[
https://issues.apache.org/jira/browse/ARTEMIS-5787?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18056453#comment-18056453
]
Dmitriy Nozdrya edited comment on ARTEMIS-5787 at 2/4/26 2:55 PM:
------------------------------------------------------------------
Hello [~jbertram] , [~clebertsuconic] !
After a brief analysis of the problem, I suspected that the
SendAcknowledgementHandler doesn't always work correctly at the BridgeImpl
class level when running cluster bridges.
When an error occurs while a message is being sent to broker_2 (broker_1 ----
redistribution_msg ----> broker_2) during redistribution, in some cases,
[sendAcknowledged|https://github.com/apache/artemis/blob/11717d8d7e7fa43b8030cbeca79e4d005e517b25/artemis-server/src/main/java/org/apache/activemq/artemis/core/server/cluster/impl/BridgeImpl.java#L476]
method is executed and the message is removed from the queue when acknowledge
method is called, although
[sendFailed|https://github.com/apache/artemis/blob/11717d8d7e7fa43b8030cbeca79e4d005e517b25/artemis-server/src/main/java/org/apache/activemq/artemis/core/server/cluster/impl/BridgeImpl.java#L468]
should be executed and the message should be resent to broker_2.
Such behavior is possible?
was (Author: JIRAUSER284724):
Hello [~jbertram] , [~clebertsuconic] !
After a brief analysis of the problem, I suspected that the
SendAcknowledgementHandler doesn't always work correctly at the BridgeImpl
class level when running cluster bridges.
When an error occurs while a message is being sent to broker_2 (broker_1 ----
redistribution_msg ----> broker_2) during redistribution, in some cases,
[sendAcknowledged|https://github.com/apache/artemis/blob/11717d8d7e7fa43b8030cbeca79e4d005e517b25/artemis-server/src/main/java/org/apache/activemq/artemis/core/server/cluster/impl/BridgeImpl.java#L476]
method is executed and the message is removed from the queue when acknowledge
method is called, although
[sendFailed|https://github.com/apache/artemis/blob/11717d8d7e7fa43b8030cbeca79e4d005e517b25/artemis-server/src/main/java/org/apache/activemq/artemis/core/server/cluster/impl/BridgeImpl.java#L468]
should be executed and the message should be resent to broker_2.
Is this behavior possible?
> Message loss when redistribution is failed
> ------------------------------------------
>
> Key: ARTEMIS-5787
> URL: https://issues.apache.org/jira/browse/ARTEMIS-5787
> Project: Artemis
> Issue Type: Bug
> Components: AMQP, Broker
> Affects Versions: 2.33.0, 2.42.0, 2.44.0
> Environment:
> Reporter: Dmitriy Nozdrya
> Priority: Critical
> Attachments: AMQPMessageRedistributionFailedTest.java
>
>
> If an exception occurs during redistribution while receiving a message from
> the broker, some messages are lost.
> Some messages remain in the bridge queue, while others are lost.
> If errors occur during redistribution, messages must not be lost. For some
> use cases, this is unacceptable.
> Reproduction steps:
> 1. A symmetric cluster consists of two brokers (broker_1, broker_2),
> operating mode ON_DEMAND.
> 2. A producer connects to broker_1 and writes messages to the queue.
> 3. A consumer connects to broker_2 to read messages.
> 4. Simulate the exception in
> org.apache.activemq.artemis.spi.core.protocol.EmbedMessageUtil#extractRegularMessage.
>
> {code:java}
> private static Message extractRegularMessage(ICoreMessage message,
> StorageManager storageManager) {
> RuntimeException exception = new RuntimeException("Exception occurred");
> logger.error(String.format("Error occurred for message=%s", message),
> exception.getMessage(), exception);
> throw exception;
> // ActiveMQBuffer buffer = message.getReadOnlyBodyBuffer();
> //
> // if (buffer.readableBytes() < signature.length ||
> !checkSignature(buffer)) {
> // logger.trace("Message type {} was used for something other than
> embed messages, ignoring content and treating as a regular message",
> Message.EMBEDDED_TYPE);
> // return message;
> // }
> //
> // return readEncoded(message, storageManager, buffer);
> } {code}
> The test case can be reproduced using the following test-class -
> [^AMQPMessageRedistributionFailedTest.java]
> {code:java}
> package org.apache.activemq.artemis.tests.integration.cluster.distribution;
> import ...
> public class AMQPMessageRedistributionFailedTest extends ClusterTestBase
> {...}{code}
> {code:java}
> // When running the test, the failed to send log is visible
> AMQ224143: Bridge
> $.artemis.internal.sf.cluster0.1b68cb70-d032-11f0-8cc5-2a5a1ea9ca09 failed to
> send CoreMessage[messageID=178, durable=true, userID=null, priority=0,
> timestamp=0, expiration=0, durable=true, address=exampleQueue, size=102833,
> properties=TypedProperties[]]@1819912186: ActiveMQInternalErrorException
> // The result of the test execution is SUMMARY_LOG
> [main] 13:23:30,142 INFO
> [org.apache.activemq.artemis.tests.integration.cluster.distribution.AMQPMessageRedistributionFailedTest]
>
> SUMMARY:
> messageCount=100 sent in exampleQueue queue for node 0
> messageCount=0 for exampleQueue queue for node 0
> messageCount=0 for exampleQueue queue for node 1
> messageCount=88 for
> $.artemis.internal.sf.cluster0.b3f833cd-d034-11f0-a96c-2a5a1ea9ca09
> bridge_queue for node 0
> messageCount=0 for
> $.artemis.internal.sf.cluster1.b2d7336a-d034-11f0-a96c-2a5a1ea9ca09
> bridge_queue for node 1
> LOST_MESSAGES_COUNT=12
> {code}
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
