I think it would be 1 day work for me, as again, I already refactored
the adapters to support this use case. It would just be a matter of:
* Making AdatperDeploymentContext pluggable
* Writing an AdapterDeploymentContext which could accept a URI pattern
to extract the name of the realm.
Once I have that in place, you guys can fork it and implement anything
you want. What I think I want to do, is keep this SPI private until
users like you have testdrived it.
On 9/4/2014 7:17 PM, Travis De Silva wrote:
Hi Bill,
Sorry for missing your ping as this is something that we definitely
need. I was going down the keycloak.js path (since we use AngularJS as
our UI layer) but doing it on the server side is so much more elegant.
Picking the realm name from the URI is the way to go. Maybe we have it
as a query parameter rather than within the path as then it is less
invasive for the war application.
I don't understand the keycloak code base enough to comment on how we
can deploy the new AdapterDeploymentContext but what if this feature is
plugged into the current AdapterDeploymentContext and this is a feature
of the core product?
Also with regard to getting realm information from the server using a
shared client secret, or public clients, another way to do this might be
to provide an alternate way to pick the keycloak.json file by storing it
outside the war in the file system and then based on the realm name in
the uri, pick the corresponding keycloak.json file and run the
KeycloakDeployment. We can name the the files as keycloak-{realmname}.json
Note we can keep the current functionality where it can pick it from
within the war but if the file is missing, currently its throwing an
exception. Maybe before we throw the exception, we also check the file
system as well.
Then maybe we don't need to load this on request but can have a
directory scanner and whenever a new file is added or removed, it will
automatically pick it up. Sort of how the JBoss/Wildfly deployment
scanner works. On each request of course it will need to pick the
correct realm to perform the authentication.
This might be more elegant but once again I don't know enough of the
core keycloak code to comment if doing this is more complex than the
other option.
Obviously these changes will not go into 1.0 release but a subsequent
release (hopefully the first beta release after 1.0 :)
Therefore it might be good to give some thought and get this right. For
me this and the multi-lingual are the two key items that we need to tick
off to be able to use this in a multi tenancy environment.
Keen to know your thoughts.
Cheers
Travis
On Thu, Sep 4, 2014 at 11:15 PM, Bill Burke <bburke(a)redhat.com
<mailto:bburke@redhat.com>> wrote:
Travis, I did do most of the work for this. I think I pinged you to see
if you still wanted the feature, but never followed through. I'm sorry.
All this would require a shared client secret, or public clients. It
would require you to extract the realm name somehow based on the current
HTTP request. Probably a URI pattern.
There is an AdapterDeploymentContext class. This class has a method:
KeycloakDeployment resolveDeployment(HttpFacade)
This method get's called every request. You would extend this class and
override resolveDeployment and create (and then cache) your
KeycloakDeployment based on the incoming HTTP request.
The only problem is that the current code has no way for you to plug in
a new implementation of the AdapterDeploymentContext.
On 9/4/2014 2:36 AM, Travis De Silva wrote:
> Hi Stian,
>
> You proposed solution would not cover the use case where we can
create
> tenants at runtime as the realm config in the keycloak.json would be
> hard coded into the war.
>
> I had discussed this identical use case a while ago on this forum and
> Bill was planning to refactor the adapters to support this use case.
> Unfortunately he got caught up in other tasks and was not able to
> proceed on this.
>
> The discussion thread is here
>
http://lists.jboss.org/pipermail/keycloak-user/2014-March/000062.html
>
> Basically what I believe Bill suggested which would meet this use
case
> is to:
>
> 1. Have a shared secret between clients for all realms.
> 2. The adapter would just extract the realm name from the request,
> invoke on the keycloak server to get the public information about
> the realm (i.e. public key) and then cache the information
locally.
>
> The key bit here is extracting the realm name from the request
and then
> pulling the realm info from the keycloak server.
>
> I had a look at the keycloak source code and I believe the magic
happens
> in the KeycloakServletExtension class under the
> org.keycloak.adapters.undertow package for my use case (since I
deploy
> it on wildfly)
>
> What I have got stumped is that this class gets loaded when my war is
> deployed and I am wondering how I can do it per request (if the
info is
> not already cached locally)
>
> Maybe with the imminent release of 1.0 (btw congrats for the
great work
> to everyone in the team and for Bill and your leadership), maybe we
> should start thinking about this multi tenancy use case to be
included
> in future releases.
>
> I believe that SaaS models are going to be popular and having this
> feature added will make keycloak a major player in this space.
>
> Cheers
> Travis
>
>
>
>
> _______________________________________________
> keycloak-user mailing list
> keycloak-user(a)lists.jboss.org <mailto:keycloak-user@lists.jboss.org>
>
https://lists.jboss.org/mailman/listinfo/keycloak-user
>
--
Bill Burke
JBoss, a division of Red Hat
http://bill.burkecentral.com
_______________________________________________
keycloak-user mailing list
keycloak-user(a)lists.jboss.org <mailto:keycloak-user@lists.jboss.org>
https://lists.jboss.org/mailman/listinfo/keycloak-user