm failed, at which point, ret would be used in the conditional
> deciding whether to do a log message. Furthermore, libvirt.c
> virDomainMigrateVersion3 has the same bug with a dead assignment to
> ret; we should solve it in the same way for both implementations.
>
Eric, I haven't figured out a apposite error number in virterror.h,
whether we need to define a new error number like this:
VIR_ERR_MIGRATE_CONFIRM_FAILED = 75, /* a migration finished, but
confirmation failed, there is a domain probably left in 'paused' state
on source.*/
That's if we want to report the error back to the caller and fail the
API. But the migration happened - it is running on the destination, so
I think we have to make the API return success. Which means that we
don't need a new VIR_ERR_* value, rather, we just need a new VIR_WARN()
call to guarantee a log entry but nothing further.
--
Eric Blake eblake(a)redhat.com +1-801-349-2682
Libvirt virtualization library
http://libvirt.org