On 10/7/14, 11:27 PM, "Michael Dürig" <mdue...@apache.org> wrote:
> > >On 7.10.14 10:30 , Tommaso Teofili wrote: >> 2014-10-07 18:15 GMT+02:00 Rob Ryan <rr...@adobe.com>: >> >>> I'm not so sure about the conclusion that because the events of >>>interest >>> are local you can safely use userData. IIUC under heavy load local >>>events >>> may be aggregated to mitigate the load and in that process 'localness' >>>is >>> lost... >>> >> >> maybe I missed that piece of information from the linked page, where did >> you get that? > >That information is missing from the page. But Rob is right, under some >circumstances local events get aggregated such that the user data >becomes unavailable (i.e. the commit boundaries get lost). This happens >when the commit rate is too hight for event processing to keep up with. >There is some effort under OAK-1368 to further improve this. However >this will probably still take some time. Until then there is also the >option to enable a kind of flow control through the CommitRateLimiter, >which allows to throttle the commit rate until listeners have caught up. Hi, You are saying that under load the local events are dispatched as external events, right? So if a listener acts only on local events it will miss some events? But is it guaranteed that all events dispatched as local events have the user data filled if available? Regards, Marius
