Open Stack

2016-03-18 12:06 GMT+09:00 reedip banerjee <reedip14 at gmail.com>:
> Dear All Neutron Developers and Reviewers,
>
> I have a query/concern related to the parsing of options in
> python-neutronclient.
> I would like to bring this up, as it "may" also impact the transition of the
> CLIs to the openstack client as well.
>
> NeutronClient is pretty special in its behavior, and has one pretty powerful
> feature of parsing extra options. This feature states that, if the CLI does
> not support an option but the API does, and the user passes a value for this
> option, then the "unsupported" CLI option is parsed , and forwarded to the
> Neutron Server for processing.
>
> Example:
> Currently "neutron net-create" does not support --router:external. If you
> see the output of "neutron net-create -h" you would not find
> "--router-external". However, this option is supported in the API since Juno
> [2]. So therefore , if a user executes the following CLI
> " neutron net-create TestNetwork --router-external"
>
> then [1] would be observed as an output.

We can use "--router:external" since the initial support of neutron
router (Folsom).
You are talking about the situation that it is not described in the
help message.
It is better to have it in the help message.

> Now the query/concern comes next....
> Any option which is not supported by the CLI is open to the above parsing.
> Therefore , for net-create and net-update, all the following are possible:
>
> neutron net-create --router:external=True TESTNetwork --(A)
> neutron net-create --router:external TESTNetwork  --(B)
> neutron net-create TESTNetwork --router:external --(C)
> neutron net-create TESTNetwork --router:external=True --(D)
> neutron net-create TESTNetwork --router:external True --(E)
>
> However, user is not aware of the --router:external option because it is not
> visible in the HELP section ( this is true for other CLI options as well).
>
> In order to demonstrate these options to the User, we have to update
> add_known_arguments function to display them. And once they are known to the
> CLI, the parsing changes, and some of the options from (A) to (E) may not be
> supported ( Please see [3] for an ongoing, though now dormant, discussion ).
>
> Note that this discussion is not limited only to net-create, but possibly
> other CLIs as well which do not completely expose the Options which the API
> can support.I am , however, taking the net-create example as a case-study.
>
> I would like to know how we can move forward in this regards:
>
> -- Should NeutronClient continue to support all options from (A) to (E), but
> deprecate some of them in Openstack Client?

As Richard said, we don't plan to support the extra arguments in
openstackclient,
so this is not a problem in the openstackclient migration.
In OpenStack Client, we should follow OpenStack client option convention.
I don't think it is a DEPRECATION of some options. It is a DERECATION
of neutronclient CLI itself.

> -- Should we deprecate them in NeutronClient itself, so that the users are
> comfortable with the options when the migration to Openstack Client occurs?

My vote is to keep the current behavior.
On the other hand I don't think we need to show all available patterns.

Akihiro


>
> -- Any other suggestions....
>
> [1]: http://paste.openstack.org/show/491032/
>
> [2]:
> http://docs.openstack.org/juno/install-guide/install/apt/content/neutron_initial-external-network.html
>
> [3]: https://review.openstack.org/#/c/137279/
>
> --
> Thanks and Regards,
> Reedip Banerjee
> IRC: reedip
>
>
>
>
>
> __________________________________________________________________________
> OpenStack Development Mailing List (not for usage questions)
> Unsubscribe: OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>