Open Stack

We settled on 1251920.

https://review.openstack.org/57509 is the fix for that bug.

Note that Oslo was fixed on Jun 28th, nova hasn't synced since then.
If we were using oslo as a library we would have had the fix as soon
as olso did a release.

These are the references to strong_store - and thus broken in nova
trunk (and if any references exist in H, in H too):
./nova/network/neutronv2/__init__.py:58:        if not
hasattr(local.strong_store, 'neutron_client'):
./nova/network/neutronv2/__init__.py:59:
local.strong_store.neutron_client = _get_client(token=None)
./nova/network/neutronv2/__init__.py:60:        return
local.strong_store.neutron_client
./nova/openstack/common/rpc/__init__.py:102:    if
((hasattr(local.strong_store, 'locks_held')
./nova/openstack/common/rpc/__init__.py:103:         and
local.strong_store.locks_held)):
./nova/openstack/common/rpc/__init__.py:108:                 {'locks':
local.strong_store.locks_held,
./nova/openstack/common/local.py:47:strong_store = threading.local()
./nova/openstack/common/lockutils.py:173:        if not
hasattr(local.strong_store, 'locks_held'):
./nova/openstack/common/lockutils.py:174:
local.strong_store.locks_held = []
./nova/openstack/common/lockutils.py:175:
local.strong_store.locks_held.append(name)
./nova/openstack/common/lockutils.py:217:
local.strong_store.locks_held.remove(name)
./nova/tests/network/test_neutronv2.py:1837:
local.strong_store.neutron_client = None


So we can expect lockutils to be broken, and rpc to be broken. Clearly
they are being impacted more subtly than the neutron client usage.

-Rob


On 21 November 2013 07:44, Robert Collins <robertc at robertcollins.net> wrote:
> Which of these bugs would be appropriate to use for the fix to
> strong_store - it affects lockutils and rpc, both of which are going
> to create havoc :)
>
> -Rob
>
> On 21 November 2013 07:19, Salvatore Orlando <sorlando at nicira.com> wrote:
>> I've noticed that
>> https://github.com/openstack/nova/commit/85332012dede96fa6729026c2a90594ea0502ac5
>> stores the network client in local.strong_store which is a reference to
>> corolocal.local (the class, not the instance).
>>
>> In Russell's example instead the code accesses local.store which is an
>> instance of WeakLocal (inheriting from corolocal.local).
>>
>> Perhaps then Roman's findings apply to the issue being observed on the gate.
>>
>> Regards,
>> Salvatore
>>
>>
>> On 20 November 2013 18:32, Russell Bryant <rbryant at redhat.com> wrote:
>>>
>>> On 11/20/2013 12:21 PM, Alex Gaynor wrote:
>>> > Nope, you're totally right, corolocal.local is a class, whose instances
>>> > are the actual coroutine local storage.
>>>
>>> But I don't think his example is what is being used.
>>>
>>> Here is an example using the openstack.common.local module, which is
>>> what nova uses for this.  This produces the expected output.
>>>
>>> http://paste.openstack.org/show/53687/
>>>
>>>
>>> https://git.openstack.org/cgit/openstack/nova/tree/nova/openstack/common/local.py
>>>
>>> For reference, original example from OP:
>>> http://paste.openstack.org/show/53686/
>>>
>>> --
>>> Russell Bryant
>>>
>>> _______________________________________________
>>> OpenStack-dev mailing list
>>> OpenStack-dev at lists.openstack.org
>>> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>>
>>
>>
>> _______________________________________________
>> OpenStack-dev mailing list
>> OpenStack-dev at lists.openstack.org
>> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>>
>
>
>
> --
> Robert Collins <rbtcollins at hp.com>
> Distinguished Technologist
> HP Converged Cloud



-- 
Robert Collins <rbtcollins at hp.com>
Distinguished Technologist
HP Converged Cloud