ARTEMIS-5733 Core bridge can lock on flow control

[AntonRoskvist]

I've observed this issue happening in live environments for a long time but it's very uncommon. In the environment where it happens most frequently it's no more than something like once a month or so.

I've been unable to find a really exact way of triggering the issue artificially, but the added test testBridgeHandlesReachingZeroCredits does it fairly reliably.

I'm open to changing the test if anyone has any suggestions on how to precisely trigger the issue... or just remove it and accept the change based on static analysis.

[AntonRoskvist]

So to clarify the change... if the AsynchronousProducerCreditsImpl receives credits and lands on a balance of exactly 0. The callback to the bridge will set blockedOnFlowControl = true based on balance <= 0 but the following handling of credit starvation evaluates balance < 0 so it is not executed.

Following this, the bridge will return all calls to handle with HandleStatus.BUSY so it will never attempt to get more credits, effectively locking it.

[jbertram]

Nice find! A single character bug fix.

I recommend removing your existing test and adding this to the tests in activemq-core-client:

package org.apache.activemq.artemis.core.client.impl;

import org.junit.jupiter.api.Test;
import org.junit.jupiter.api.Timeout;
import org.mockito.Mockito;

public class AsynchronousProducerCreditsImplTest {

   @Test
   @Timeout(10)
   public void testZeroCredits() throws Exception {
      ClientSessionInternal session = Mockito.mock(ClientSessionInternal.class);
      AsynchronousProducerCreditsImpl producerCredits = new AsynchronousProducerCreditsImpl(session, null, 0, Mockito.mock(ClientProducerFlowCallback.class));
      producerCredits.receiveCredits(0);
      Mockito.verify(session).sendProducerCreditsMessage(0, null);
   }
}

This test is simpler, faster, and reproduces the problem every time.

[AntonRoskvist]

Thanks @jbertram , that's a much better test!

[jbertram]

I'll merge as soon as the PR builds complete successfully!

[jbertram]

The failure of testReclaimingAfterConcurrentAddsAndDeletesTx on JDK 21 looks completely unrelated. I'm merging.