As I talked to Tim this morning on IRC, there is an issue with ordering on the AIO Callbacks.

Which is the same issue we had before with replication. As there's nothing to hold on non-persistence, non persistence sends were happening out of order (when mixing with persistent).

I could fix that by adding a method syncCallback on the SequentialFiles. This method won't perform any IO, but it will make sure the callback will be called only after the current pending IO operations are performed.

I'm now doing a further check on the Producer. It should be no reason to sync if there are no pending persistent messages to be sent. So I will add extra check to make sure there wouldn't be any syncs on those cases.


Also, this will not do any context switch if there is no need to sync, the producer will just send to the queues directly.

The OperationContext (old ReplicationContext, now renamed) has a check for...

if pending==0
 runnable.run