[openstack-dev] [nova]nova not work with eventlet 0.16.0

Sean Dague sean at dague.net
Wed Jan 14 13:01:41 UTC 2015

On 01/14/2015 07:58 AM, Ihar Hrachyshka wrote:
> On 01/14/2015 01:44 PM, Daniel P. Berrange wrote:
>> On Wed, Jan 14, 2015 at 07:39:59AM -0500, Sean Dague wrote:
>>> On 01/14/2015 04:08 AM, Dr. Jens Rosenboom wrote:
>>>> Am 14/01/15 um 05:17 schrieb Adam Gandelman:
>>>>> So eventlet 0.16.x has started hitting slaves and breaking stable
>>>>> branches
>>>>> (its not like we weren't warned :\ )
>>>>> https://bugs.launchpad.net/nova/+bug/1410626
>>>>> Should hopefully be resolved by eventlet versions caps in icehouse +
>>>>> juno's
>>>>> requirements:
>>>>> https://review.openstack.org/#/q/I4bbbeb5bf9c22ed36f5c9a74fec6b487d2c15697,n,z
>>>> It was discussed already in the context of
>>>> https://bugs.launchpad.net/nova/+bug/1407685 that capping eventlet
>>>> would
>>>> not be the proper solution, instead there is a fix already in master
>>>> which has backports to icehouse + juno pending:
>>>> https://review.openstack.org/#/c/145955/
>>>> https://review.openstack.org/#/c/146096/
>>>> Or is this new bug triggered by a different piece of code relating to
>>>> the newly released eventlet 0.16.1?
>>> Capping in stable/* is fine. A real openstack installation will not
>>> randomly rev all the dependent libraries to a latest version. Assuming
>>> that they will do so is misguided at best.
>>> There was real concern in getting master to work correctly, which was
>>> done. But for stable, capping is the right answer.
>> The patch that went into master is pretty trivial really. I don't see a
>> reason to not just copy that into stable too.
> Gate issues are complicated due to the fact that currently we have
> multiple failures for nova:
> 1) eventlet patch missing/eventlet not capped;
> 2) boto patch missing/boto not capped.
> Currently if you want to pass gate, you have two options:
> a) squash boto and eventlet patches and merge them in;
> b) merge in an update to requirements.txt that will cap both eventlet
> and boto.
> The latter seems to be more straightforward. If you are interested in
> making sure stable branches work ok with new boto/eventlet releases, you
> will have time to resolve it while gate will be in good shape and not
> broken (including master). After both boto and eventlet patches are
> merged in all stable branches, we may reconsider uncapping them.

And also matches the fact that we've agreed that stable requirements are
going to be fully capped, the implementation is just lagging on that.


Sean Dague

More information about the OpenStack-dev mailing list