[
https://issues.apache.org/activemq/browse/AMQ-1244?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_39604
]
Marcos Sanz commented on AMQ-1244:
----------------------------------
I assume that the only communication channel that two brokers have in common is
the database. What about leveraging on the usual concept of a heartbeat? Let
the Lock-Table have a datetime row and let the active broker update
periodically the row with the current (database) time. If the standby broker
detects that the row value diverges from the current time more than a delta,
then the active broker is stalled/dead. Advantages of this technique:
a) No transaction kept open for an undetermined time window
b) To my eyes, this is even more effective in detecting a failure that the
current mechanism, for it requires now the active broker to actually undertake
an action,and not just remain inactive holding a token.
Comments?
> DatabaseLocker implementation impedes database replication
> ----------------------------------------------------------
>
> Key: AMQ-1244
> URL: https://issues.apache.org/activemq/browse/AMQ-1244
> Project: ActiveMQ
> Issue Type: Bug
> Components: Message Store
> Affects Versions: 4.1.1
> Environment: Latest ActiveMQ snapshot, Sybase ASE 12.5.x
> Reporter: Marcos Sanz
> Priority: Critical
>
> The current implementation of the JDBC Master/Slave feature makes one broker
> (the master) acquire a lock on a database object. In Sybase, this has been
> implemented with the command:
> LOCK TABLE foo IN EXCLUSIVE MODE
> This command can only be executed within a transaction, see:
> http://manuals.sybase.com/onlinebooks/group-as/asg1250e/sqlug/@Generic__BookTextView/54552;pt=54651
> This implies that for the whole lifespan of the ActiveMQ-process there is an
> open transaction in the RDBMS. This is a problem in a professional
> environment making use of a database replication server: The open transaction
> impedes that the transaction log in the active database is emptied, then the
> stable queue at the replication server won't be purged and will steadily grow
> up to infinitum. We have been able to observe this behaviour.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
