Keycloak on Quarkus is basically a Quarkus Extension. For more details about extensions, please take a look at [Writing Your Own Extension](https://quarkus.io/guides/writing-extensions) guide.
As an extension, the server can be embedded in any Quarkus application by adding the following dependency:
```
<dependencies>
<dependency>
<groupId>org.keycloak</groupId>
<artifactId>keycloak-quarkus-server</artifactId>
</dependency>
</dependencies>
```
Just like any other extension, the server extension has two main modules:
*`deployment`
*`runtime`
Within the `deployment` module you'll find the implementation for the build steps that run when (re)augmenting the application. This module
is responsible for all metadata processing, such as:
* Lookup SPIs and their providers and enable/disable them accordingly
* Create a closed-world assumption about the providers that should be registered to the `KeycloakSessionFactory`
* Customizations to how Hibernate, Resteasy, Liquibase, Infinispan, and other dependencies are configured and bootstrapped
* Creating Jandex indexes for some server dependencies
The output of this module is bytecode generated by Quarkus that improves the server startup time and memory usage so that any processing
done during (re)augmentation does not happen again when actually starting the server. Note that the code from this module does
not run at all at runtime (when running the server) but only when building the project (the `server` module in particular, more on that later) or when triggering
a re-augmentation when running the `build` command.
Within the `runtime` module you'll find the code that is run at runtime when starting and running the server. The main link between
the `deployment` and `runtime` modules is the `org.keycloak.quarkus.runtime.KeycloakRecorder` class. The `KeycloakRecorder` holds
a set of methods that are invoked from build steps that will end up in recorded bytecode. The recorded bytecode executes those methods at runtime
just like they were called when a build step is run, with the exact same arguments and values, if any.
As an example, let us look at how the SPIs and their providers are discovered and how the `KeycloakSessionFactory` is created.
Quarkus will run the `org.keycloak.quarkus.deployment.KeycloakProcessor.configureProviders` build step whenever you are (re)augmenting the server. The main outcome from this method
is to invoke the `org.keycloak.quarkus.runtime.KeycloakRecorder.configSessionFactory` so that the code from this method is executed at
*runtime* with a closed-world assumption about the providers that should be registered to the `KeycloakSessionFactory`. At runtime the code from that method is executed
and a `KeycloakSessionFactory` is created without any processing to discover SPIs and their providers.
There are a few reasons why we have Keycloak as a Quarkus extension:
* More control over the build-time augmentation and runtime stages.
* More flexibility when extending Quarkus itself and the extensions used by the server (e.g.: Hibernate, Resteasy, etc.)
* Make it easier to embed the server into Quarkus applications
* Make it possible to allow Keycloak developers to customize the server distribution
## Keycloak Server
The `server` module holds the Keycloak Server itself. It is basically a regular Quarkus application using the `keycloak-quarkus-server` extension. If you look at the
[pom.xml](server/pom.xml) from this module you'll notice that it is a very simple Quarkus application with the bare minimum configuration
to actually build and run the server.
As Quarkus application, the server is a [mutable](https://quarkus.io/guides/reaugmentation#how-to-re-augment-a-quarkus-application) application using the
[mutable-jar](server/src/main/resources/application.properties) package type. As a mutable application, Keycloak is able to allow
users to configure some aspects of the server without having to re-build this module, something impractical from a user perspective.
The mutability of the server is directly related to the `build` command. As mentioned before, the build steps from the `deployment` module
are only run when (re)augmenting the server and when running the `build` command, the server will indicate to Quarkus that the build steps
should re-run and the recorded bytecode should be updated to reflect any change to the server configuration.
From a Quarkus perspective, the server is also a [Command Mode Application](https://quarkus.io/guides/command-mode-reference) and provides
a CLI based on [Picocli](https://picocli.info/). As such, there is a single entrypoint that executes the code to execute the CLI and bootstrap the server. This entry point
is the `org.keycloak.quarkus.runtime.KeycloakMain` class from the `runtime` module.
## Keycloak Distribution
The server distribution is created by build the `dist` module. This module basically consists of packaging the `deployment`, `runtime`,
and `server` modules artifacts and their dependencies to a `ZIP` or tarball file.
Within this directory you'll find the directory structure of the distribution and what is included in it.
## Running the server in development mode
As a regular Quarkus application, you are able to run the `server` module in [dev mode](https://quarkus.io/guides/maven-tooling#dev-mode) just like any regular application:
```
cd server
mvn clean quarkus:dev -Dquarkus.args="start-dev"
```
You can set any command or configuration option to the server by setting the `quarkus.args` environment variable.
When running in dev mode, you can benefit from the dev mode capabilities from Quarkus but with some limitations. The main limitations you'll find
at the moment are:
* Changes are only automatically reflected at runtime if you are changing resources from the `deployment`, `runtime`, and `server` modules. Other modules, such as `keycloak-services` still rely on Hot Swap in Java debuggers to reload classes.
* There is nothing in the Dev UI related to the server itself, although you can still change some configuration from there.
* There are some limitations when passing some options when running in dev mode. You should expect more improvements in this area.
We are working to improve the dev experience, and you should expect improvements over time.
## Debugging the server distribution
The `kc.sh|bat` script allows you to remotely debug the distribution. For that, you should run the server as follows:
```
kc.sh --debug start-dev
```
By default, the debug port is available at `8787`.
An additional environment variable `DEBUG_SUSPEND` can be set to suspend the JVM, when launched in debug mode. The `DEBUG_SUSPEND` variable supports the following values:
*`y` - The debug mode JVM launch is suspended
*`n` - The debug mode JVM is started without suspending
Suspending the JVM when in debug mode is useful if you want to debug the early stages of the bootstrap code.
When making changes to the `deployment`, `runtime`, or `server` modules, you can update the distribution with the new artifacts by executing
the following command:
```
mvn -DskipTests clean install
```
After the `quarkus` module and sub-modules are built, you can update the distribution as follows:
```
cp -r server/target/lib ${KC_HOME_DIR}
```
In the example above, the `${KC_HOME_DIR}` variable points to the root directory of the distribution.
You should also be able to update a server dependency directly. For that, copy the jar to the following location:
In order to avoid using external services for DNS resolution, the tests are executed using a local host file by setting the `-Djdk.net.hosts.file=${project.build.testOutputDirectory}/hosts_file` system property.
* Change the dependencies we are using from Quarkus
* Run a build to make sure the server extension is not broken
The steps still require a lot of manual work, and we should be improving this.
### Changing the Quarkus version
To change the Quarkus version, you can run the following script:
```bash
./set-quarkus-version.sh <version>
```
The `set-quarkus-version.sh` script is enough to change the version for all dependencies we are
using from Quarkus.
The `<version>` should point to a branch or a tag from Quarkus repository.
It is also possible to change to a snapshot version by running:
```bash
./set-quarkus-version.sh
```
### Run a local build
After changing the dependency versions, you can run a local build to make sure the server extension is not broken by API changes and if
all tests are passing:
```
mvn clean install
```
### Changing versions of JDBC Extensions
It might happen that when upgrading a version for any of the JDBC extensions (e.g.: `quarkus-jdbc-postgresql`) you also need to make sure the server extension is using the same JDBC Drivers.
For that, you should look at the `deployment` module of the corresponding JDBC extension from Quarkus (e.g.: https://github.com/quarkusio/quarkus/blob/main/extensions/jdbc/jdbc-postgresql/deployment/src/main/java/io/quarkus/jdbc/postgresql/deployment/JDBCPostgreSQLProcessor.java) to check if they match with the drivers used by the server extension by looking at the `org.keycloak.config.database.Database` class.
### Changing versions of Quarkiverse dependencies
Make sure the Quarkiverse dependencies are also updated and in sync with the Quarkus version you are upgrading.
For now, we only have this Quarkiverse dependency:
* https://github.com/quarkiverse/quarkus-vault
### What can go wrong when upgrading?
The perfect scenario is that after performing all the steps above the server extension will compile, the distribution can be built,
and all tests will pass.
However, it is expected breaking changes between Quarkus upgrades that break the integration code we have in both [deployment](deployment) and [runtime](runtime) modules. When this happens,
you should understand what is breaking and upgrade the integration code accordingly.
