rave-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Chris Geer <ch...@cxtsoftware.com>
Subject Re: Rave Data Structures
Date Tue, 12 Nov 2013 04:43:53 GMT
Better late than never. I started to document my thoughts on the wiki.

https://wiki.apache.org/rave/RESTAPI

Chris


On Thu, Oct 24, 2013 at 11:09 AM, Chris Geer <chris@cxtsoftware.com> wrote:

> On Sun, Oct 6, 2013 at 1:56 PM, Matt Franklin <m.ben.franklin@gmail.com>wrote:
>
>> On Tue, Sep 24, 2013 at 1:02 PM, Chris Geer <chris@cxtsoftware.com>
>> wrote:
>>
>> > On Tue, Sep 24, 2013 at 9:54 AM, Michael Jett <mikesjett@gmail.com>
>> wrote:
>> >
>> > > On Mon, Sep 23, 2013 at 7:28 PM, Chris Geer <chris@cxtsoftware.com>
>> > wrote:
>> > >
>> > > > Some additional comments below:
>> > > >
>> > > > On Sun, Sep 22, 2013 at 7:01 PM, Erin Noe-Payne <
>> > > erin.noe.payne@gmail.com
>> > > > >wrote:
>> > > >
>> > > > > Hey all, sorry I've been absent from this thread. I've caught
up
>> now
>> > > > > and have a few thoughts:
>> > > > >
>> > > > > - As discussed before, the REST api should have CRUD endpoints
>> that
>> > > > > support all data retrieval and manipulation for any case that
we
>> wish
>> > > > > for RAVE to allow. These endpoints should not be optimized for
>> > > > > specific client application / library needs.
>> > > > > - In addition we have some number of endpoints (currently one)
>> that
>> > > > > optimize for specific client needs. In this case we are talking
>> about
>> > > > > the pages for render endpoint, which serves 3 purposes:
>> > > > >   1) Filter to the set of pages by context and identifier
>> > > > >   2) Aggregate nested data for one read operation that gives
all
>> data
>> > > > > necessary to render
>> > > > >   3) Attach special properties to regionWidgets needed to render
>> the
>> > > > > iframes (rpc tokens)
>> > > > > - The data structure we retrieve from the render endpoint should
>> not
>> > > > > be different from the models sent / consumed by the CRUD
>> endpoints.
>> > > > >
>> > > >
>> > > > So using your example above....the regionWidget CRUD model should
>> have
>> > > rpc
>> > > > token?
>> > > >
>> > > >
>> > > > > The render endpoint is read-only, so any data manipulation that
>> > > > > follows will be interacting with the CRUD endpoints. This is
not
>> > > > > specific to angular - any client technology will be consuming
the
>> > > > > render endpoint and feeding back that data in the case of updates.
>> > > > > Again, render endpoint is only filtering, aggregating nested
data,
>> > > > > attaching render properties to regionWidget. Otherwise I don't
>> see a
>> > > > > reason the data models should differ.
>> > > > >
>> > > > > -A regionwidget should probably just have a mutable title
>> property.
>> > At
>> > > > > time of creation the title can be copied from the widget it is
an
>> > > > > instance of, but afterwards it can be changed by the user. F.E
if
>> I
>> > > > > have 3 instances of a map widget or a weather widget or something
>> on
>> > > > > my page, I may want to title them differently anyway.
>> > > > >
>> > > >
>> > > > I support this +1 as long as it's mutable.
>> > > >
>> > > > >
>> > > > > - I don't really know the best way to deal with relationships.
My
>> > > > > instinct is that resources that have a one-to-X relationship
>> should
>> > > > > simply have the id of that relationship as a property. For
>> example,
>> > > > > each region has a pageId property. Each regionWidget has a pageId
>> and
>> > > > > regionId property. Then these relationships are atomically mutable
>> > via
>> > > > > a PUT request. Many-to-many relationships should be modeled via
>> their
>> > > > > own resource. So for example user-to-users friend should
>> correspond
>> > to
>> > > > > a friends resource that can be POSTed and DELETEd.
>> > > > >
>> > > >
>> > > > My opinion is our data model is bad on the server. A RegionWidget
>> > > shouldn't
>> > > > have a RegionID or PageID because they really shouldn't be mutable.
>> I
>> > > > realize this is a backend problem, not a REST problem exactly. I'm
>> just
>> > > > arguing we don't extend that to the REST interface so we can fix the
>> > > > backend problem later. I don't like a situation where you can have
a
>> > url
>> > > > and data that can contradict.
>> > > >
>> > >
>> > > +1 - I think that this is a great point Chris. What planning or
>> > > coordination would it take to fix this backend problem?
>> > >
>> >
>> > I don't think fixing the backend problem is our primary concern as we
>> can
>> > work around it. My recommendation would be to design the REST API in
>> such a
>> > way as it makes the most sense, then we can fix the backend.
>> >
>> > >
>> > >
>> > > >
>> > > > > - Angular doesn't really care about this stuff. We have started
>> > > > > implementing according to what exists currently. If the api
>> changes,
>> > > > > then it will create some refactoring on the client. That's fine
-
>> > > > > $resource is just json objects anyway, so it can handle it.
>> > However...
>> > > > > - We really just need consistency. My impression so far is that
we
>> > > > > don't really have any practical expertise on building a robust
>> REST
>> > > > > api around a complicated data set. Short of that expertise
>> showing up
>> > > > > on the list, we really just need to rally around one strategy
and
>> > > > > implement it consistently.
>> > > > >
>> > > >
>> > > > Well, that's not entirely true ;)  At the end of the day, this data
>> > model
>> > > > really isn't all that complex (the rendering part). Page contains
>> 0-n
>> > > > Regions which contain 0-n Region Widgets. There is no reason why a
>> > > > RegionWidget needs a reference to the page it's on directly, that
>> > should
>> > > be
>> > > > derived by which resource it's retrieved from. It's that way on the
>> > > backend
>> > > > because our backed just mirrors the SQL table and it's foreign
>> keys. If
>> > > we
>> > > > had started this project with a NoSQL database, we never would have
>> had
>> > > > those attributes on those objects (probably wouldn't have used JPA
>> > > either).
>> > > >
>> > > > Using Mongo as an example, we never would have done this for a
>> page...
>> > > >
>> > > > {
>> > > >   id: 123,
>> > > >   regions: [
>> > > >     {
>> > > >       id: 332,
>> > > >       page_id: 123,
>> > > >       region_widgets: [
>> > > >         {
>> > > >           id: 454,
>> > > >           region_id: 332
>> > > >         }
>> > > >       ]
>> > > >     }
>> > > >   ]
>> > > > }
>> > > >
>> > > > We would have stored page as a single document and not broken
>> Regions
>> > and
>> > > > RegionWidgets out into separate top level objects.
>> > > >
>> > > >
>> > > That makes sense. The REST interface would give us an opportunity to
>> > > consolidate and hide the relational keys, presenting objects in a
>> > "natural"
>> > > state. (i.e. pages, users) Rather than breaking it down into
>> proprietary
>> > > sub-object jargon. (i.e. RegionWidgets...)
>> > >
>> >
>> > Bingo!! Now we are on the same page.
>> >
>>
>> I think this all makes sense.  The question now is if anyone has captured
>> it in Jira or started implementing it.  I was looking at the Page rest
>> model object today in trunk and it is still the deep one we created to
>> support the full "rendered" object.
>>
>
> I don't believe anyone has done any work here yet. I will take a stab at
> something this week.
>
>>
>>
>> >
>> > >
>> > >
>> > > > Chris
>> > > >
>> > > >
>> > > > >
>> > > > > On Wed, Sep 18, 2013 at 1:29 PM, Chris Geer <
>> chris@cxtsoftware.com>
>> > > > wrote:
>> > > > > > Dan,
>> > > > > >
>> > > > > > My proposal is that on the server side, we do not add the
extra
>> > > > > attributes
>> > > > > > to the REST model. Instead we have the render endpoint that
will
>> > > return
>> > > > > the
>> > > > > > full object tree with the extra attributes. So on the server
>> side
>> > > there
>> > > > > > would be a different object like you said i.e.
>> > RenderedRegionWidget.
>> > > An
>> > > > > > alternative is to not create an objet and just build the
JSON
>> > > directly
>> > > > > for
>> > > > > > the render endpoint.
>> > > > > >
>> > > > > > I'm also ok with the server ignoring unknown attributes
on a
>> > > > save/update.
>> > > > > >
>> > > > > > Chris
>> > > > > >
>> > > > > >
>> > > > > > On Wed, Sep 18, 2013 at 3:56 AM, Dan Gornstein <
>> dan@gornstein.com>
>> > > > > wrote:
>> > > > > >
>> > > > > >> Chris,
>> > > > > >>
>> > > > > >> I think I was a bit confused before and might have been
>> inaccurate
>> > > in
>> > > > > what
>> > > > > >> I was trying to say. I have been thinking about it a
bit more
>> had
>> > a
>> > > > few
>> > > > > >> more questions for you.
>> > > > > >>
>> > > > > >> The endpoint which returns information for rendering
a page is
>> a
>> > > > nested
>> > > > > >> object and I agree should be read only, meaning you
cannot
>> post to
>> > > the
>> > > > > same
>> > > > > >> endpoint with the nested data model to save your object.
In the
>> > > > angular
>> > > > > >> branch right now, when we hit this endpoint we break
down the
>> > > response
>> > > > > into
>> > > > > >> their individual angular resource objects (page, regions,
>> > > > regionWidgets)
>> > > > > >> which point to the CRUD interfaces.
>> > > > > >>
>> > > > > >> The way the pages for render endpoint works right now
is to use
>> > > > > pageService
>> > > > > >> to get the canonical model of a page and transform it
into the
>> > REST
>> > > > > model
>> > > > > >> and send it through the DefaultRenderService to prepare
the
>> > > > > regionWidgets.
>> > > > > >> This means it ends up returning the REST model of
>> RegionWidgets in
>> > > the
>> > > > > >> pages for render endpoint. So if you wanted to add the
pageId
>> and
>> > > > title,
>> > > > > >> they would have to be properties on the RegionWidget
REST
>> model.
>> > Are
>> > > > you
>> > > > > >> proposing that we make this pages for render endpoint
be
>> compiled
>> > > with
>> > > > > >> different model objects other than the REST ones? Would
there
>> be
>> > > > > something
>> > > > > >> along the lines of RegionWidgetRender, which has the
extra
>> > > properties
>> > > > > but
>> > > > > >> does not have CRUD operations?
>> > > > > >>
>> > > > > >> If this was the case and we did not add pageId and title
to the
>> > REST
>> > > > > models
>> > > > > >> of RegionWidgets, what will happen if we hit the pages
for
>> render
>> > > > > endpoint
>> > > > > >> and in angular still broke them down into their individual
>> > resource
>> > > > > >> objects? Would you be allowed to save a RegionWidget
to
>> > > > > >>
>> > > /pages/<page-id>/regions/<region-id>/regionwidgets/<region-widget-id>
>> > > > > which
>> > > > > >> had title and pageId on it and the server would ignore
it, or
>> > would
>> > > > you
>> > > > > >> expect the server to throw an error?
>> > > > > >>
>> > > > > >> Thanks,
>> > > > > >> Dan
>> > > > > >>
>> > > > > >>
>> > > > > >>
>> > > > > >> On Mon, Sep 16, 2013 at 1:21 PM, Michael Jett <
>> > mikesjett@gmail.com>
>> > > > > wrote:
>> > > > > >>
>> > > > > >> > I'd like to pause this discussion for a brief moment,
and
>> advise
>> > > > that
>> > > > > we
>> > > > > >> > watch this APIGee video.
>> > > > > >> >
>> > > http://www.youtube.com/watch?feature=player_embedded&v=QpAhXa12xvU
>> > > > > >> >
>> > > > > >> > This is something that Erin pointed us to early
on, and
>> serves
>> > as
>> > > a
>> > > > > basis
>> > > > > >> > for a good RESTful architecture. There are too
many good
>> points
>> > > made
>> > > > > that
>> > > > > >> > will allow our API to scale and be intuitive for
other
>> > developers.
>> > > > > >> >
>> > > > > >> > It's very important that we are on the same page.
I believe a
>> > lot
>> > > of
>> > > > > >> > "format" questions will be answered by following
the APIGee
>> > > > > guidelines.
>> > > > > >> >
>> > > > > >> > - Mike
>> > > > > >> >
>> > > > > >> >
>> > > > > >> > On Sun, Sep 15, 2013 at 6:30 PM, Chris Geer <
>> > > chris@cxtsoftware.com>
>> > > > > >> wrote:
>> > > > > >> >
>> > > > > >> > > Since we've been having a lot of discussions
on data
>> > structures
>> > > > > lately
>> > > > > >> I
>> > > > > >> > > wanted to write down what my suggestions were.
These aren't
>> > 100%
>> > > > > >> complete
>> > > > > >> > > examples but show the relationships
>> > > > > >> > >
>> > > > > >> > > CRUD Interfaces
>> > > > > >> > >
>> > > > > >> > > Page (/pages/<page-id>)
>> > > > > >> > > {
>> > > > > >> > >   id: <id>,
>> > > > > >> > >   owner: <owner>,
>> > > > > >> > >   regionIds: [
>> > > > > >> > >     <id>, <id>, <id>   (region
order is based on order in
>> > list,
>> > > > not
>> > > > > >> field
>> > > > > >> > > on region object)
>> > > > > >> > >   ]
>> > > > > >> > > }
>> > > > > >> > >
>> > > > > >> > > Region (/pages/<page-id>/regions/<region-id>
>> > > > > >> > > {
>> > > > > >> > >   id: <id>,
>> > > > > >> > >   regionWidgetIds: [
>> > > > > >> > >     <id>, <id>, <id>   (widget
order is based on order in
>> > list,
>> > > > not
>> > > > > >> field
>> > > > > >> > > on region widget object)
>> > > > > >> > >   ]
>> > > > > >> > > }
>> > > > > >> > >
>> > > > > >> > > Region Widget
>> > > > > >> > >
>> > > > >
>> > (/pages/<page-id>/regions/<region-id>/regionwidgets/<region-widget-id>
>> > > > > >> > > {
>> > > > > >> > >   id: <id>,
>> > > > > >> > >   widgetId: <widgetId>,
>> > > > > >> > >   collapsed: <collapsed>,
>> > > > > >> > >   ...
>> > > > > >> > > }
>> > > > > >> > >
>> > > > > >> > > Region Widget
>> > > > > >> > > Properties
>> > > > > >> > >
>> > > > > >> >
>> > > > > >>
>> > > > >
>> > > >
>> > >
>> >
>> (/pages/<page-id>/regions/<region-id>/regionwidgets/<region-widget-id/properties/<propertyId>)
>> > > > > >> > > {
>> > > > > >> > >   ....
>> > > > > >> > > }
>> > > > > >> > >
>> > > > > >> > > Widget (/widgets/<widget-id>
>> > > > > >> > > {
>> > > > > >> > >   id: <id>
>> > > > > >> > >   title: <title>
>> > > > > >> > > }
>> > > > > >> > >
>> > > > > >> > >
>> > > > > >> > > Render Interface (custom mime-type) (Read
Only)
>> > > > > >> > >
>> > > > > >> > > /pages/<page-id>
>> > > > > >> > > {
>> > > > > >> > >   id: <id>,
>> > > > > >> > >   regions: [
>> > > > > >> > >     {
>> > > > > >> > >       id: <id>,
>> > > > > >> > >       regionWidgets: [
>> > > > > >> > >         {
>> > > > > >> > >           id: <id>,
>> > > > > >> > >           widgetId: <widgetId>,
>> > > > > >> > >           title: <title>,
>> > > > > >> > >           properties: [
>> > > > > >> > >             {
>> > > > > >> > >                key: <key>,
>> > > > > >> > >                value: <value>
>> > > > > >> > >              }
>> > > > > >> > >           ]
>> > > > > >> > >         }
>> > > > > >> > >       ]
>> > > > > >> > >     }
>> > > > > >> > >   ]
>> > > > > >> > > }
>> > > > > >> > >
>> > > > > >> > > You should also be able to render sub elements
below a
>> page so
>> > > for
>> > > > > >> > example,
>> > > > > >> > >
>> > > > > >> > > /pages/<page-id>/regions/<region-id>
with the custom
>> mime-type
>> > > > would
>> > > > > >> > render
>> > > > > >> > > a single region.
>> > > > > >> > >
>> > > > > >> > >
>> > > > > >> > > Obviously there is still room for uncertainty
in some
>> places.
>> > > For
>> > > > > >> > example,
>> > > > > >> > > what happens if your have a region with three
region
>> widgets
>> > > then
>> > > > > you
>> > > > > >> > save
>> > > > > >> > > the region but only include two ids in the
list?
>> Personally, I
>> > > > think
>> > > > > >> that
>> > > > > >> > > should delete the missing regionWidget because
that list
>> > denotes
>> > > > > >> > ordering.
>> > > > > >> > > The reason I don't like an "order" attribute
on the sub
>> > objects
>> > > is
>> > > > > that
>> > > > > >> > > what if you save two sub objects with the
same order (which
>> > > would
>> > > > > >> happen
>> > > > > >> > if
>> > > > > >> > > you ever wanted to swap two objects in order
because you
>> have
>> > to
>> > > > > update
>> > > > > >> > > them then save each one so they would have
the same order
>> at
>> > > least
>> > > > > >> > > momentarily on the server)?
>> > > > > >> > >
>> > > > > >> > > Anyway, my 2-cents.
>> > > > > >> > >
>> > > > > >> > > Chris
>> > > > > >> > >
>> > > > > >> >
>> > > > > >>
>> > > > >
>> > > >
>> > >
>> >
>>
>
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message