Details
-
Bug
-
Status: Closed
-
Major
-
Resolution: Not A Bug
-
5.15.2
-
None
-
None
Description
We have ActiveMq 5.15.2 in following configuration:
- PostgreSQL for persistance
- two nodes, one in standby
- JDBC master slave with shared database
- static cluster discovery
Everything seams to be fine, failover works as expected, but sometimes during failover we are observing following exception:
28.02.2018 09:28:54,207 WARN [ActiveMQ NIO Worker 6] org.apache.activemq.transaction.LocalTransaction - Store COMMIT FAILED:java.io.IOException: Batch entry 2 INSERT INTO ACTIVEMQ_MSGS(ID, MSGID_PROD, MSGID_SEQ, CONTAINER, EXPIRATION, PRIORITY, MSG, XID) VALUES (7095330, 'xxx-1-1519303952070-3:862:8:1', 15, 'queue://abc', 1520411334073, 4, ?, NULL) was aborted: FEHLER: doppelter Schl▒sselwert verletzt Unique-Constraint ▒activemq_msgs_pkey▒ Detail: Schl▒ssel ▒(id)=(7095330)▒ existiert bereits. Call getNextException to see other errors in the batch. at org.apache.activemq.util.IOExceptionSupport.create(IOExceptionSupport.java:46) at org.apache.activemq.store.jdbc.TransactionContext.commit(TransactionContext.java:209) at org.apache.activemq.store.jdbc.JDBCPersistenceAdapter.commitTransaction(JDBCPersistenceAdapter.java:516) at org.apache.activemq.store.memory.MemoryTransactionStore$Tx.commit(MemoryTransactionStore.java:113) at org.apache.activemq.store.memory.MemoryTransactionStore.commit(MemoryTransactionStore.java:270)
ActiveMQ propagates this exception directly to the client.
I am assuming, that due to a failover some clients did not get acknowledgment and message has been resent to a new node.
If I am correct ActiveMq should just ignore duplicated message, or both messages should be stored in database. But latest it's not possible, because ACTIVEMQ_MSGS#ID is a PK