[openstack-dev] [swift] [keystone] Keystone v3 API domains in Swift
Yee, Guang
guang.yee at hp.com
Wed Jan 23 20:39:08 UTC 2013
Swift ACLs currently can be specified as
<project/tenant ID>:<username>
<project/tenant name>:<username>
*:<username>
IDs are always globally unique regardless. Names (username, project/tenant
name, etc), on the other hand, are currently required to be globally unique.
But there are attempts to relax the globally uniqueness on names constraint.
The notion of private domain is one of them. I am still unclear on how
private domains will impact the existing services.
1. How does private domain work in conjunction with public domains? If
I have a user jdoe in private domain XYZ and another jdoe in a public domain
ABC, how does not affect the existing Swift ACLs? Seems to me, private
domain or not, Swift ACL will have to change once we relax this constraint.
2. How does private domain affect middleware? Are we going to have a
X-Domain-ID, X-Domin-Name, and X-Is-Private-Domain headers?
I think we need a full impact assessment before moving forward with this.
Guang
From: Dolph Mathews [mailto:dolph.mathews at gmail.com]
Sent: Wednesday, January 23, 2013 9:52 AM
To: OpenStack Development Mailing List
Subject: Re: [openstack-dev] [swift] [keystone] Keystone v3 API domains in
Swift
On the surface, this seems to be defeating the purpose of introducing
domains in the first place (allowing domain administrators to manage
isolated groups of users & projects). I'm not sure why you'd need to switch
away from tenant_id/project_id.
-Dolph
On Wed, Jan 23, 2013 at 4:23 AM, Alexandra Shulman-Peleg
<SHULMANA at il.ibm.com> wrote:
Hi,
I would like to get back to this discussion and specify the exact syntax of
ACLs that can be used when removing the global uniqueness constraint on user
names. I wander whether we really need to prefix both the project_name and
the username with the domain id? Especially, since on ACLs we mainly need to
properly identify the user and not the project. So the notion of a project
may not be required in this context? For example, in NFSv4 ACLs (also
adopted by CDMI) users are identified by username at domain. So I wander
whether on ACLs, in V3 we can simply switch from tenant_id:username to
domain_id:username? This seems to fulfill the identification requirements
and will give a very simple solution for the migration of existing v2
customers to private domains in V3 - assigning the new domain_id to match
the old tenant_id will allow preserving all of the stored containers without
the need to modify the containers' meta data.
Best Regards,
Alex.
From: "Yee, Guang" <guang.yee at hp.com>
To: OpenStack Development Mailing List
<openstack-dev at lists.openstack.org>,
Date: 11/01/2013 10:31 PM
Subject: Re: [openstack-dev] [swift] [keystone] Keystone v3 API
domains in Swift
_____
As long as Swift URL stay the same we should be OK. Frankly, there aren't
any strong arguments for changing it at this point. Whenever we remove the
globally uniqueness constraint on names, new Swift ACLs probably will need
to switch over to using namespacing.
domain_name.project_name:domain_name.username
something like that. Existing Swift ACLs should work fine since if the given
domain is the default (migrated) system domain, auth_token middleware should
not set the domains headers.
Guang
-----Original Message-----
From: David Chadwick [ <mailto:d.w.chadwick at kent.ac.uk>
mailto:d.w.chadwick at kent.ac.uk]
Sent: Friday, January 11, 2013 8:36 AM
To: OpenStack Development Mailing List
Subject: Re: [openstack-dev] [swift] [keystone] Keystone v3 API domains in
Swift
Hi Chuck
On 11/01/2013 15:59, Chuck Thier wrote:
> The Tenant_ID is in the URL
> (https://{SWIFT_IP}/v1/AUTH_{TENANT_ID}/{CONTAINER}/{OBJECT})
>
> I think we have beaten this part to death a bit now, as we seem to all
> agree that we can continue this pattern with the V3 API. The one
> concern that I still have is how the ACLs will work, and weather or
> not that will need to change.
>
> I'm also curious how the Keystone V3 API will work alongside V2 apis.
My opinion (only, I dont speak for anyone else) is as follows:
1. A v2 API system has no problems as it is working OK today
2. A v3 API system only, with domains added, should work OK tomorrow
otherwise the v3 API has problems
3. So the main point as you say is how do v2 and v3 systems interwork. I
suggest there is an intercept module, say in the Keystone pipeline, that
knows it is operating in a v2/v3 environment, and when it receives a v2
request already containing a tenant_ID it knows it will comprise
domain:project and it can unpack it, and give the separate elements to
the rest of the V3 code for processing as in a v3 system. When the
intercept module receives a v2 request that needs a tenant ID returning
to it, it will encode up the domain and project as a tenant ID, and
return this to the v2 client. The v2 client will be blissfully unaware
that what it thinks is a tenant ID is actually a combination of domain
and project.
regards
david
>
> --
> Chuck
>
> On Thu, Jan 10, 2013 at 4:16 AM, David Chadwick <d.w.chadwick at kent.ac.uk>
wrote:
>> You have to ask, where does the Swift client get the tenant_Id from? Isnt
it
>> Keystone? So if Keystone returns project_ID:tenant_Id to swift as the
>> project_id string, then Swift can continue to use this as if nothing has
>> changed. Its just a string whose content has no meaning to Swift, but
whose
>> content does have meaning to Keystone. The Swift policy simply needs to
>> change the value of the tenant_id in its policy to the new value and it
>> should work the same
>>
>> regards
>>
>> David
>>
>>
>> On 09/01/2013 20:21, heckj wrote:
>>>
>>> Given that domains are a segmentation of projects/tenants, then I
wouldn't
>>> expect to want to change it from a project_id representation to anything
>>> else.
>>>
>>> -joe
>>>
>>> On Jan 9, 2013, at 12:13 PM, Chuck Thier <cthier at gmail.com> wrote:
>>>>
>>>> Things are always easy, until you start thinking about backwards
>>>> compatibility. The storage urls for swift with keystone are currently
>>>> keyed off of the tenant_id (soon to be project_id), so you end up with
>>>> an endpoint url that looks something like
>>>> http://{SWIFT_IP}/v1/AUTH_{TENANT_ID} if you change that by adding
>>>> the domain, then you break any current users in your system, and you
>>>> can't use v2 and v3 auth contracts simultaneously.
>>>>
>>>> --
>>>> Chuck
>>>>
>>>> On Wed, Jan 9, 2013 at 1:37 PM, David Chadwick
<d.w.chadwick at kent.ac.uk>
>>>> wrote:
>>>>>
>>>>> I would have thought that the solution is conceptually rather
>>>>> straightforward. If domains can have their own project names and
>>>>> usernames,
>>>>> then you prefix the names with the domain ID or domain name to make
them
>>>>> globally unique again.
>>>>>
>>>>> regards
>>>>>
>>>>> David
>>>>>
>>>>>
>>>>>
>>>>> On 09/01/2013 19:14, Yee, Guang wrote:
>>>>>>
>>>>>>
>>>>>> Yes. Swift ACLs <tenant_id>:<user_name>, <tenant_id>:<user_name>, and
>>>>>> *:<user_name> will be impacted if project (formely tenant) name and
>>>>>> user
>>>>>> name are no longer globally unique. We'll need to figure out a
>>>>>> migration
>>>>>> path before relaxing that constraint.
>>>>>>
>>>>>>
>>>>>> Guang
>>>>>>
>>>>>>
>>>>>> -----Original Message-----
>>>>>> From: Chuck Thier [ <mailto:cthier at gmail.com>
mailto:cthier at gmail.com]
>>>>>> Sent: Wednesday, January 09, 2013 10:48 AM
>>>>>> To: OpenStack Development Mailing List
>>>>>> Subject: Re: [openstack-dev] [swift] [keystone] Keystone v3 API
domains
>>>>>> in
>>>>>> Swift
>>>>>>
>>>>>> Se responses inline:
>>>>>>
>>>>>> On Wed, Jan 9, 2013 at 4:01 AM, Henry Nash
<henryn at linux.vnet.ibm.com>
>>>>>> wrote:
>>>>>>>
>>>>>>>
>>>>>>> So there are a couple of issues intertwined in this thread:
>>>>>>>
>>>>>>> 1) Uniqueness of identifiers in Swift given the keystone Identity v3
>>>>>>> api.
>>>>>>> This is the issue of whether Swift uses tenant names (now called
>>>>>>> project
>>>>>>> names) at all to uniquely identify any objects - if it did, then it
>>>>>>> would
>>>>>>> need to also consider storing a domain name or id. From the
>>>>>>> discussion,
>>>>>>
>>>>>>
>>>>>> it
>>>>>>>
>>>>>>>
>>>>>>> sounds like tenant/project ID is used instead, which (from a
>>>>>>> uniqueness
>>>>>>> point of view) is fine. A separate issue exists needs to be
discussed
>>>>>>> around swift ACLs and whether username potentially becoming unique
>>>>>>> only
>>>>>>> within a domain will have an impact.
>>>>>>>
>>>>>>
>>>>>> For AuthN, you are correct, in that it only relies on tenant/project
>>>>>> ID. So, nothing has to be changed from that perspective. AuthZ is a
>>>>>> little more tricky. For ACLs with keystone, they are set as
>>>>>> TENANT:USER in any of the following patterns:
>>>>>>
>>>>>> *:user_name - that user from any tenant has access
>>>>>> tenant_id:user_name - that user from that tenant id has access
>>>>>> tenant_name:user_name - that user from that tenant name has access
>>>>>>
>>>>>> If project_name will not be unique in v3, then the
>>>>>> tenant_name:user_name format may have to be deprecated.
>>>>>>
>>>>>> I would be interested to hear from providers that are using keystone
>>>>>> with swift and hear which of the above use cases they are using.
>>>>>>
>>>>>>
>>>>>>> 2) Given that keystone identity v3 domains are likely to be usually
>>>>>>> used
>>>>>>
>>>>>>
>>>>>> to
>>>>>>>
>>>>>>>
>>>>>>> represent an enterprise (or "account holder" in common cloud
>>>>>>> terminology)
>>>>>>> and contain the collection of projects owned by that enterprise, is
it
>>>>>>> important for Swift to have that domain knowledge? Will there be
>>>>>>
>>>>>>
>>>>>> operations
>>>>>>>
>>>>>>>
>>>>>>> either within swift (or more likely layered on top of swift) that
need
>>>>>>
>>>>>>
>>>>>> that
>>>>>>>
>>>>>>>
>>>>>>> information? E.g. How would someone layer a billing engine on top
of
>>>>>>
>>>>>>
>>>>>> swift
>>>>>>>
>>>>>>>
>>>>>>> that could collate all the swift containers that were part of one
>>>>>>> domain?
>>>>>>> Obviously that engine could call keystone with each project_id in
turn
>>>>>>> and
>>>>>>> find the domain_id.....but that sounds pretty inefficient.
>>>>>>>
>>>>>>
>>>>>> As is, containers can already be collated for a given tenant/project
>>>>>> id. The containers for a domain is then an aggregate of the project
>>>>>> ids associated to that domain.
>>>>>>
>>>>>> I think the default should be that domains are not mapped in swift.
I
>>>>>> believe that this will also be required to facilitate backwards
>>>>>> compatibility, which brings up another interesting question -- Is
>>>>>> there an expectation that people will be able to run keystone auth
>>>>>> v2.0 and v3.0 side by side?
>>>>>>
>>>>>> --
>>>>>> Chuck
>>>>>>
>>>>>> _______________________________________________
>>>>>> OpenStack-dev mailing list
>>>>>> OpenStack-dev at lists.openstack.org
>>>>>> <http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev>
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>>>>>>
>>>>>>
>>>>>>
>>>>>> _______________________________________________
>>>>>> OpenStack-dev mailing list
>>>>>> OpenStack-dev at lists.openstack.org
>>>>>> <http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev>
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> OpenStack-dev mailing list
>>>>> OpenStack-dev at lists.openstack.org
>>>>> <http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev>
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>>>>
>>>>
>>>> _______________________________________________
>>>> OpenStack-dev mailing list
>>>> OpenStack-dev at lists.openstack.org
>>>> <http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev>
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>>>
>>>
>>>
>>> _______________________________________________
>>> OpenStack-dev mailing list
>>> OpenStack-dev at lists.openstack.org
>>> <http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev>
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>>>
>>
>> _______________________________________________
>> OpenStack-dev mailing list
>> OpenStack-dev at lists.openstack.org
>> <http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev>
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>
> _______________________________________________
> OpenStack-dev mailing list
> OpenStack-dev at lists.openstack.org
> <http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev>
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>
_______________________________________________
OpenStack-dev mailing list
OpenStack-dev at lists.openstack.org
<http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev>
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
[attachment "smime.p7s" deleted by Alexandra Shulman-Peleg/Haifa/IBM]
_______________________________________________
OpenStack-dev mailing list
OpenStack-dev at lists.openstack.org
<http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev>
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
_______________________________________________
OpenStack-dev mailing list
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/20130123/78cbe0c8/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/x-pkcs7-signature
Size: 6186 bytes
Desc: not available
URL: <http://lists.openstack.org/pipermail/openstack-dev/attachments/20130123/78cbe0c8/attachment.bin>
More information about the OpenStack-dev
mailing list