Thanks Sachin for your answer and reviews.
My test is very simple for the moment:
for i in {1..100}; do echo "#################### $i ####################";
neutron router-gateway-clear R1; neutron router-gateway-set R1 public; done
I run that test on a devstack environment. So it's a a mono-node setup and
I plan to run it on a multi config nodes platform next.
For the moment, I still have sometime (very rare) the error we discuss
above.
And another error I wasn't able to understand on the VNC API code.
The methods '_cassandra_[RESOURCE]_read' generated method code fails with
that error [1]. It seems the resource is not complete in the 'obj_uuid'
Column family (no fq_name key). How that's possible? The resource is
created in several steps into the db? Any thoughts? I met that error
principally with 'instance-ip' resource and sometimes with
'virtual-machine-interface' with the test I described.
[1] http://paste.openstack.org/show/203945/
Édouard.
On Wed, Apr 15, 2015 at 3:18 AM, Sachin Bansal <[email protected]> wrote:
> Hi Edouard,
>
> NoIdError on getting back_refs is actually a bug. We can fix it by
> either tracking which fields were read during the last read of the object
> or more simply, providing a different API that does not re-read the object.
> I think we should fix it for 2.20.
>
> For others, I looked at your reviews [1] and [2]. I have sent some
> comments on [1] and approved [2]. Please also port them to all other
> branches (2.0, 2.1, 2.2 and master). For any other crashes you see, if we
> can explain why that exception is happening, we should except it.
> Otherwise, it may be better to identify the root cause of the exception.
>
> Also, if you are using a test script for this stress test, please send
> it to me. I will also try it out myself.
>
> Thanks for running these tests!
>
> Sachin
>
>
>
> On Apr 14, 2015, at 9:53 AM, Édouard Thuleau <[email protected]> wrote:
>
> Hi,
>
> I'm facing lot of issues with schema and SVC monitor code when I stress
> them.
> For example, I've a simple test where I set and clear the Neutron gateway
> on a router in a loop.
> That means for Contrail, I create then delete a SNAT service
> instance continually.
>
> I've notice lot of bugs where the code does not handle correctly
> exceptions like 'NoIdError' or 'RefsExistError'. I'm starting to fix them
> for the specific above test and I'm close to stabilize that test case (I
> only worked on R1.10 for the moment, and put all in big reviews [1] and
> [2]).
>
> But sometimes, I get an exception 'NoIdError' (like that one [3]) in
> schema or SVC code when they use vnc_api client code to list a back refs of
> a resource. Indeed, the vnc_api client calls the API to get resource back
> refs (ie. [4]).
> Should I re-work schema and SVC code to take care of the case every time
> they tried to get back references on a resources?
>
> [1] https://review.opencontrail.org/#/c/8977
> [2] https://review.opencontrail.org/#/c/8978/
> [3] http://paste.openstack.org/show/203906/
> [4]
> https://github.com/Juniper/contrail-generateDS/blob/master/ifmap_frontend.py#L807
>
> Édouard.
>
>
>
_______________________________________________
Dev mailing list
[email protected]
http://lists.opencontrail.org/mailman/listinfo/dev_lists.opencontrail.org
