I got to try out the scenario I described above (ie) physically remove the network cable from the Master.
Basically the database did not realize that the client had disconnected abruptly, and there was still a lock on the database table, (which was now an invalid stale lock), and would never get released. After some further digging I realised that there was an issue raised about this, and also explains the exact behaviour I experienced. https://issues.apache.org/activemq/browse/AMQ-1958 https://issues.apache.org/activemq/browse/AMQ-1958 I know that there's nothing that can be done really at the ActiveMQ level, its more of a DB issue really. I did look into maybe setting the tcp_keepalive to a low value which might force the (invalid)connection to close quicker at a TCP level, but I didn't want to mess around with stuff like that at such a low level. Any lock options at the DB level were specific to the amount of time you wait until you get a lock, not the amount of time you can hold onto it. In the end I wrote a DB specific script which basically kills any sessions for a particular IP, which in turn releases the lock and the slave can kick in. If anyone knows of a different (perhaps better) way of implementing this please let us know. Gary Tully wrote: > > You have got to ask the JDBC driver or database documentation. I imagine > there is a lock expiry option that can be tweaked. > > 2009/6/29 colonelx <z...@zacburke.com> > >> >> Hi, >> >> I have 2 brokers setup in a JDBC Master Slave Configuration. >> >> Each broker is on a different machine. BoxA and BoxB >> The database is also on a different machine BoxC. (mysql 5.1) >> >> The failover is working fine when I manually kill any instance of >> ActiveMQ >> on BoxA or BoxB. (Ctrl-C) >> >> However, can anyone tell me what will happen if the if the Master >> instance >> looses its network connection ? >> Will its lock on the database get dropped immediately ? or will it still >> hold a (now invalid) lock on the database ? >> I ask this, because if this happens , there could possibly be a condition >> where the slave won't ever be able to get a lock on the database because >> the >> master has lock which did not get correctly released and manual >> intervention >> would be needed in order to fix up the problem. >> >> I know that I could test this by plugging in and out the physical network >> connections, but I'm not really in a position to do this on any of the >> above >> boxes. >> >> Thanks. >> >> -- >> View this message in context: >> http://www.nabble.com/JDBC-Master-Slave-Question-tp24251198p24251198.html >> Sent from the ActiveMQ - User mailing list archive at Nabble.com. >> >> > > > -- > http://blog.garytully.com > > Open Source Integration > http://fusesource.com > > -- View this message in context: http://www.nabble.com/JDBC-Master-Slave-Question-tp24251198p24273534.html Sent from the ActiveMQ - User mailing list archive at Nabble.com.
