fix(keycloak): tests on aarch64, use image from [jboss -> quay], change supported version [16+ -> 18+]

[max-pfeiffer]

• jboss/keycloak is discontinued, adapting the official Docker image: https://quay.io/repository/keycloak/keycloak
• switched to the latest image version with ARM support

fixes #451
fixes #483

[santi]

Great work! Tested locally on my Apple Silicon M3 and works wonders 👍 Just a small nitty-gritty detail, decide yourself if you want to do it or not.

[matheusvnm]

@max-pfeiffer, allow me to give my 2 cents: The new Keycloak version has a specific endpoint for readiness checks (/health/ready as can be seen here). It can be activated by setting the KC_HEALTH_ENABLED=true. Why not use it to reliably check the state of the system?

You could do something like below:

def _configure(self):
    self.with_env('KEYCLOAK_ADMIN', self.username)
    self.with_env('KEYCLOAK_ADMIN_PASSWORD', self.password)
    self.with_env('KC_HEALTH_ENABLED', 'true')
    self.with_command('start-dev')

@wait_container_is_ready(requests.exceptions.ConnectionError, requests.exceptions.ReadTimeout)
def _connect(self):
    response = requests.get(f"{self.get_url()}/health/ready", timeout=1)
    response.raise_for_status()

[max-pfeiffer]

@matheusvnm Good point! I was not aware that these endpoints became introduced. I agree: this would be a better way to implement a readiness probe.

Problem is here: it looks like these endpoints became introduced with v18.0. So we would cut off all the older versions which people might still be using.

But we also want to have ARM support here. Images with ARM64 support are available since v17.0, I just checked the container registry.

So supporting versions from 18.0 to 24.0.1 could be an acceptable trade off.

[santi]

Keycloak themselves only support the latest major version, so shrinking the supported version range to 18.x+ sounds fine by me.

[alexanderankin]

if anyone really needs ancient versions im sure we can put together some kind of code sample with extending the core class DockerContainer with whatever they need, but it doesn't need to be in the library that gets published to pypi. I think we have enough of a consensus on that to merge? i even remember seeing something about trying to get keycloak to maintain their own testcontainers, so im not trying to overthink this.