On Mon, Oct 15, 2018 at 5:26 PM Mark Thomas <ma...@apache.org> wrote:
> > and we shouldn't pollute that with side uses. Is that > > doable in 9, or should it be added to the "10" todo list ? > > No strong view. If it can be done without impacting the external API > then no objections here. > This looks ok compatibility wise. I will add this refactoring after 9.0.13. Now, looking at things: - There are two more heavyweight uses of an executor in clustering: MessageDispatchInterceptor and RecieverBase. This might be moveable too after review, IMO there's no gain to have multiple small internal pools. - The async timeout threads of the connectors is an obvious item that could be moved to that executor (it runs only once per second). I am adding the necessary get/set APIs on ProtocolHandler and Endpoint. Looking at other items: - I could obviously avoid using a dedicated accept thread for NIO2 but it's more complex. So maybe later. - tribes.transport.bio ... Why does this still exist ? Rémy
