On Mon, Jul 8, 2013 at 1:52 PM, Paul Gale <paul.n.g...@gmail.com> wrote:
> Hi, > > I've read where message cursors should be used when PFC is turned off or > for when consumers fall behind producers. When PFC is turned off, you have to make sure that the broker can deal with more messages coming in than it can handle especially in the case of non-persistent messages (offline to disk, purge old messages, expire messages, etc, etc) > The doc states that in the normal > case, so-called, where consumers keep up with producers, does not involve a > pending cursor of any kind. The configured cursor is only engaged when a > consumer falls behind its producer. > Pending cursors are used throughout the architecture to buffer messages that are pending for dispatch. In the normal case, messages are cached in memory (persistent and non-persistent) thus can be delivered much faster. If consumers fall behind, then they are no longer cached and are expected to live in a cursor capable of handling overflow (disk for persistent messages, filependingmessagecursor for non-persistent). > > What happens when a consumer catches back up with the producer: does the > configured cursor disengage resulting in messages being passed directly to > the dispatch queue, the way it did before the consumer fell behind? If not, > why? > If consumer catches back up, cache will be used again and all will be good. > > Thanks, > Paul > -- *Christian Posta* http://www.christianposta.com/blog twitter: @christianposta
