On Tue, Sep 16, 2014 at 1:00 AM, Fawad Khaliq <fa...@plumgrid.com> wrote:

> Folks,
> I have had discussions with some folks individually about this but I would
> like bring this to a broader audience.
> I have been playing with security groups and I see the notion of 'default'
> security group seems to create some nuisance/issues.
> There are list of things I have noticed so far:
>    - Tenant for OpenStack services (normally named service/services) also
>    ends up having default security group.
>    - Port create operation ends up ensuring default security groups for
>    all the tenants as this completely seems out of the context of the tenant
>    the port operation takes place. (bug?)
>    - Race conditions where if system is stressed and Neutron tries to
>    ensure the first default security group and in parallel another call comes,
>    Neutron ends up trying to create multiple default security groups as the
>    checks for duplicate groups are invalidated as soon as the call make past a
>    certain point in code.
> Right this is a bug. We should catch this foreign key constraint exception
here and pass as the default security group was already created. We do
something similar here when ports are created as there is a similar race
for ip_allocation.

>    -
>    - API performance where orchestration chooses to spawn 1000 tenants
>    and we see unnecessary overhead
>    - For plugins that use RESTful proxy backends require the backend
>    systems to be up at the time neutron starts. [Minor, but affects some
>    packaging solutions]
This is probably always a requirement for neutron to work so I don't think
it's related.

> To summarize, is there a way to disable default security groups? Expected
> answer is no; can we introduce a way to disable it? If that does not make
> sense, should we go ahead and fix the issues around it?

I think we should fix these bugs you pointed out.

> I am sure some of you must have seen some of these issues and solved them
> already. Please do share how do tackle these issues?
> Thanks,
> Fawad Khaliq
> _______________________________________________
> OpenStack-dev mailing list
> OpenStack-dev@lists.openstack.org
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
OpenStack-dev mailing list

Reply via email to