[Pulp-dev] Master-detail inheritance in Pulp 3

Wed Jul 24 14:08:42 UTC 2019

I like your solution using default_related_name [0] manually, as Brian
noted [1], it's more explicit and therefore more pythonic.

That in mind, Daniel's alternative, not using model inheritance for the
Content models [2], while less simple a change initially, potentially had
significant performance gains and is also more explicit and pythonic.

Should we still pursue this more complex fix for the improvements to
bulk_create since we'd rather have breaking changes early in development
than need to address them later?

Or am I putting the cart before the horse by seeking optimization too early?

[0] https://pulp.plan.io/issues/4681#note-19
[1] https://pulp.plan.io/issues/4681#note-20
[2] https://pulp.plan.io/issues/4681#note-11

Dana Walker

She / Her / Hers

Software Engineer, Pulp Project

Red Hat <https://www.redhat.com>

dawalker at redhat.com
<https://www.redhat.com>

On Wed, Jul 24, 2019 at 8:24 AM David Davis <daviddavis at redhat.com> wrote:

> I want to bump this thread again. We've only had one person weigh in and
> this is a major change that'll affect all Pulp 3 plugins that we need to
> address soon. Please respond here or on the issue with feedback.
>
> David
>
>
> On Sun, Jul 21, 2019 at 10:49 AM Brian Bouterse <bbouters at redhat.com>
> wrote:
>
>> Thanks for the investigation and demo patch. I posted a +1 for the manual
>> option with reasoning here: https://pulp.plan.io/issues/4681#note-20
>>
>> Other ideas and perspectives are welcome. I hope we can resolve this
>> issue soon as we approach RC4.
>>
>> On Wed, Jul 17, 2019 at 12:55 PM David Davis <daviddavis at redhat.com>
>> wrote:
>>
>>> I did some investigation and posted my findings[0]. Basically, it would
>>> be possible to solve this problem by defining default_related_name either
>>> manually or automatically on detail models. I don't know if we want to go
>>> this route so feedback is appreciated.
>>>
>>> [0] https://pulp.plan.io/issues/4681#note-19
>>>
>>> David
>>>
>>>
>>> On Mon, Apr 29, 2019 at 2:16 PM David Davis <daviddavis at redhat.com>
>>> wrote:
>>>
>>>> It seems like most people are in favor of setting the OneToOneField or
>>>> perhaps the default_related_name on the detail model. I think there’s also
>>>> some interest in seeing how we can do this automatically for plugins. I’ve
>>>> added this feedback to the issue:
>>>>
>>>> https://pulp.plan.io/issues/4681#note-8
>>>>
>>>> David
>>>>
>>>>
>>>> On Wed, Apr 24, 2019 at 6:22 AM Ina Panova <ipanova at redhat.com> wrote:
>>>>
>>>>> I would avoid making changes in class naming. So +1 for the
>>>>> OneToOneField definition.
>>>>>
>>>>>
>>>>> --------
>>>>> Regards,
>>>>>
>>>>> Ina Panova
>>>>> Software Engineer| Pulp| Red Hat Inc.
>>>>>
>>>>> "Do not go where the path may lead,
>>>>>  go instead where there is no path and leave a trail."
>>>>>
>>>>>
>>>>> On Tue, Apr 23, 2019 at 6:45 PM David Davis <daviddavis at redhat.com>
>>>>> wrote:
>>>>>
>>>>>> The default_related_name setting is something that django provides.
>>>>>> Subclasses can also explicitly define their OneToOneField parent link as
>>>>>> well:
>>>>>>
>>>>>> content_ptr = models.OneToOneField(Content, on_delete=models.CASCADE,
>>>>>> parent_link=True, related_name='rpm_package')
>>>>>>
>>>>>> I am not sure what you mean by 'robust' but if a plugin subclass
>>>>>> doesn't do either of these, it may not work with other plugins.
>>>>>>
>>>>>> I think the question now would be whether we should just document
>>>>>> this or try to do it automagically for plugins?
>>>>>>
>>>>>> David
>>>>>>
>>>>>>
>>>>>> On Tue, Apr 23, 2019 at 12:31 PM Brian Bouterse <bbouters at redhat.com>
>>>>>> wrote:
>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> On Tue, Apr 23, 2019 at 11:02 AM David Davis <daviddavis at redhat.com>
>>>>>>> wrote:
>>>>>>>
>>>>>>>> I think I found another solution that might work best: defining
>>>>>>>> 'default_related_name' on subclassed master-detail models. So Package in
>>>>>>>> pulp_rpm would define its default_related_name as "rpm_package".
>>>>>>>>
>>>>>>> Would we be making 'default_related_name' or is that something
>>>>>>> Django is providing? If it's something Pulp would be providing perhaps
>>>>>>> defining the explicit one-to-one field is better. Any plugin that takes the
>>>>>>> step of defining the one-to-one field will insulate themselves from other
>>>>>>> plugins. If plugins don't take that step they will still work, just not as
>>>>>>> robustly. Am I thinking about this correctly?
>>>>>>>
>>>>>>>
>>>>>>>> David
>>>>>>>>
>>>>>>>>
>>>>>>>> On Tue, Apr 23, 2019 at 10:29 AM David Davis <daviddavis at redhat.com>
>>>>>>>> wrote:
>>>>>>>>
>>>>>>>>> I wanted to email the pulp-dev list about a major problem[0] that
>>>>>>>>> was recently encountered in Pulp 3 that affects how the Pulp 3 plugin API
>>>>>>>>> functions.
>>>>>>>>>
>>>>>>>>> # Problem
>>>>>>>>>
>>>>>>>>> In the plugin API we rely on inheritance to allow plugin writers
>>>>>>>>> to import functionality into their plugin. This includes models such as
>>>>>>>>> Remote and Content that are inherited by plugins. We rely on django's
>>>>>>>>> multi-table inheritance[1] for these models.
>>>>>>>>>
>>>>>>>>> Behind the scenes, django defines a OneToOneField and a reverse
>>>>>>>>> accessor. This field is not namespace so if two subclasses have the same
>>>>>>>>> name, you get an error ("Reverse accessor for 'Package.content_ptr' clashes
>>>>>>>>> with reverse accessor for 'Package.content_ptr'.")
>>>>>>>>>
>>>>>>>>> To give an actual example, both the Debian and RPM plugins
>>>>>>>>> implement a Package class. This causes an error to be raised when a user
>>>>>>>>> installs both plugins. Django tries to define a 'package' reverse accessor
>>>>>>>>> for both subclasses and blows up.
>>>>>>>>>
>>>>>>>>> # Potential Solutions
>>>>>>>>>
>>>>>>>>> ## Class Naming
>>>>>>>>>
>>>>>>>>> The first solution I can think of which is probably also the
>>>>>>>>> simplest and most straightforward would be to require plugin writers to
>>>>>>>>> namespace their master/detail subclass names. So Package would be
>>>>>>>>> RpmPackage. This places the onus on plugin writers to name their
>>>>>>>>> master/detail classes correctly.
>>>>>>>>>
>>>>>>>>> ## Defining OneToOneField
>>>>>>>>>
>>>>>>>>> The other solution would be to either manually define the
>>>>>>>>> OneToOneField on the subclasses[2] and specify a namespaced field name.
>>>>>>>>> There may be a way to do this dynamically (ie magically) in the parent
>>>>>>>>> somehow as well.
>>>>>>>>>
>>>>>>>>> ## Abstract Class
>>>>>>>>>
>>>>>>>>> Lastly, we could redefine master models as abstract classes[3]. I
>>>>>>>>> can think of at least one or two places (e.g. content field on
>>>>>>>>> RepositoryVersionContent, publisher field on Publication) that would have
>>>>>>>>> to switch their relationships to generic relationships in order to
>>>>>>>>> accommodate this change.
>>>>>>>>>
>>>>>>>>> There might be other solutions I am not thinking of so feel free
>>>>>>>>> to propose something. Also, quick feedback would be greatly appreciated as
>>>>>>>>> this is going to be a major change in our plugin API.
>>>>>>>>>
>>>>>>>>> [0] https://pulp.plan.io/issues/4681
>>>>>>>>> [1]
>>>>>>>>> https://docs.djangoproject.com/en/2.2/topics/db/models/#multi-table-inheritance
>>>>>>>>> [2]
>>>>>>>>> https://docs.djangoproject.com/en/2.2/topics/db/models/#specifying-the-parent-link-field
>>>>>>>>> [3]
>>>>>>>>> https://docs.djangoproject.com/en/2.2/topics/db/models/#abstract-base-classes
>>>>>>>>>
>>>>>>>>> David
>>>>>>>>>
>>>>>>>> _______________________________________________
>>>>>>>> Pulp-dev mailing list
>>>>>>>> Pulp-dev at redhat.com
>>>>>>>> https://www.redhat.com/mailman/listinfo/pulp-dev
>>>>>>>>
>>>>>>> _______________________________________________
>>>>>> Pulp-dev mailing list
>>>>>> Pulp-dev at redhat.com
>>>>>> https://www.redhat.com/mailman/listinfo/pulp-dev
>>>>>>
>>>>> _______________________________________________
> Pulp-dev mailing list
> Pulp-dev at redhat.com
> https://www.redhat.com/mailman/listinfo/pulp-dev
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/pulp-dev/attachments/20190724/192b94e1/attachment.htm>