What do you mean by you're "pretty sure clustering/session replication is configured correctly".  Either its working or it isn't.   Have you tested a simple application WITHOUT keycloak and with Round Robin to determine if replication is set?    Because from what you are describing and what Stian said to you over and over again is that it does not look like HTTP session replication is working for you.  Keycloak client adapter stores that you are authenticated within the HttpSession.  It is possible that session replication is lagging I think...only if you have async replication turned on.  I don't know what the default is in Wildfly 10.

Also, a forbidden exception might mean that your web.xml requires a role set for the user for the URL you are accessing and the user does not have that role assigned to them.

On 1/25/2016 2:33 PM, Christian Beikov wrote:
I don't want to be rude but you aren't helping me at all so I'd like to ask someone else from the team about this. I tried to explain multiple times that I already configured clustering in my Wildfly server, thus I configured session replication.
Either I don't understand something about session replication or I require a configuration that you don't mention anywhere.
I copied over the cache container from the standalone-ha configurations and configured the JGroups Subsystem which is what is described in the official documentation: https://docs.jboss.org/author/display/WFLY10/High+Availability+Guide

I think I gave you enough information about my situation which you seemed to ignore completely. I would very much appreciate if I got a clear answer to my question especially since I am not asking for general configuration help, but for this special case which to me seems like a problem/limitation that should be explicitly documented.

I am pretty sure clustering/session replication is configured correctly since I am using a custom cache container with a similar configuration like the web cache container which I know replicates correctly. So here comes my question once again, is it possible that the replication just lags behind which makes the usage of round robin completely impossible in the login flow? Or is there some kind of special configuration I have to do which differs from the standard cluster configuration as provided by the wildfly distribution? Or is this maybe even an implementation limiation of the server adapter?

Regards,
Christian

Am 25.01.2016 um 19:04 schrieb Stian Thorgersen:
Try google for wildfly replicate http sessions

On 25 January 2016 at 15:53, Christian Beikov <christian.beikov@gmail.com> wrote:
I just wrote that I configured clustering for my application just like in the standlone-ha.xml of my Wildfly 10 CR4.
I configured the jgroups subsystem and the distributed caches for web sessions as it is done in standalone-ha.xml of Wildfly.
If there is anything else that should be configured, can you please point me to that configuration option?

Regards,
Christian


Am 25.01.2016 um 15:45 schrieb Stian Thorgersen:
HTTP session replicate is not enabled by default. You need to enable it for your application.

On 25 January 2016 at 14:39, Christian Beikov <christian.beikov@gmail.com> wrote:
The documentation states, that the default token-store is "session" and as I wrote before, I have setup clustering on my Wildfly 10 CR4 like in standalone-ha.xml, so the session should already be replicated.

Regards,
Christian


Am 25.01.2016 um 14:20 schrieb Stian Thorgersen:
Your issue doesn't have anything to do with the Keycloak server side user sessions, they don't require sticky sessions. 

Your issue is down to the http session on the adapter side not being replicated by default. For the adapter you've got 3 choices: sticky session, replicated session or stateless. Which is best depends on your application.


On 25 January 2016 at 14:05, Christian Beikov <christian.beikov@gmail.com> wrote:
I don't have a problem with sticky sessions and I will definitively configure them, but I am curious. What is the reason for the problems with round robin in this test scenario? Are the infinispan caches not replicated fast enough or is there an implementation limitation in the adapters?

Regards,
Christian


Am 25.01.2016 um 08:58 schrieb Stian Thorgersen:
By default the adapters will require sticky sessions, please refer to http://keycloak.github.io/docs/userguide/keycloak-server/html/applicationClustering.html for more information

On 22 January 2016 at 12:48, Christian Beikov <christian.beikov@gmail.com> wrote:
Hello,

I am running some tests with my application cluster being secured by a
single keycloak server instance and I encountered problems with the adapter.

My application cluster contains 2 nodes and is load balanced by nginx.
For testing purposes, I enabled round robin load balancing which is
probably the "cause" for my issues.

When I access a secured page, I get redirected to keycloak and
everything is fine. When I then login, and keycloak redirects me back to
the application, I get to a different application cluster node because
of round robin. On that node, apparently the initial information of the
client session is not available and I get redirected to keycloak login
page again. Then keycloak redirects me back to the application, this
time to the original node, and says that access is forbidden.

I suppose the web session caches are not in sync but I just used the
default cache containers as they are defined in standalone-ha.xml of my
Wildlfy 10 CR4. Clustering with jgroups works, as I use other
distributed caches too which work just fine.

We are using Keycloak 1.8.0.CR2 on a Wildfly 10 CR4

Regards,
Christian
_______________________________________________
keycloak-dev mailing list
keycloak-dev@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/keycloak-dev










_______________________________________________
keycloak-dev mailing list
keycloak-dev@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/keycloak-dev

-- 
Bill Burke
JBoss, a division of Red Hat
http://bill.burkecentral.com