Jano wrote:
> Thinking about this, we could perhaps slightly change the consumer model
> and, instead of generating ten events at once, generate one for the first
> reader and when it triggers, generate one event at a time for the next
> reader. This way we reduce by a factor of ten the pending events.

Possibly... it depends on the use case again, I suppose. Your suggestion 
sounds similar to the way things propagate through blogs, whereas the 
current behaviour is more like a one-writer-many-readers model.

I've committed some changes to SVN so the readers aren't informed until 
after the search leaves the node's queue. That should keep the number of 
pending events under control, but it will also change the shape of the 
graphs, because the number of requests will be affected by throttling.

Cheers,
Michael

Reply via email to