Train vmdk driver doesn't copy data into vmdk from glance correctly

[Ryan Fuerst]

Upgraded from OpenStack Stein (openstack-ansible deployment methods) to Train. The VMware vmdk driver when creating a volume from an Image doesn't copy any data over into the vmdk on the ESX side. Everything says things worked correctly, the volume gets created, openstack logs say its good but when you log into ESX the volume doesn't actually have any data in it and is 0 size. downgrade to stein it works correctly, train and above this doesn't work. And all cinder log files say it copied everything over successfully.

If you try to launch and Instance from this volume you get OS not found (because their is no data in the volume)

[Ryan Fuerst (rfuerst42)]

esx log files

[Ryan Fuerst (rfuerst42)]

I see this marked Fixed Release, what is the fix?

[Ryan Fuerst (rfuerst42)]

This works in Stein, fails in Train and above

[Ryan Fuerst (rfuerst42)]

Line 1263 in the train version

                http_method='POST') <--- that seems to be the code change between stein & train causing this problem.

[Ryan Fuerst (rfuerst42)]

Commit

https://github.com/openstack/cinder/commit/748ab113d68260fecd0f978b8de0e6b5f62e38f5

Seems to be when this defect was introduced against normal Datastores (I do not have a VVOL to test against)

[Xiangfei Zhu (xiangfeiz)]

@Ryan, if you remove http_method='POST' or change POST to PUT, is creating a volume from image successful? I'm asking because you may have the env handy for a test.

[Ryan Fuerst (rfuerst42)]

If I remove that line then things work, haven't tried changing it to PUT. Currently I have a normal Datastore environment not one with a VVOL setup. Not sure when I'll be able to test if that change breaks when using VVOLS or not.

[Ryan Fuerst (rfuerst42)]

Tested a VVOL setup, same results, no data in the volume created in the datastore.