Hazelcast 3.12.12 - Semaphore slowness performance

[BertuNet]

Hi all,
i've a question for who get the same problem.
We are using a cp subsystem (3 nodes) and a distribuited semaphore.
We've created more groups (with a specific key) to handle the lock.
Our spring-boot services take from less then a second to 9-10 second to get the semaphore.
Sometimes we get errors like:

• LEADER_LOCAL query failed
  java.util.concurrent.RejectedExecutionException: Task com.hazelcast.spi.impl.AbstractCompletableFuture$ExecutionCallbackRunnable@323b4c45 rejected from ThreadPoolExecutor{name='hz:async'}
• Could not get CP group info of CPGroupId{name='KEY_1234', seed=0, commitIndex=1514619}
  com.hazelcast.core.OperationTimeoutException: RaftQueryOp invocation failed to complete due to operation-heartbeat-timeout

Anyone has an idea about?
What may i check? Is there a way to tuning semaphore performance?

Thank u,

Regards