JDBC Master/Slave: After database failure, stale db connections are reused in
DefaultdatabaseLocker
---------------------------------------------------------------------------------------------------
Key: AMQ-1263
URL: https://issues.apache.org/activemq/browse/AMQ-1263
Project: ActiveMQ
Issue Type: Bug
Components: Message Store
Affects Versions: 4.x
Environment: Linux x86, Postgresql 8.1
Reporter: Eric Anderson
Attachments: patch.txt
Start two brokers in a JDBC master/slave configuration. One will become
master, the other slave.
Shutdown the database.
The slave will immediately begin emitting emitting database connection
exceptions messages from the DefaultdatabaseLocker.
Bring up the database again.
The slave continues to emit database error messages and all subsequent attempts
by the DefaultdatabaseLocker to issue its locking query will fail.
The problem is that the DefaultdatabaseLocker grabs one db connection at
startup and never releases it. Therefore unless the JDBC connection has an
auto-reconnect feature, the slave broker is dead in the water, because its
Connection is bad. The fix is to allocate a new Connection and
PreparedStatement on every iteration of the database locker loop, and free the
statement/connection on any Exception.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
