AttributeError raised during roll_back_live_migration_at_destination

[Zhenyu Zheng]

In roll_back_live_migration_at_destination, we treat migrate_data as a LiveMigrateData object, but it is actually a dict:
https://github.com/openstack/nova/blob/master/nova/compute/manager.py#L5176-L5198
and
https://github.com/openstack/nova/blob/master/nova/compute/manager.py#L5509-L5514

We should not transform object to dict in the above code, instead, we should transform
dict to object.

We have logic here :
https://github.com/openstack/nova/blob/master/nova/compute/manager.py#L5484-L5491
but we directly pass it down to driver here:
https://github.com/openstack/nova/blob/master/nova/compute/manager.py#L5512-L5514
and it has not been handled in libvirt driver:
https://github.com/openstack/nova/blob/master/nova/virt/libvirt/driver.py#L6406-L6412

This will cause AttributeError.

[Matt Riedemann (mriedem)]

What git hash are you testing against? The code links no longer line up.

[Matt Riedemann (mriedem)]

Where is the compute manager code converting an object to a dict?

If _do_live_migration does not get an object for migrate_data, it converts it to one here:

https://github.com/openstack/nova/blob/2d7a851199b12355b36a3982ea0fc7bb5a7cd11f/nova/compute/manager.py#L5174

And then migrate_data (as an object) should be passed down to _rollback_live_migration here:

https://github.com/openstack/nova/blob/2d7a851199b12355b36a3982ea0fc7bb5a7cd11f/nova/compute/manager.py#L5195

And even if _rollback_live_migration gets a dict for migrate_data, it should convert it to an object here:

https://github.com/openstack/nova/blob/2d7a851199b12355b36a3982ea0fc7bb5a7cd11f/nova/compute/manager.py#L5486

Are you saying it's the call to compute rpcapi that's converting it to a dict?

https://github.com/openstack/nova/blob/2d7a851199b12355b36a3982ea0fc7bb5a7cd11f/nova/compute/manager.py#L5511

[Matt Riedemann (mriedem)]

Also, are you live migrating from 2 mitaka nodes, or are you live migrating from liberty->mitaka (or vice-versa)?

[Zhenyu Zheng (zhengzhenyu)]

@Matt Riedemann, Hi, sorry for the delay, I was on vacation, I realized I was in the middle of your BP implementation when I find out this bug, I will close this one.