[
https://issues.apache.org/activemq/browse/AMQ-2880?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Stan Lewis updated AMQ-2880:
----------------------------
Attachment: xa-test-case.txt
patch_for_xa_transaction.txt
Here's a test for XATransaction. I'm finding however that catching the
exception within commit and calling rollback() from there doesn't seem to fix
the issue. Instead it seems like the message is effectively lost, as when the
dummy persistent store throws it's exception the message is out of the
database, I wouldn't have expected that. Also to properly rollback in the
client I had to call XAResource.start() for some reason, also a bit unexpected.
Anyway, am also attaching the changes I had done to XATransaction.java.
> Exception thrown during commit leads to message loss
> ----------------------------------------------------
>
> Key: AMQ-2880
> URL: https://issues.apache.org/activemq/browse/AMQ-2880
> Project: ActiveMQ
> Issue Type: Bug
> Components: Broker
> Affects Versions: 5.4.0
> Reporter: Stan Lewis
> Assignee: Gary Tully
> Fix For: 5.4.1
>
> Attachments: patch_for_xa_transaction.txt, test-case.txt,
> test-case.txt, xa-test-case.txt
>
>
> In cases where JDBC persistence is used and the database server is under a
> fair bit of load it's sometimes possible for table/row locks to time out,
> which means you'll see exceptions such as:
> java.sql.BatchUpdateException: Lock wait timeout exceeded; try restarting
> transaction
> at
> com.mysql.jdbc.PreparedStatement.executeBatchSerially(PreparedStatement.java:1693)
> at com.mysql.jdbc.PreparedStatement.executeBatch(PreparedStatement.java:1108)
> at
> org.apache.commons.dbcp.DelegatingStatement.executeBatch(DelegatingStatement.java:297)
> at
> org.apache.commons.dbcp.DelegatingStatement.executeBatch(DelegatingStatement.java:297)
> at
> org.apache.activemq.store.jdbc.TransactionContext.executeBatch(TransactionContext.java:98)
> at
> org.apache.activemq.store.jdbc.TransactionContext.executeBatch(TransactionContext.java:80)
> at
> org.apache.activemq.store.jdbc.TransactionContext.commit(TransactionContext.java:161)
> at
> org.apache.activemq.store.jdbc.JDBCPersistenceAdapter.commitTransaction(JDBCPersistenceAdapter.java:504)
> at
> org.apache.activemq.store.memory.MemoryTransactionStore$Tx.commit(MemoryTransactionStore.java:109)
> at
> org.apache.activemq.store.memory.MemoryTransactionStore.commit(MemoryTransactionStore.java:220)
> at org.apache.activemq.transaction.XATransaction.commit(XATransaction.java:73)
> at
> org.apache.activemq.broker.TransactionBroker.commitTransaction(TransactionBroker.java:176)
> at
> org.apache.activemq.broker.MutableBrokerFilter.commitTransaction(MutableBrokerFilter.java:103)
> at
> org.apache.activemq.broker.TransportConnection.processCommitTransactionTwoPhase(TransportConnection.java:430)
> at org.apache.activemq.command.TransactionInfo.visit(TransactionInfo.java:102)
> at
> org.apache.activemq.broker.TransportConnection.service(TransportConnection.java:309)
> at
> org.apache.activemq.broker.TransportConnection$1.onCommand(TransportConnection.java:185)
> at
> org.apache.activemq.transport.TransportFilter.onCommand(TransportFilter.java:69)
> at
> org.apache.activemq.transport.WireFormatNegotiator.onCommand(WireFormatNegotiator.java:113)
> at
> org.apache.activemq.transport.InactivityMonitor.onCommand(InactivityMonitor.java:217)
> at
> org.apache.activemq.transport.TransportSupport.doConsume(TransportSupport.java:83)
> at org.apache.activemq.transport.tcp.TcpTransport.doRun(TcpTransport.java:219)
> at org.apache.activemq.transport.tcp.TcpTransport.run(TcpTransport.java:201)
> at java.lang.Thread.run(Thread.java:636)
> In this case the connection to the database is fine and what we should do is
> retry the transaction as it's a temporary failure condition. Instead what
> happens is the broker moves to the next message in the queue, leaving the
> current message in the database. The message won't show up in the web
> console and cannot be consumed by any consumers until the broker is restarted.
> Attached is a test case that simulates the error condition in a controlled
> fashion by using a subclassed JDBCPersistenceAdapter that will throw an
> exception in commitTransaction as necessary. So the producer sends 10
> messages and then the consumer tries to consume them, during this time the
> persistence adapter will throw an exception during commitTransaction. Then
> the condition is cleared and the consumer can consume all 10 messages,
> however the consumer only consumes 9 messages, the 1st message in the
> sequence is still in the database. Hopefully the logging makes this clear.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
