Answering Stian's and Marek's email at the same time.
Just FYI - I've created a ticket to track this work -
On Wed, Jul 25, 2018 at 8:16 PM Marek Posolda <mposolda(a)redhat.com> wrote:
IMO Creating keystore files programatically as Peter suggested,
better than creating them through maven-antrun-plugin. It seems that
creating with maven will mean lots of the stuff in pom.xml files,
however generating programatically will probably allow some better
reusability (common utility class in the testsuite to easily generate
keystore in various locations). Also programatic generation will work
when running the test from Intellij Idea without previous maven build.
Perhaps I didn't make it clear enough - the JKS autogeneration script will
mostly be executed in our test suite. I've just checked if you can run any
of the tests with a clean repo (without invoking `mvn package` or `mvn
install`) and it failed with . So it seems you always need to execute
Maven prior to invoking any tests from the IDE (at least from the
The programmable approach Peter suggested looks very nice. However, I'm not
sure if this is a good fit for us since we modify app servers during the
build . Those servers expect a keystore (and sometimes a truststore too)
to be in certain places upon the server bootstrap. When the Arquillian
testsuite if fired, it's simply to late to generate keystores on the fly.
Sebastian, yesterday I didn't mention on the call that we
test X509 authentication with IBM JDK. I recalled it just now :) The
thing is, that X509 tests are not executed during default build as they
require Wildfly (they don't work on embedded undertow) and also
phantomJS (no other browser is supported). So I am pretty sure they are
not executed during the various matrix jobs for various java versions
(Vasek Muzikar can correct me if I am wrong). Which means they probably
don't work on IBM JDK too.
In Infinispan we had exactly the same problem. Keystores generated by
OpenJDK (or Oracle JDK) didn't work for IBM JDK. Therefore we added this
 little trick.
On 25/07/18 15:50, Stian Thorgersen wrote:
> Sounds like this could be a time-consuming task and the tests are working
> at the moment.
I think I've solved the worst case - in some of the scenarios (like
`DemoServletsAdapterTest#testOIDCUiLocalesParamForwarding`) we expect the
certificate in keystore to be the same as the one from the realm JSON
configuration. I might be wrong here (I'm still the "new guy") but the
reason for this is that keystore.jks is used by the adapter whereas the
auth server uses the certificate from json file. Those need to match.
Therefore, I had to add a small piece of the code that does automatic
replacement of PEM certificates in the configuration . I could probably
use some latest syntactic sugar (Optional#flatmap and Streams) but I think
the old way is more readable in this case (but of course, comments are more
than welcome, especially on the PR).
So not sure I see a need to start a bulk refactoring here.
Even though it might seem like a redundant thing to do, I would still
prefer to replace all existing JKS entries. The main reason for this is
that I'm a bit afraid that we'll end with a mixture of pushed JKS files and
automatically generated ones.
Let's make a deal. I'll look more into this today and will see how many of
those I'll manage to replace. Maybe it won't be that bad after all...
> The json file you're mentioning has a cert in pem format in
the json file
> that is then stored in the DB. This doesn't have anything to do with
> keystores/truststores. We expect Keycloak to be able to handle a cert
> provided in the json file (or through admin endpoints) and store it in
> Now for SSL tests with the server itself I can see the value in what you
> are proposing. Keystore/truststore for Keycloak https connections and
> truststores for mutual SSL could probably be improved. However, it seems
> be working at the moment, so not sure this is required.
> On Wed, 25 Jul 2018 at 15:30, Sebastian Laskawiec <slaskawi(a)redhat.com>
>> Thanks Darran and Peter for the insight!
>> I've just noticed that generating JKS files is one thing but we will
>> need to pre-process RealmRepresentation and replace all private/public
>> (like this one for example ) with values from keystores/truststores.
>> At the moment I'm thinking about replacing  with something like this:
>> "jwt.credential.certificate" :
>> similarly to private keys (if needed):
>> "privateKey" :
>> Later on, while converting JSON to RealmRepresentation I plan to
>> all nodes and replace replace those which start with private-key or
>> Let me know if you have a better idea.
>> On Wed, Jul 25, 2018 at 2:07 PM Peter Skopek <pskopek(a)redhat.com>
>>> Hi Sebastian,
>>> I think that generation of Keystore on the fly is the only way to go.
>>> You can use Java Keystore API generate it entirely from Java code (no
>>> need to run maven).
>>> Here  is an example of utility class used in wildfly testsuite to
>>> generate Vault (basically just keystore).
>>> It will be easy to strip vault related stuff and use it in our
>>> Note that there is one more drawback of storing generated JKS files,
>>> which is that in some cases they are not binary compatible between
>>> different JDK (particularly Oracle and IBM).
>>> On Wed, Jul 25, 2018 at 1:23 PM Sebastian Laskawiec <
>>>> Together with Sebi we are working on Certificate-based authentication
>>>> the clients. Our work will require adding at least 2-3 keystores to
>>>> codebase with different DNs and I think this might be a good
>>>> revisit the way we handle JKS files in the tests.
>>>> Currently we push JKS files directly into our repo, which has a couple
>>>> - it is hard to figure out what's inside the JKS, it requires
>>>> for password (usually in some JSON configuration file or hardcoded in
>>>> test) and using keytool (or some similar one) to explore its content.
>>>> - It is not git-friendly. Every time we update JKS content we
>>>> store another binary file (git doesn't understand binary file
>>>> can not diff it).
>>>> - we use many different naming and password schemes in our tests.
>>>> - it is hard to migrate all keystores to pkcs12 at the same time (JKS
>>>> format is deprecated) 
>>>> I believe most of the issues could be addressed by generating JKS
>>>> the fly - during the build. In Infinispan we did it with Maven Ant
>>>> . I already created a very limited POC for Keycloak and you can
>>>> out here . Unfortunately, the process of reverse engineering all
>>>> files is quite time-consuming, so I would like to know your opinion
>>>> moving on.
>>>> Of course, generating JKS files on the fly has some drawbacks:
>>>> - It increases build time (~1s per keytool invocation, and we probably
>>>> have more than 30 of them).
>>>> - it makes testing from IDE a bit harder, you need to run Maven and
>>>> test resources before doing anything. A common workaround is to use
>>>> clean install -DskipTests` and then opening your IntelliJ.
>>>> Please let me know what you think.
>>>>  https://github.com/keycloak/keycloak/pull/5410
>>>> keycloak-dev mailing list
>> keycloak-dev mailing list
> keycloak-dev mailing list