Package org.keycloak.quarkus.runtime.integration.health

Class ReactiveHealthHandler

java.lang.Object
org.keycloak.quarkus.runtime.integration.health.ReactiveHealthHandler
All Implemented Interfaces:
io.vertx.core.Handler<io.vertx.ext.web.RoutingContext>
Direct Known Subclasses:
ReactiveLivenessHandler
public abstract class ReactiveHealthHandler extends Object implements io.vertx.core.Handler<io.vertx.ext.web.RoutingContext>
This adds the possibility to have a non-blocking health handler in Quarkus.

Without a non-blocking health check, all liveness and readiness probes will enqueue in the worker thread pool. Under high load of if there is a lot of blocking IO happening (for example, during Keycloak cluster rebalancing), this leads to probes being queued. Queued probes would lead to timeouts unless the timeouts are configured to 10-20 seconds. Reactive probes avoid the enqueueing in the worker thread pool for all non-blocking probes, which will be the default for the (otherwise empty) liveness probe. For the readiness probe, this depends on the implementation of the specific readiness probes.

This is a workaround until quarkusio/quarkus#35100 is available in a regular Quarkus version. Then these classes can be removed.

Author:
Alexander Schwartz

  • Constructor Details

    • ReactiveHealthHandler

      public ReactiveHealthHandler()

  • Method Details

    • handle

      public void handle(io.vertx.ext.web.RoutingContext context)
      Specified by:
      handle in interface io.vertx.core.Handler<io.vertx.ext.web.RoutingContext>

    • getHealth

      protected abstract io.smallrye.mutiny.Uni<io.smallrye.health.SmallRyeHealth> getHealth()