2009/1/20 Gordon Sim <[email protected]>: > Marnie McCormack wrote: >> >> I'm looking at the implementation/design work for Flow To Disk on the Java >> Broker and would interested in your thoughts. >> >> The JIRA is: >> >> https://issues.apache.org/jira/browse/QPID-949 >> >> Please have a look and add a comment/reply to this thread if you have any >> ideas. >> >> C++-ers - is there some documentation about how the C++ broker does this ? > > Not really. For a queue has been selected to use the flow to disk policy, > when a message is enqueued that would exceed the configured limit (in > message count or aggregate size) its content is released from memory > (keeping the headers in memory). When such messages are delivered to a > subscriber the data is read in again from the store. > > The actual implementation is in need of some refactoring in places, but > thats the rough approach. >
I've added some comments to the JIRA on how I would approach the flow-to-disk work in Java... There's a bit more in my head but I am still snowed under with AMQP stuff at the moment. https://issues.apache.org/jira/browse/QPID-949?focusedCommentId=12665483#action_12665483 I think the key is to approach in small manageable chunks rather than trying to do something too complicated too soon (e.g. in the first instance rely on manually configured queue limits rather than having the broker attempt to dynamically "allocate" memory to queues). Cheers, Rob --------------------------------------------------------------------- Apache Qpid - AMQP Messaging Implementation Project: http://qpid.apache.org Use/Interact: mailto:[email protected]
