Quarkus SmallRye Fault Tolerance @Bulkhead missing queue size configuration

Answer the exact question first

Yes, you can bound the number of pending requests for a @Bulkhead without touching the shared Vert.x worker pool. SmallRye Fault Tolerance exposes a queueSize attribute that is mapped to the Quarkus configuration property quarkus.smallrye-fault-tolerance.bulkhead.queue-size. Setting this property to a positive integer enables a bounded queue; a value of 0 (the default) disables queuing and causes immediate rejections.

How to configure it

1. Ensure you are running Quarkus 3.x or later. The queueSize support was added in SmallRye Fault Tolerance 2.2, bundled with Quarkus 3.0.
2. In src/main/resources/application.properties add:

   quarkus.smallrye-fault-tolerance.bulkhead.queue-size=10
   

   This sets a global queue limit of 10 for all bulkheads.
3. Alternatively, annotate a specific bean or method:

   @Bulkhead(value = 5, queueSize = 10)
   public String doWork() {
       // …
   }
   

   Here the bulkhead will allow 5 concurrent executions and queue up to 10 additional calls.

Isolation of the queue from the global Vert.x pool

• The queue is implemented by the underlying executor that the bulkhead uses. Quarkus creates a dedicated executor for each bulkhead instance, so the queue length does not depend on quarkus.vertx.worker-pool-size.
• If you need tighter isolation, create a custom ExecutorService and register it as a CDI bean, then inject it into the bulkhead via the executor attribute (available in SmallRye 2.3+). Example:

  @ApplicationScoped
  public class CustomExecutorProducer {
      @Produces
      @BulkheadExecutor
      public ExecutorService executor() {
          return Executors.newFixedThreadPool(5, new ThreadFactoryBuilder()
              .setNameFormat("bulkhead‑pool‑%d")
              .build());
      }
  }
  
  @Bulkhead(value = 2, queueSize = 5, executor = "@BulkheadExecutor")
  public void handle() { … }
  

Does Quarkus plan to add a new maxQueueSize attribute?

No additional attribute is planned; the existing queueSize covers the same requirement. The property name and semantics are stable in Quarkus 3.x and newer.

Recommended practices to bound pending tasks

• Set queueSize to a value that reflects your memory budget and desired back‑pressure level.
• Monitor smallrye.fault.tolerance.bulkhead.queue.size metrics (Micrometer, Prometheus) to detect saturation.
• Use separate bulkhead configurations for different service layers to avoid one component starving another.
• When possible, combine @Bulkhead with @Timeout or @Retry to release queued requests sooner.

Missing diagnostic detail

To tailor the configuration precisely, could you confirm the Quarkus version you are using? The queueSize property is only effective from Quarkus 3.0 onward; earlier releases ignore it.