[AMQ-6847] Immediate poison ACK after move from DLQ leads to message loss

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 5.14.5, 5.15.2
• Fix Version/s: 5.14.6, 5.15.3, 5.16.0
• Assignee: Gary Tully
• Reporter: Alvin Kwekel
• Created: 24/Oct/17 12:59
• Updated: 19/Dec/17 15:21
• Resolved: 27/Oct/17 10:42

Description

Setup

• setMaximumRedeliveries(0)
• Throw RuntimeException (Poison ACK) in consumer directly after receiving message

Scenario

• Move message from DLQ to original input queue (either web console or JMX retryMessages)
• Processing of message fails again directly
• The message is considered to be still on the DLQ on the rejection and ActiveMQ logs "Not adding duplicate to DLQ"

Introducing a delay before throwing the exception in the client will get around the issue.

I initially noticed the issue when using an AMQP reject (Apache Qpid Proton) after which I reproduced similar conditions (instant poison ACK) it with the JMS client. The attached Java app will reproduce the issue on 5.14.5 and 5.15.2.

Might be related to AMQ-5752.

2017-10-24 13:38:11,275 | DEBUG | Not adding duplicate to DLQ: ID:xxx-32848-1508845049112-6:1:1:1:1, dest: queue://TEST | org.apache.activemq.broker.region.policy.AbstractDeadLetterStrategy | ActiveMQ Transport: tcp:///127.0.0.1:36360@61616