I have created a File System Events stream and start it running.
What I find odd, particularly in light of the API design which is
clearly designed to handle large volumes of events, is that I never
get any big batches of events passed to my callback routine, even
when I start with a stream that asks for hundreds of thousands of
historical events.
Every time my callback function is called, the numEvents parameter
is 1. It processes one event
That's just an artifact of how historical events are
dealt with. With live events that happen while your
stream is running they will be batched.
When processing thousands of historical events, most of the CPU time
is wasted in the run loop. According to Shark & sample, only 10% of
the time is spent in my callback function. The other 90% is spent in
the run loop. This makes me think that I could get up to an order of
magnitude improvement in performance if the stream would just pass
more than 1 event at a time to my callback.
There is already a bug filed to batch historical events.
It may happen for SnowLeopard.
