fix(operator): Scale statefulset to 0 to prepare for update (#30450)
When performing a keycloak update, the operator is supposed to make sure that potential database migrations are run with only one pod active. This change makes the operator scale down the stateful set to zero pods in preparation for the update. The next reconciliation loop will scale the stateful set back up and change the image, making sure migrations are being run on the first pod that is brought up. This also makes sure that the rollover works even if the infinispan versions are incompatible. (ref: #30449) Signed-off-by: Schmidt, Sascha (sasschmidt) <sascha.schmidt@breuninger.de>
This commit is contained in:
parent
333f1fb7ee
commit
13ef6fb1c8
1 changed files with 2 additions and 2 deletions
|
@ -461,13 +461,13 @@ public class KeycloakDeploymentDependentResource extends CRUDKubernetesDependent
|
|||
var reconciledContainer = reconciledDeployment.getSpec().getTemplate().getSpec().getContainers().get(0);
|
||||
|
||||
if (!previousContainer.getImage().equals(reconciledContainer.getImage())
|
||||
&& previousDeployment.getStatus().getReplicas() > 1) {
|
||||
&& previousDeployment.getStatus().getReplicas() > 0) {
|
||||
// TODO Check if migration is really needed (e.g. based on actual KC version); https://github.com/keycloak/keycloak/issues/10441
|
||||
Log.info("Detected changed Keycloak image, assuming Keycloak upgrade. Scaling down the deployment to one instance to perform a safe database migration");
|
||||
Log.infof("original image: %s; new image: %s", previousContainer.getImage(), reconciledContainer.getImage());
|
||||
|
||||
reconciledContainer.setImage(previousContainer.getImage());
|
||||
reconciledDeployment.getSpec().setReplicas(1);
|
||||
reconciledDeployment.getSpec().setReplicas(0);
|
||||
|
||||
reconciledDeployment.getMetadata().getAnnotations().put(Constants.KEYCLOAK_MIGRATING_ANNOTATION, Boolean.TRUE.toString());
|
||||
}
|
||||
|
|
Loading…
Reference in a new issue