Open Stack

It seems like there's some reasonable support for going in the direction of HATEOS-esque links for defining capabilities... I suppose it'd be nice to see one of the more extreme examples taken to its logical conclusion. Could someone more familiar with the breadth of Nova's extensions write out what it would look like if you actually provided links for every action reasonably available on a server?


-          Gabriel

From: Dolph Mathews [mailto:dolph.mathews at gmail.com]
Sent: Monday, May 13, 2013 9:26 AM
To: OpenStack Development Mailing List
Subject: Re: [openstack-dev] [openstack-tc] Proposal for API version discovery


On Thu, May 9, 2013 at 8:40 AM, Mark Washenberger <mark.washenberger at markwash.net<mailto:mark.washenberger at markwash.net>> wrote:
I'm not sure that is really 100% of the problem here, but we are definitely doing links in an inefficient way these days, due to an early misconception some of us on the Nova team had about *how* to do hyperlinks in json.

Json schema [1] provides ways to indicate, _in the schema of an object_, how it is linked to other objects. For example, if I had an object

GET /widgets/foo
{
  'name': 'foo',
  'self': '/widgets/foo',
  'bar': '/widgets/foobar',
}

I might use a schema to communicate that 'self' is a self link, and that 'bar' is a related resource by publishing a schema

GET /schemas/object
{
   'type': 'object',
   'properties': {
      'name': {'type': 'string'},
      'self': {'type': 'string'},
      'bar': {'type': 'string'},
   },
   'links': [
      {'rel': 'self', 'href': '{self}'},
      {'rel': 'related', 'href': '{bar}'}
   ]
}

What this schema says is that name, self, and bar are string types, that you can construct a self link using the value of the "self" property, and that you can construct a related link using the value of the "bar" property.

For some reason I cannot remember, we neglected to realize that link descriptor objects [2] were only intended for json schema objects, not the regular objects described by the schemas. In the process, we made hyperlinks in our apis really inefficient by loading them up with unnecessary 'rels' and 'hrefs'. We also made finding relevant links difficult by requiring clients to iterate through an array rather than just explicitly looking up the relevant properties.

Glance v2 api fixes this problem. But we also tried to go all-in on being schema driven. This combined approach is fine for Glance, but I'm not sure that the "schema-driven" part is necessarily that useful for others. In any case, we needn't have such bloated json hyperlinks for future openstack apis. For example, if you want to have actions listed for a server, you could just do

{
  "id": "abcd",
  ...
  "actions": {
    "stop": "/servers/abcd/stop",
    "snapshot": "/servers/abcd/snapshot"
  }
}


+1; identity api v3 addresses this as well, but uses a "links" container to avoid namespace pollution.

  {
    "entity": {
      "id": string,
      "name": string
      "links": {
        "self": url,
        "related_collection": url
      }
    }
  }

https://github.com/openstack/identity-api/blob/master/openstack-identity-api/src/markdown/identity-api-v3.md#create-an-entity


which is a bit more compressed.

[1] - http://tools.ietf.org/html/draft-zyp-json-schema-03
[2] - http://tools.ietf.org/html/draft-zyp-json-schema-03#section-6.1

On Wed, May 8, 2013 at 10:33 PM, Mark McLoughlin <markmc at redhat.com<mailto:markmc at redhat.com>> wrote:
On Wed, 2013-05-08 at 18:59 +0000, Gabriel Hurley wrote:
> > -----Original Message-----
> > From: Vishvananda Ishaya [mailto:vishvananda at gmail.com<mailto:vishvananda at gmail.com>]
> > Sent: Tuesday, May 07, 2013 3:44 PM
>
> > It is relatively easy to include the info at servers/<uuid> however:
> > {
> >  "features": ["OS-EXT-SRV-ATTR", "os-start-stop"],
> >  "server": {
> >    "OS-EXT-SRV-ATTR:host": "1169a68456af48238da47b1d5957a714",
> >    "OS-EXT-SRV-ATTR:hypervisor_hostname": "fake-mini",
> >    "OS-EXT-SRV-ATTR:instance_name": "instance-00000001",
> >    "links": [
> >      {"rel": "self",
> >       "href":"http://example.org/v3/servers/009e9aca-b765-11e2-855a-
> > 008cfa10b980"},
> >      {"rel": "start",
> >       "href":"http://example.org/v3/servers/009e9aca-b765-11e2-855a-
> > 008cfa10b980/start"
> >       "feature": "os-start-stop"},
> >      {"rel": "stop",
> >       "href":"http://example.org/v3/servers/009e9aca-b765-11e2-855a-
> > 008cfa10b980/stop"
> >       "feature": "os-start-stop"}.
> >    ]
> >   }
> > }
>
> Tht perfectly illustrates my biggest concern with doing this on a
> per-resource basis.
Per-resource does allow you to expose whether a user has permission to
perform an action, though.

>  The list of "links" gets unconscionably enormous. If I do a "list"
> call and retrieve 100 instances each with 100 links because you have
> 20 extensions the whole thing becomes a mess. I think it's much more
> sane (though less explicit) to understand what's provided by the
> service and extrapolate that to the individual resources therein.
Yeah, lots of links can be messy. How about URI templates?

Maybe we have them as a /capabilities/templates resource:

  {"templates": [
     {"rel": "server_start_action",
      "href": "http://example.org/v3/servers/{server_uuid}<http://example.org/v3/servers/%7Bserver_uuid%7D>"
     }
  ]}

then just do:

  "links": [
    {"rel": "start", "template": "server_start_action"}
  ]

Just a thought ...

Cheers,
Mark.


_______________________________________________
OpenStack-TC mailing list
OpenStack-TC at lists.openstack.org<mailto:OpenStack-TC at lists.openstack.org>
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-tc


_______________________________________________
OpenStack-dev mailing list
OpenStack-dev at lists.openstack.org<mailto:OpenStack-dev at lists.openstack.org>
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-dev/attachments/20130514/bfc784ec/attachment.html>