[ARTEMIS-1568] Unsettled AMQP messages are lost when Receiver Link is opened on remote cluster member

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.4.0, 2.5.0, 2.6.0
• Fix Version/s: 2.6.1
• Component/s: AMQP, Broker
• Labels: amqp bridge broker cluster clustering

Environment

Fedora 26, openJDK8, AMQP .Net Lite client

• Assignee: Clebert Suconic
• Reporter: Todd Baert
• Created: 20/Dec/17 18:54
• Updated: 31/May/18 17:18
• Resolved: 31/May/18 17:17

Description

EDIT: This is observed in the current SNAPSHOT, as well as 2.4.0 (though the exception is slightly different there)

In a 2-broker cluster, if AMQP messages are sent to broker0, and then a receiver link (consumer) is connected to broker1, the messages are removed from broker0, and attempted to be forwarded across the cluster bridge, but are lost. broker1 shows exception:

ERROR [org.apache.activemq.artemis.core.server] AMQ224016: Caught exception: ActiveMQIllegalStateException[errorType=ILLEGAL_STATE message=AMQ119029: No address configured on the Server''s Session]
	at org.apache.activemq.artemis.core.server.impl.ServerSessionImpl.send(ServerSessionImpl.java:1393) [artemis-server-2.5.0-SNAPSHOT.jar:2.5.0-SNAPSHOT]
	at org.apache.activemq.artemis.core.server.impl.ServerSessionImpl.send(ServerSessionImpl.java:1327) [artemis-server-2.5.0-SNAPSHOT.jar:2.5.0-SNAPSHOT]
	at org.apache.activemq.artemis.core.server.impl.ServerSessionImpl.send(ServerSessionImpl.java:1320) [artemis-server-2.5.0-SNAPSHOT.jar:2.5.0-SNAPSHOT]
	at org.apache.activemq.artemis.core.protocol.core.ServerSessionPacketHandler.onSessionSend(ServerSessionPacketHandler.java:661) [artemis-server-2.5.0-SNAPSHOT.jar:2.5.0-SNAPSHOT]
	at org.apache.activemq.artemis.core.protocol.core.ServerSessionPacketHandler.onMessagePacket(ServerSessionPacketHandler.java:264) [artemis-server-2.5.0-SNAPSHOT.jar:2.5.0-SNAPSHOT]
	at org.apache.activemq.artemis.utils.actors.Actor.doTask(Actor.java:33) [artemis-commons-2.5.0-SNAPSHOT.jar:2.5.0-SNAPSHOT]
	at org.apache.activemq.artemis.utils.actors.ProcessorBase.executePendingTasks(ProcessorBase.java:66) [artemis-commons-2.5.0-SNAPSHOT.jar:2.5.0-SNAPSHOT]
	at org.apache.activemq.artemis.utils.actors.OrderedExecutor.doTask(OrderedExecutor.java:42) [artemis-commons-2.5.0-SNAPSHOT.jar:2.5.0-SNAPSHOT]
	at org.apache.activemq.artemis.utils.actors.OrderedExecutor.doTask(OrderedExecutor.java:31) [artemis-commons-2.5.0-SNAPSHOT.jar:2.5.0-SNAPSHOT]
	at org.apache.activemq.artemis.utils.actors.ProcessorBase.executePendingTasks(ProcessorBase.java:66) [artemis-commons-2.5.0-SNAPSHOT.jar:2.5.0-SNAPSHOT]
	at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149) [rt.jar:1.8.0_151]
	at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624) [rt.jar:1.8.0_151]
	at java.lang.Thread.run(Thread.java:748) [rt.jar:1.8.0_151]

This is similar (but not identical to) https://issues.apache.org/jira/browse/ARTEMIS-1542 which was fixed a few weeks back. In ARTEMIS-1542, the receiver links are present on both brokers are the time the message are sent. They are now load balanced and received properly.

In this issue, the receiver link is created after the messages already exist in broker0. I'm not sure if the desired behavior is for these messages simply to stay on broker0 and not cross over to broker1, but right now, they are sent over the cluster bridge and lost forever.

In this case, it looks like the 2nd arg to ClusterConnectionBridge.beforeForward() is null. I have attached some native AMQP client code that demonstrates this issue ( AMQP .Net Lite client, c#):

using System;
using System.Collections.Generic;
using System.Threading;
using Amqp.Framing;
using Amqp.Extensions;
using Amqp.Sasl;
using Amqp.Types;

namespace Amqp.Extensions.Examples
{
    class Program
    {
        static void Main(string[] args)
        {
            // create receiver link on broker0
            Connection connection0 = new Connection(new Address("amqp://localhost:5672"));       
            Session session0 = new Session(connection0);
            ReceiverLink receiver0 = new ReceiverLink(session0, "test", "orders");
            
            // send messages to broker0
            SenderLink sender = new SenderLink(session0, "sender", "orders");
            Message message = new Message("a message!");
            
            for (var i = 0; i < 5; i++)
            {
                sender.Send(message);
                Thread.Sleep(100);
            }

            // receive 1 of 5, works as expected...
            Message m = receiver0.Receive(TimeSpan.FromSeconds(1));
            Console.WriteLine(m.Body);
            receiver0.Accept(m);
            session0.Close();
            connection0.Close();

            // create receiver link on broker1
            Connection connection1 = new Connection(new Address("amqp://localhost:5673"));       
            Session session1 = new Session(connection1);
            ReceiverLink receiver1 = new ReceiverLink(session1, "test", "orders");

            // these 4 messages are removed from broker0 (ack'd) but never delivered. NPE seen in logs on broker1
            for (var i = 0; i < 4; i++)
            {   
                m = receiver1.Receive(TimeSpan.FromSeconds(1));
                if (m != null)
                {
                    Console.WriteLine(m.Body);
                    receiver1.Accept(m);
                }
            }

            session1.Close();
            connection1.Close();
        }
    }
}

Issue Links

• is duplicated by: ARTEMIS-1858 Expired messages are not traversing bridge with AMQP