Thursday, September 30, 2010


Recently I've stuck in the problem when the Streams checkpoint wasn't updating the required_checkpoint_scn and applied_scn fields in dba_capture. This caused a lot of problems as RMAN couldn't remove the processed logs from the system (it uses the required_checkpoint_scn to determine whether the Streams still needs those logs). As a result our DBA has raised an SR
with Oracle, which has resulted into th advice to reset the aq_tm_processes parameter.

It's turned out that starting from 10g aq_tm_processes sets the number of queue monitor slave processes dedicated to monitor the PERSISTENT queues in the system rather than the buffered ones. If you set the value for this parameter as 10 than ALL of the salve monitors will look at the persistent queues and NONE of then will monitor the buffered queues. That was the case
in our system.

Oracle advises to remove that parameter from init.ora/spfile at all and bounce the instance. That would enable Oracle to manage automatically how many processes will watch persistent/buffered queues. If instance bouncing isn't an option in your environment, Oracle advises setting it to 1 which can be done dynamically. Pay attention that manual setting the parameter to 0 will totally disable the queue monitoring and is highly unrecommended (haven't checked what it could lead to though).

1 comment: